{"repo": "akashmahlaz/fireclaw", "files": {"PLAN.md": {"994174926b7f2828eec0d73419c60cdd8095081a": {"t": 1785747204.1210766, "n": 16035, "h": "9c945bfc6092f13e", "c": "# 🔥 FireClaw — The Plan (Research-Backed v2)\n\n## Current Priority (Updated May 2026)\n\n### Deployment is the #1 focus — this is what brings business.\n\n**What we deploy:**\n1. **OpenClaw** (our modified version) — Priority #1. Verify it works end-to-end.\n2. **Ad-Chain Verify Agent** (OpenClaw base + ad-tech modifications) — Priority #2. Client waiting.\n   - Pricing follows ad-tech market rates (not same as OpenClaw)\n   - Better UI/UX specific to ad verification workflows\n   - Scripts in `/scripts` folder contain initial attempts\n3. **SEO Agent** (OpenClaw base + SEO capabilities) — Priority #3.\n   - Site audits, auto backlink building, ranking analysis\n   - Connect with user's site for ongoing monitoring\n4. **All other agents** — Marked as \"Coming Soon\" in the marketplace\n\n**Deployment UX:**\n- User chooses an agent from the marketplace (catalog page)\n- Goes to a dedicated page for that agent type\n- Clicks deploy → pays → sees \"Your agent is deploying\" status\n- Dashboard shows deployed agents only (no deployment wizard inside dashboard)\n\n**Branch strategy:** Only working in `main` branch. Remove unused branches.\n\n---\n\n## What Is FireClaw?\n\n**FireClaw is the multi-agent management platform for OpenClaw.**\n\nDeploy, manage, and monitor multiple AI agents from one dashboard. Built for agencies, teams, and power users who need more than a single agent.\n\nThink: **Vercel for AI agents** — not another Hostinger. \n\n---\n\n## Why Not Just Use Hostinger?\n\nHostinger offers 1-click OpenClaw at **$5.99/mo** with 100K+ agents deployed. They win on:\n- Price ($5.99 vs anything we could charge for single agent)\n- Brand trust (4M+ users)\n- Infrastructure (global data centers)\n\n**But Hostinger can't do:**\n- ❌ Manage 10 agents from one dashboard\n- ❌ White-label agents for your clients\n- ❌ Pre-built agent templates (sales bot, support bot)\n- ❌ Cross-agent analytics\n- ❌ Agency billing (charge your clients through your dashboard)\n- ❌ Agent marketplace / skill store\n- ❌ India-first pricing in ₹\n\n**FireClaw's wedge: Multi-agent management, agency tools, and templates.**\n\n---\n\n## The Pitch (30 seconds)\n\n> \"Managing AI agents for your clients? Deploy 10 OpenClaw agents in 5 minutes, manage them all from one dashboard, white-label them, and charge your clients — starting at ₹1,999/month.\"\n\n---\n\n## Revenue Model (Research-Backed)\n\n### Pricing (India-First)\n\n| Plan | Agents | Price/mo | Infra Cost | Margin |\n|------|--------|----------|------------|--------|\n| **Personal** | 1 | ₹499 (~$6) | ~$8 (shared) | ~25% |\n| **Business** | 3 | ₹1,999 (~$24) | ~$20 (shared) | ~17% |\n| **Agency** | 10 | ₹4,999 (~$60) | ~$50 (shared) | ~17% |\n| **Enterprise** | Unlimited | Custom | Custom | 30%+ |\n\n**Why these prices work:**\n- Personal at ₹499 matches Hostinger's $5.99 in value\n- Business/Agency have NO equivalent competitor (Hostinger = 1 agent/plan)\n- Margins improve with density (more agents per server)\n\n### Infrastructure Strategy (NOT 1 EC2 per agent)\n\nInstead of dedicated instances, use **shared containerized infrastructure:**\n\n| Server | Type | Cost/mo | Agents Hosted | Cost/Agent |\n|--------|------|---------|---------------|------------|\n| Shared | t3.xlarge (4 vCPU, 16 GB) | ~$122 | 10-15 | $8-12 |\n| Shared | t3.2xlarge (8 vCPU, 32 GB) | ~$244 | 25-30 | $8-10 |\n| Dedicated | t3.micro (for personal) | ~$11 | 1 | $11 |\n\n*With 1yr Reserved Instances: costs drop 35-40%.*\n\n### Add-On Revenue\n| Add-On | Price |\n|--------|-------|\n| Extra agent | ₹299/mo |\n| WhatsApp messages (over 1000/mo) | ₹1/message |\n| Premium templates | ₹499-999 one-time |\n| White-label branding | Included in Agency+ |\n| Bundled AI credits (no BYOK needed) | ₹499/mo for 10K messages |\n\n### Revenue Projections\n\n| Milestone | Users | MRR | ARR |\n|-----------|-------|-----|-----|\n| Month 3 | 60 | ₹75K (~$900) | ~$10.8K |\n| Month 6 | 250 | ₹300K (~$3.6K) | ~$43K |\n| Month 12 | 700 | ₹1M (~$12K) | ~$144K |\n| Month 24 | 3,000 | ₹5M (~$60K) | ~$720K |\n\n---\n\n## Target Customers (Revised)\n\n### Phase 1 — Indian Agencies & Freelancers (Month 1-3)\n- Digital marketing agencies deploying chatbots for clients\n- Freelance developers offering \"AI assistant\" as a service\n- **Why they pay:** Managing 5-10 client agents from one place saves hours/week\n\n### Phase 2 — SMBs & Startups (Month 3-6)\n- Small businesses wanting AI for support/sales\n- Startups that need AI agents across departments\n- **Why they pay:** Templates get them started in minutes, not days\n\n### Phase 3 — Global Agencies (Month 6+)\n- International agencies managing AI fleets\n- White-label resellers\n- **Why they pay:** White-label + client billing = new revenue stream for them\n\n---\n\n## Competitive Landscape (Research-Backed)\n\n| Competitor | What | Price | Gap |\n|-----------|------|-------|-----|\n| **Hostinger OpenClaw** | 1-click single agent | $5.99/mo | No multi-agent, no agency tools |\n| **Dante AI** | AI chatbots | $40-400/mo | Chatbot only, not full agent, expensive |\n| **Botpress** | Chatbot builder | $0-495/mo | Visual builder, not OpenClaw, no WhatsApp focus |\n| **Chatfuel** | WhatsApp/IG bots | $69/mo | Simple bots, not AI agents |\n| **Tidio** | Customer service AI | $24-749/mo | Support-focused, not general agents |\n| **ManyChat** | Marketing automation | $15-100/mo | Flow-based, not AI agents |\n| **Railway/Render** | Generic PaaS | $5-85/mo | Not AI-specific, complex |\n\n**FireClaw's unique position:** Only platform for managing multiple OpenClaw agents with agency tools.\n\n---\n\n## Phases (Revised)\n\n### Phase 1: MVP — Agency Dashboard (3 weeks)\n- Landing page targeting Indian agencies\n- Multi-agent deploy from one dashboard\n- 3 pre-built templates (sales bot, support bot, personal assistant)\n- Stripe + Razorpay billing\n- 10-20 beta agencies\n- **Goal:** Prove agencies will pay for multi-agent management\n\n### Phase 2: Templates & Marketplace (3 weeks)\n- 10+ agent templates\n- Skill marketplace (install skills per agent)\n- White-label setup\n- Client billing (agencies charge their clients through us)\n- **Goal:** 100 paying users\n\n### Phase 3: Scale (Ongoing)\n- Multi-region (Mumbai, Singapore, US-East)\n- Agent analytics dashboard\n- Mobile app for monitoring\n- API for programmatic agent management\n- Community template submissions\n- **Goal:** ₹1M MRR\n\n---\n\n## Tech Stack (Revised)\n\n### Frontend (fireclaw.ai)\n- Next.js 16 + TypeScript\n- Tailwind CSS + shadcn/ui\n- Stripe + Razorpay for payments\n- authjs for authentication \n\n### Backend (Provisioning Engine)\n- Node.js API\n- **Docker + Docker Compose** (not bare EC2 per agent)\n- AWS SDK v3 (EC2 for host servers, not per agent)\n- MongoDB Atlas (user data, agent registry)\n- Bull/BullMQ (provisioning queue)\n\n### Agent Infrastructure\n- **Shared host servers** running Docker\n- Each agent = isolated Docker container with OpenClaw\n- Caddy reverse proxy (routing subdomains to containers)\n- Automated health checks + restart\n- Central logging (agent metrics → dashboard)\n\n### Why Docker Over Bare EC2\n| Approach | Cost for 10 agents | Management |\n|----------|-------------------|------------|\n| 10x t3.micro | ~$106/mo | 10 servers to manage |\n| 1x t3.xlarge + Docker | ~$125/mo | 1 server, 10 containers |\n| 2x t3.large + Docker | ~$132/mo | 2 servers, 5 containers each (HA) |\n\nDocker is similar cost but dramatically simpler to manage and scale.\n\n---\n\n## Key Risks & Mitigations\n\n| Risk | Mitigation |\n|------|-----------|\n| Hostinger drops price or adds multi-agent | Move fast, build community moat |\n| OpenClaw changes licensing | We contribute upstream, stay aligned |\n| Low margins on Personal plan | Personal is acquisition funnel → upsell to Business/Agency |\n| Container density limits | Auto-scale to new host servers |\n| India market slow to adopt | Parallel global launch at USD pricing |\n\n---\n\n## Key Metrics\n\n- **Agents deployed** — total across all users\n- **Agents per user** — avg (target: 2.5+)\n- **MRR** — monthly recurring revenue\n- **Provisioning time** — target: <60 seconds\n- **Uptime per agent** — target: 99.5%+\n- **Churn** — target: <5%/month\n- **CAC** — cost to acquire (target: <₹500)\n- **LTV** — lifetime value (target: >₹15,000)\n\n---\n\n## Multi-Provider API Key + Live Model Selection\n\n### Overview\nTwo places users configure AI providers with the same flow:\n1. **FireClaw Deployment Wizard** — during onboarding (Next.js app)\n2. **OpenClaw Dashboard** — topbar key icon popover (Lit UI)\n\n### Flow\n1. User opens provider panel → sees list of supported providers worldwide\n2. User selects a provider → enters their API key\n3. System **fetches live models** from the provider API using that key\n4. Models displayed with auto-select on the latest/best model\n5. User confirms → key + selected model saved to OpenClaw config\n6. Gateway restarts → agent uses the new model\n\n### Supported Providers (Worldwide)\n| Provider | Config Key | Models API | API Key Prefix |\n|----------|-----------|------------|----------------|\n| OpenAI | openai | GET /v1/models | sk- |\n| Anthropic | anthropic | GET /v1/models | sk-ant- |\n| Google Gemini | google | GET /v1beta/models | AI... |\n| Groq | groq | GET /openai/v1/models | gsk_ |\n| Mistral | mistral | GET /v1/models | — |\n| OpenRouter | openrouter | GET /api/v1/models | sk-or- |\n| DeepSeek | deepseek | GET /models | sk- |\n| Together AI | together | GET /v1/models | — |\n| Fireworks AI | fireworks | GET /inference/v1/models | — |\n| Perplexity | perplexity | GET /models | pplx- |\n| Cohere | cohere | GET /v2/models | — |\n| xAI (Grok) | xai | GET /v1/models | xai- |\n| MiniMax | minimax | GET /v1/models | — |\n| Cerebras | cerebras | GET /v1/models | csk- |\n| SambaNova | sambanova | GET /v1/models | — |\n| Qwen (DashScope) | qwen | GET /compatible-mode/v1/models | sk- |\n\n### Suggestion Banner\nShow a recommendation: \"Try MiniMax — powerful models at affordable pricing\" (shown when no key is configured).\n\n### Config Patch Format\n```json\n{\n  \"models\": {\n    \"providers\": {\n      \"<provider>\": {\n        \"apiKey\": \"<key>\",\n        \"api\": \"openai-completions\"\n      }\n    }\n  },\n  \"agents\": {\n    \"defaults\": {\n      \"model\": \"<provider>/<model-id>\"\n    }\n  }\n}\n```\n\n### Live Model Fetching\n- Client-side fetch to provider APIs (most support CORS or use OpenAI-compatible endpoints)\n- Fallback: proxy through OpenClaw gateway if CORS blocked\n- Parse response → display model list → auto-select latest\n- Show model metadata: name, context window, pricing tier\n\n---\n\n## Infrastructure Roadmap\n\n### Phase A: Shared VPS Multi-Tenancy (Cost Optimisation)\n\n**Problem:** Current architecture creates one Hetzner VPS per agent (€3.79/agent on CX21).\nStarter plan at $7/agent barely covers infra cost. Margins are too thin.\n\n**Solution:** Shared VPS — pack 3 agents onto one CX32.\n\n#### Architecture\n\n```\nCX32 host (4 vCPU, 8GB, €9.29/mo)  ← \"cluster server\"\n├── nginx (port 443) — routes by subdomain\n├── agent-abc123  (openclaw container, port 18789, 2GB RAM limit)\n├── agent-def456  (openclaw container, port 18790, 2GB RAM limit)\n└── agent-ghi789  (openclaw container, port 18791, 2GB RAM limit)\n```\n\nEach agent gets subdomain: `{agent-id}.agents.fireclaw.app` → nginx → `localhost:1878N`\n\n#### Cost comparison\n\n| Model | Server | Cost/agent |\n|-------|--------|-----------|\n| Current (dedicated) | CX21 per agent | €3.79 |\n| Shared Starter | 3 agents / CX32 | €3.10 |\n| Shared Growth | 5 agents / CX42 | €2.90 |\n\n#### DB changes needed\n\nNew `hosts` collection:\n```ts\n{\n  _id: ObjectId,\n  serverId: number,        // Hetzner server ID\n  serverIp: string,\n  tier: \"shared\" | \"dedicated\",\n  maxSlots: number,        // 3 for CX32, 5 for CX42\n  usedSlots: number,\n  region: string,\n  status: \"active\" | \"draining\" | \"decommissioned\"\n}\n```\n\n`provision.ts` logic:\n1. If plan === \"starter\" → find host with `usedSlots < maxSlots` in same region\n2. If none found → create new CX32, add to `hosts`, set `maxSlots: 3`\n3. Run docker container on host at next available port (18789 + usedSlots)\n4. Increment `usedSlots`\n5. If plan === \"growth\"+ → create dedicated CX21 as before\n\n#### Plan tiers\n\n| Plan | Infrastructure |\n|------|---------------|\n| Starter ($7/agent) | Shared CX32, 2GB RAM cap, 3 agents/server |\n| Growth ($21 for 3) | Dedicated CX21 per agent |\n| Agency ($70 for 10) | Dedicated CX32 per agent |\n\n---\n\n### Phase B: Fireclaw Platform on Hetzner (No Cold Starts)\n\n**Problem:** Vercel cold starts (~500ms) hurt interactive dashboard. Vercel Pro + functions can spike to $100+/mo at scale. `maxDuration=800` is expensive on Vercel.\n\n**Solution:** Self-host Fireclaw platform on Hetzner using Coolify.\n\n#### Architecture\n\n```\nHetzner CX42 (8 vCPU / 16GB / €29/mo)   ← \"platform server\"\n├── Coolify (open-source PaaS, free)\n│   ├── fireclaw-web  (Next.js, PM2, always-on)\n│   ├── mongodb       (replica, backups to R2)\n│   └── caddy         (SSL termination, routing)\n└── GitHub webhook → Coolify → zero-downtime deploy\n```\n\n#### Benefits vs Vercel\n\n| | Vercel | Hetzner + Coolify |\n|---|---|---|\n| Monthly cost | $20+ (spikes) | €29 fixed |\n| Cold starts | 200–800ms | None (PM2 always-on) |\n| Long-running functions | $$ (maxDuration) | Free (it's just Node) |\n| Deploy | Push to main | Push to main (same) |\n| SSL | Automatic | Automatic (Caddy) |\n| Logs | Vercel dashboard | Coolify dashboard |\n\n#### Migration steps\n1. Provision CX42 in Frankfurt (same region as agent VPSes)\n2. Install Coolify: `curl -fsSL https://get.coolify.io | bash`\n3. Add GitHub repo → Coolify auto-detects Next.js\n4. Set all `.env.local` vars in Coolify environment panel\n5. Add DNS A record: `fireclaw.ai → CX42 IP`\n6. Test → cut over → remove Vercel project\n\n#### Storage\n- MongoDB on the same CX42 (with daily dumps to Cloudflare R2)\n- Agent media/files → Cloudflare R2 (free egress, S3-compatible)\n\n---\n\n### Phase C: OpenClaw Upstream PR — Multi-Provider API Key\n\n**Status:** Draft PR openclaw/openclaw#83570 — CLOSED pending fixes.\n\n**Fixes needed before re-opening:**\n\n1. `api` type mapping — most providers (Groq, Mistral, Together, etc.) use the `/v1/chat/completions` endpoint which maps to `openai-completions`, NOT `openai-responses`. Only actual OpenAI should use `openai-responses`. Anthropic stays `anthropic-messages`.\n\n2. Verify Google Gemini — needs `google-vertex` or `openai-completions` depending on endpoint used.\n\n3. End-to-end test on a live OpenClaw instance before re-opening.\n\n**Correct `api` mapping:**\n```\nopenai       → openai-responses      (Responses API)\nanthropic    → anthropic-messages\ngoogle       → openai-completions    (Gemini OpenAI-compat endpoint)\ngroq         → openai-completions\nmistral      → openai-completions\nopenrouter   → openai-completions\ndeepseek     → openai-completions\ntogether     → openai-completions\nfireworks    → openai-completions\nperplexity   → openai-completions\nxai          → openai-completions\nminimax      → openai-completions\nqwen         → openai-completions\ncohere       → openai-completions\nvoyage       → openai-completions    (embeddings only)\nreplicate    → openai-completions\n```\n\n---\n\n### Phase D: MCP Server + CLI\n\n#### MCP Server (`app/api/mcp/route.ts`)\nLets Claude Desktop, Cursor, and any MCP client manage Fireclaw agents programmatically.\n\nTools to expose:\n- `deploy_agent(template, plan, name)` → starts provisioning, returns agent ID\n- `list_agents()` → all agents for authenticated user\n- `get_agent_status(id)` → status, logs, domain\n- `update_agent_prompt(id, prompt)` → live system prompt push\n- `delete_agent(id)` → destroys VPS and removes DB record\n\nAuth: Bearer token from user profile API keys table.\n\n#### CLI (`packages/cli/`)\nnpm package `fireclaw-cli` for paid users.\n\n```bash\nnpx fireclaw login           # OAuth → stores API key in ~/.fireclaw\nnpx fireclaw deploy          # interactive: pick template → name → plan → deploy\nnpx fireclaw list            # table of all agents + status\nnpx fireclaw logs <id>       # stream provision + runtime logs\nnpx fireclaw ssh <id>        # proxy SSH to agent VPS (Growth+ plan)\nnpx fireclaw delete <id>     # destroy agent\n```\n\nDB addition needed: `api_keys` collection per user (hashed, last-used timestamp).\n"}, "3cb0b789e6087be3efc0dbf131fef90798c17bd2": {"t": 1785771360.105774, "n": 16035, "h": "9c945bfc6092f13e", "c": "# 🔥 FireClaw — The Plan (Research-Backed v2)\n\n## Current Priority (Updated May 2026)\n\n### Deployment is the #1 focus — this is what brings business.\n\n**What we deploy:**\n1. **OpenClaw** (our modified version) — Priority #1. Verify it works end-to-end.\n2. **Ad-Chain Verify Agent** (OpenClaw base + ad-tech modifications) — Priority #2. Client waiting.\n   - Pricing follows ad-tech market rates (not same as OpenClaw)\n   - Better UI/UX specific to ad verification workflows\n   - Scripts in `/scripts` folder contain initial attempts\n3. **SEO Agent** (OpenClaw base + SEO capabilities) — Priority #3.\n   - Site audits, auto backlink building, ranking analysis\n   - Connect with user's site for ongoing monitoring\n4. **All other agents** — Marked as \"Coming Soon\" in the marketplace\n\n**Deployment UX:**\n- User chooses an agent from the marketplace (catalog page)\n- Goes to a dedicated page for that agent type\n- Clicks deploy → pays → sees \"Your agent is deploying\" status\n- Dashboard shows deployed agents only (no deployment wizard inside dashboard)\n\n**Branch strategy:** Only working in `main` branch. Remove unused branches.\n\n---\n\n## What Is FireClaw?\n\n**FireClaw is the multi-agent management platform for OpenClaw.**\n\nDeploy, manage, and monitor multiple AI agents from one dashboard. Built for agencies, teams, and power users who need more than a single agent.\n\nThink: **Vercel for AI agents** — not another Hostinger. \n\n---\n\n## Why Not Just Use Hostinger?\n\nHostinger offers 1-click OpenClaw at **$5.99/mo** with 100K+ agents deployed. They win on:\n- Price ($5.99 vs anything we could charge for single agent)\n- Brand trust (4M+ users)\n- Infrastructure (global data centers)\n\n**But Hostinger can't do:**\n- ❌ Manage 10 agents from one dashboard\n- ❌ White-label agents for your clients\n- ❌ Pre-built agent templates (sales bot, support bot)\n- ❌ Cross-agent analytics\n- ❌ Agency billing (charge your clients through your dashboard)\n- ❌ Agent marketplace / skill store\n- ❌ India-first pricing in ₹\n\n**FireClaw's wedge: Multi-agent management, agency tools, and templates.**\n\n---\n\n## The Pitch (30 seconds)\n\n> \"Managing AI agents for your clients? Deploy 10 OpenClaw agents in 5 minutes, manage them all from one dashboard, white-label them, and charge your clients — starting at ₹1,999/month.\"\n\n---\n\n## Revenue Model (Research-Backed)\n\n### Pricing (India-First)\n\n| Plan | Agents | Price/mo | Infra Cost | Margin |\n|------|--------|----------|------------|--------|\n| **Personal** | 1 | ₹499 (~$6) | ~$8 (shared) | ~25% |\n| **Business** | 3 | ₹1,999 (~$24) | ~$20 (shared) | ~17% |\n| **Agency** | 10 | ₹4,999 (~$60) | ~$50 (shared) | ~17% |\n| **Enterprise** | Unlimited | Custom | Custom | 30%+ |\n\n**Why these prices work:**\n- Personal at ₹499 matches Hostinger's $5.99 in value\n- Business/Agency have NO equivalent competitor (Hostinger = 1 agent/plan)\n- Margins improve with density (more agents per server)\n\n### Infrastructure Strategy (NOT 1 EC2 per agent)\n\nInstead of dedicated instances, use **shared containerized infrastructure:**\n\n| Server | Type | Cost/mo | Agents Hosted | Cost/Agent |\n|--------|------|---------|---------------|------------|\n| Shared | t3.xlarge (4 vCPU, 16 GB) | ~$122 | 10-15 | $8-12 |\n| Shared | t3.2xlarge (8 vCPU, 32 GB) | ~$244 | 25-30 | $8-10 |\n| Dedicated | t3.micro (for personal) | ~$11 | 1 | $11 |\n\n*With 1yr Reserved Instances: costs drop 35-40%.*\n\n### Add-On Revenue\n| Add-On | Price |\n|--------|-------|\n| Extra agent | ₹299/mo |\n| WhatsApp messages (over 1000/mo) | ₹1/message |\n| Premium templates | ₹499-999 one-time |\n| White-label branding | Included in Agency+ |\n| Bundled AI credits (no BYOK needed) | ₹499/mo for 10K messages |\n\n### Revenue Projections\n\n| Milestone | Users | MRR | ARR |\n|-----------|-------|-----|-----|\n| Month 3 | 60 | ₹75K (~$900) | ~$10.8K |\n| Month 6 | 250 | ₹300K (~$3.6K) | ~$43K |\n| Month 12 | 700 | ₹1M (~$12K) | ~$144K |\n| Month 24 | 3,000 | ₹5M (~$60K) | ~$720K |\n\n---\n\n## Target Customers (Revised)\n\n### Phase 1 — Indian Agencies & Freelancers (Month 1-3)\n- Digital marketing agencies deploying chatbots for clients\n- Freelance developers offering \"AI assistant\" as a service\n- **Why they pay:** Managing 5-10 client agents from one place saves hours/week\n\n### Phase 2 — SMBs & Startups (Month 3-6)\n- Small businesses wanting AI for support/sales\n- Startups that need AI agents across departments\n- **Why they pay:** Templates get them started in minutes, not days\n\n### Phase 3 — Global Agencies (Month 6+)\n- International agencies managing AI fleets\n- White-label resellers\n- **Why they pay:** White-label + client billing = new revenue stream for them\n\n---\n\n## Competitive Landscape (Research-Backed)\n\n| Competitor | What | Price | Gap |\n|-----------|------|-------|-----|\n| **Hostinger OpenClaw** | 1-click single agent | $5.99/mo | No multi-agent, no agency tools |\n| **Dante AI** | AI chatbots | $40-400/mo | Chatbot only, not full agent, expensive |\n| **Botpress** | Chatbot builder | $0-495/mo | Visual builder, not OpenClaw, no WhatsApp focus |\n| **Chatfuel** | WhatsApp/IG bots | $69/mo | Simple bots, not AI agents |\n| **Tidio** | Customer service AI | $24-749/mo | Support-focused, not general agents |\n| **ManyChat** | Marketing automation | $15-100/mo | Flow-based, not AI agents |\n| **Railway/Render** | Generic PaaS | $5-85/mo | Not AI-specific, complex |\n\n**FireClaw's unique position:** Only platform for managing multiple OpenClaw agents with agency tools.\n\n---\n\n## Phases (Revised)\n\n### Phase 1: MVP — Agency Dashboard (3 weeks)\n- Landing page targeting Indian agencies\n- Multi-agent deploy from one dashboard\n- 3 pre-built templates (sales bot, support bot, personal assistant)\n- Stripe + Razorpay billing\n- 10-20 beta agencies\n- **Goal:** Prove agencies will pay for multi-agent management\n\n### Phase 2: Templates & Marketplace (3 weeks)\n- 10+ agent templates\n- Skill marketplace (install skills per agent)\n- White-label setup\n- Client billing (agencies charge their clients through us)\n- **Goal:** 100 paying users\n\n### Phase 3: Scale (Ongoing)\n- Multi-region (Mumbai, Singapore, US-East)\n- Agent analytics dashboard\n- Mobile app for monitoring\n- API for programmatic agent management\n- Community template submissions\n- **Goal:** ₹1M MRR\n\n---\n\n## Tech Stack (Revised)\n\n### Frontend (fireclaw.ai)\n- Next.js 16 + TypeScript\n- Tailwind CSS + shadcn/ui\n- Stripe + Razorpay for payments\n- authjs for authentication \n\n### Backend (Provisioning Engine)\n- Node.js API\n- **Docker + Docker Compose** (not bare EC2 per agent)\n- AWS SDK v3 (EC2 for host servers, not per agent)\n- MongoDB Atlas (user data, agent registry)\n- Bull/BullMQ (provisioning queue)\n\n### Agent Infrastructure\n- **Shared host servers** running Docker\n- Each agent = isolated Docker container with OpenClaw\n- Caddy reverse proxy (routing subdomains to containers)\n- Automated health checks + restart\n- Central logging (agent metrics → dashboard)\n\n### Why Docker Over Bare EC2\n| Approach | Cost for 10 agents | Management |\n|----------|-------------------|------------|\n| 10x t3.micro | ~$106/mo | 10 servers to manage |\n| 1x t3.xlarge + Docker | ~$125/mo | 1 server, 10 containers |\n| 2x t3.large + Docker | ~$132/mo | 2 servers, 5 containers each (HA) |\n\nDocker is similar cost but dramatically simpler to manage and scale.\n\n---\n\n## Key Risks & Mitigations\n\n| Risk | Mitigation |\n|------|-----------|\n| Hostinger drops price or adds multi-agent | Move fast, build community moat |\n| OpenClaw changes licensing | We contribute upstream, stay aligned |\n| Low margins on Personal plan | Personal is acquisition funnel → upsell to Business/Agency |\n| Container density limits | Auto-scale to new host servers |\n| India market slow to adopt | Parallel global launch at USD pricing |\n\n---\n\n## Key Metrics\n\n- **Agents deployed** — total across all users\n- **Agents per user** — avg (target: 2.5+)\n- **MRR** — monthly recurring revenue\n- **Provisioning time** — target: <60 seconds\n- **Uptime per agent** — target: 99.5%+\n- **Churn** — target: <5%/month\n- **CAC** — cost to acquire (target: <₹500)\n- **LTV** — lifetime value (target: >₹15,000)\n\n---\n\n## Multi-Provider API Key + Live Model Selection\n\n### Overview\nTwo places users configure AI providers with the same flow:\n1. **FireClaw Deployment Wizard** — during onboarding (Next.js app)\n2. **OpenClaw Dashboard** — topbar key icon popover (Lit UI)\n\n### Flow\n1. User opens provider panel → sees list of supported providers worldwide\n2. User selects a provider → enters their API key\n3. System **fetches live models** from the provider API using that key\n4. Models displayed with auto-select on the latest/best model\n5. User confirms → key + selected model saved to OpenClaw config\n6. Gateway restarts → agent uses the new model\n\n### Supported Providers (Worldwide)\n| Provider | Config Key | Models API | API Key Prefix |\n|----------|-----------|------------|----------------|\n| OpenAI | openai | GET /v1/models | sk- |\n| Anthropic | anthropic | GET /v1/models | sk-ant- |\n| Google Gemini | google | GET /v1beta/models | AI... |\n| Groq | groq | GET /openai/v1/models | gsk_ |\n| Mistral | mistral | GET /v1/models | — |\n| OpenRouter | openrouter | GET /api/v1/models | sk-or- |\n| DeepSeek | deepseek | GET /models | sk- |\n| Together AI | together | GET /v1/models | — |\n| Fireworks AI | fireworks | GET /inference/v1/models | — |\n| Perplexity | perplexity | GET /models | pplx- |\n| Cohere | cohere | GET /v2/models | — |\n| xAI (Grok) | xai | GET /v1/models | xai- |\n| MiniMax | minimax | GET /v1/models | — |\n| Cerebras | cerebras | GET /v1/models | csk- |\n| SambaNova | sambanova | GET /v1/models | — |\n| Qwen (DashScope) | qwen | GET /compatible-mode/v1/models | sk- |\n\n### Suggestion Banner\nShow a recommendation: \"Try MiniMax — powerful models at affordable pricing\" (shown when no key is configured).\n\n### Config Patch Format\n```json\n{\n  \"models\": {\n    \"providers\": {\n      \"<provider>\": {\n        \"apiKey\": \"<key>\",\n        \"api\": \"openai-completions\"\n      }\n    }\n  },\n  \"agents\": {\n    \"defaults\": {\n      \"model\": \"<provider>/<model-id>\"\n    }\n  }\n}\n```\n\n### Live Model Fetching\n- Client-side fetch to provider APIs (most support CORS or use OpenAI-compatible endpoints)\n- Fallback: proxy through OpenClaw gateway if CORS blocked\n- Parse response → display model list → auto-select latest\n- Show model metadata: name, context window, pricing tier\n\n---\n\n## Infrastructure Roadmap\n\n### Phase A: Shared VPS Multi-Tenancy (Cost Optimisation)\n\n**Problem:** Current architecture creates one Hetzner VPS per agent (€3.79/agent on CX21).\nStarter plan at $7/agent barely covers infra cost. Margins are too thin.\n\n**Solution:** Shared VPS — pack 3 agents onto one CX32.\n\n#### Architecture\n\n```\nCX32 host (4 vCPU, 8GB, €9.29/mo)  ← \"cluster server\"\n├── nginx (port 443) — routes by subdomain\n├── agent-abc123  (openclaw container, port 18789, 2GB RAM limit)\n├── agent-def456  (openclaw container, port 18790, 2GB RAM limit)\n└── agent-ghi789  (openclaw container, port 18791, 2GB RAM limit)\n```\n\nEach agent gets subdomain: `{agent-id}.agents.fireclaw.app` → nginx → `localhost:1878N`\n\n#### Cost comparison\n\n| Model | Server | Cost/agent |\n|-------|--------|-----------|\n| Current (dedicated) | CX21 per agent | €3.79 |\n| Shared Starter | 3 agents / CX32 | €3.10 |\n| Shared Growth | 5 agents / CX42 | €2.90 |\n\n#### DB changes needed\n\nNew `hosts` collection:\n```ts\n{\n  _id: ObjectId,\n  serverId: number,        // Hetzner server ID\n  serverIp: string,\n  tier: \"shared\" | \"dedicated\",\n  maxSlots: number,        // 3 for CX32, 5 for CX42\n  usedSlots: number,\n  region: string,\n  status: \"active\" | \"draining\" | \"decommissioned\"\n}\n```\n\n`provision.ts` logic:\n1. If plan === \"starter\" → find host with `usedSlots < maxSlots` in same region\n2. If none found → create new CX32, add to `hosts`, set `maxSlots: 3`\n3. Run docker container on host at next available port (18789 + usedSlots)\n4. Increment `usedSlots`\n5. If plan === \"growth\"+ → create dedicated CX21 as before\n\n#### Plan tiers\n\n| Plan | Infrastructure |\n|------|---------------|\n| Starter ($7/agent) | Shared CX32, 2GB RAM cap, 3 agents/server |\n| Growth ($21 for 3) | Dedicated CX21 per agent |\n| Agency ($70 for 10) | Dedicated CX32 per agent |\n\n---\n\n### Phase B: Fireclaw Platform on Hetzner (No Cold Starts)\n\n**Problem:** Vercel cold starts (~500ms) hurt interactive dashboard. Vercel Pro + functions can spike to $100+/mo at scale. `maxDuration=800` is expensive on Vercel.\n\n**Solution:** Self-host Fireclaw platform on Hetzner using Coolify.\n\n#### Architecture\n\n```\nHetzner CX42 (8 vCPU / 16GB / €29/mo)   ← \"platform server\"\n├── Coolify (open-source PaaS, free)\n│   ├── fireclaw-web  (Next.js, PM2, always-on)\n│   ├── mongodb       (replica, backups to R2)\n│   └── caddy         (SSL termination, routing)\n└── GitHub webhook → Coolify → zero-downtime deploy\n```\n\n#### Benefits vs Vercel\n\n| | Vercel | Hetzner + Coolify |\n|---|---|---|\n| Monthly cost | $20+ (spikes) | €29 fixed |\n| Cold starts | 200–800ms | None (PM2 always-on) |\n| Long-running functions | $$ (maxDuration) | Free (it's just Node) |\n| Deploy | Push to main | Push to main (same) |\n| SSL | Automatic | Automatic (Caddy) |\n| Logs | Vercel dashboard | Coolify dashboard |\n\n#### Migration steps\n1. Provision CX42 in Frankfurt (same region as agent VPSes)\n2. Install Coolify: `curl -fsSL https://get.coolify.io | bash`\n3. Add GitHub repo → Coolify auto-detects Next.js\n4. Set all `.env.local` vars in Coolify environment panel\n5. Add DNS A record: `fireclaw.ai → CX42 IP`\n6. Test → cut over → remove Vercel project\n\n#### Storage\n- MongoDB on the same CX42 (with daily dumps to Cloudflare R2)\n- Agent media/files → Cloudflare R2 (free egress, S3-compatible)\n\n---\n\n### Phase C: OpenClaw Upstream PR — Multi-Provider API Key\n\n**Status:** Draft PR openclaw/openclaw#83570 — CLOSED pending fixes.\n\n**Fixes needed before re-opening:**\n\n1. `api` type mapping — most providers (Groq, Mistral, Together, etc.) use the `/v1/chat/completions` endpoint which maps to `openai-completions`, NOT `openai-responses`. Only actual OpenAI should use `openai-responses`. Anthropic stays `anthropic-messages`.\n\n2. Verify Google Gemini — needs `google-vertex` or `openai-completions` depending on endpoint used.\n\n3. End-to-end test on a live OpenClaw instance before re-opening.\n\n**Correct `api` mapping:**\n```\nopenai       → openai-responses      (Responses API)\nanthropic    → anthropic-messages\ngoogle       → openai-completions    (Gemini OpenAI-compat endpoint)\ngroq         → openai-completions\nmistral      → openai-completions\nopenrouter   → openai-completions\ndeepseek     → openai-completions\ntogether     → openai-completions\nfireworks    → openai-completions\nperplexity   → openai-completions\nxai          → openai-completions\nminimax      → openai-completions\nqwen         → openai-completions\ncohere       → openai-completions\nvoyage       → openai-completions    (embeddings only)\nreplicate    → openai-completions\n```\n\n---\n\n### Phase D: MCP Server + CLI\n\n#### MCP Server (`app/api/mcp/route.ts`)\nLets Claude Desktop, Cursor, and any MCP client manage Fireclaw agents programmatically.\n\nTools to expose:\n- `deploy_agent(template, plan, name)` → starts provisioning, returns agent ID\n- `list_agents()` → all agents for authenticated user\n- `get_agent_status(id)` → status, logs, domain\n- `update_agent_prompt(id, prompt)` → live system prompt push\n- `delete_agent(id)` → destroys VPS and removes DB record\n\nAuth: Bearer token from user profile API keys table.\n\n#### CLI (`packages/cli/`)\nnpm package `fireclaw-cli` for paid users.\n\n```bash\nnpx fireclaw login           # OAuth → stores API key in ~/.fireclaw\nnpx fireclaw deploy          # interactive: pick template → name → plan → deploy\nnpx fireclaw list            # table of all agents + status\nnpx fireclaw logs <id>       # stream provision + runtime logs\nnpx fireclaw ssh <id>        # proxy SSH to agent VPS (Growth+ plan)\nnpx fireclaw delete <id>     # destroy agent\n```\n\nDB addition needed: `api_keys` collection per user (hashed, last-used timestamp).\n"}, "0e29786a91776737d9a9685c9bb47686e411e557": {"t": 1785771361.5876231, "n": 16727, "h": "6a61b216dab0a611", "c": "# 🔥 FireClaw — The Plan (Research-Backed v2)\n\n## What Is FireClaw?\n\n~~One-click deploy your personal AI assistant.~~ ← Hostinger already does this at $10.99/mo.plus some ai credits charge1q234adRY \n\n**FireClaw is the multi-agent management platform for OpenClaw.**\n\nDeploy, manage, and monitor multiple AI agents from one dashboard. Built for agencies, teams, and power users who need more than a single agent.\n\nThink: **Vercel for AI agents** — not another Hostinger.\nPLAN UPDATED TODAY --let me discuss with you what is my plan for future : lets make main the deployment great because all business will bring this, so what we will do..? 1. we will support deploy openclaw that is our modified version 2. we will let user to deply ad-chain verify agant (openclaw instance same but with more modification and pricing will be according to ad-tech market not same as openclaw also with some better ui/ux, i will provide clear instruction how  to build it, actaully i have tried it but you can check files in script folder but i will also tell you clear how to do it ) 3. we will support deployment of Seo Agent (same our openclaw our modified version but specific seo capabilities and work for seo , even connect with site , do audits and auto backlink building and more ), another agents we are showing in deployment wizard we will mark it as comming soon, now lets discuss how deployment should be look like , 1 thing, we will not show deployment in dashboard , the moment user choose his agent, he will go to the dedicated page of that agent where we will show him some deploy button and click on it he will go to pay money for that agent and after pay we will show that your agent is deploying , let me know is this idea good or it is good to do all in dashboard, now come to priority, the first thing we need to give priority is the deployment of openclaw (our modified verion is working or not) then second the deployment of ad-tech becuase i have client for that, now write all this in your plan.md file and be aware of plan and git , remove git not needed brnaches, becuase we are only working in main branch, \n\n---\n\n## Why Not Just Use Hostinger?\n\nHostinger offers 1-click OpenClaw at **$5.99/mo** with 100K+ agents deployed. They win on:\n- Price ($5.99 vs anything we could charge for single agent)\n- Brand trust (4M+ users)\n- Infrastructure (global data centers)\n\n**But Hostinger can't do:**\n- ❌ Manage 10 agents from one dashboard\n- ❌ White-label agents for your clients\n- ❌ Pre-built agent templates (sales bot, support bot)\n- ❌ Cross-agent analytics\n- ❌ Agency billing (charge your clients through your dashboard)\n- ❌ Agent marketplace / skill store\n- ❌ India-first pricing in ₹\n\n**FireClaw's wedge: Multi-agent management, agency tools, and templates.**\n\n---\n\n## The Pitch (30 seconds)\n\n> \"Managing AI agents for your clients? Deploy 10 OpenClaw agents in 5 minutes, manage them all from one dashboard, white-label them, and charge your clients — starting at ₹1,999/month.\"\n\n---\n\n## Revenue Model (Research-Backed)\n\n### Pricing (India-First)\n\n| Plan | Agents | Price/mo | Infra Cost | Margin |\n|------|--------|----------|------------|--------|\n| **Personal** | 1 | ₹499 (~$6) | ~$8 (shared) | ~25% |\n| **Business** | 3 | ₹1,999 (~$24) | ~$20 (shared) | ~17% |\n| **Agency** | 10 | ₹4,999 (~$60) | ~$50 (shared) | ~17% |\n| **Enterprise** | Unlimited | Custom | Custom | 30%+ |\n\n**Why these prices work:**\n- Personal at ₹499 matches Hostinger's $5.99 in value\n- Business/Agency have NO equivalent competitor (Hostinger = 1 agent/plan)\n- Margins improve with density (more agents per server)\n\n### Infrastructure Strategy (NOT 1 EC2 per agent)\n\nInstead of dedicated instances, use **shared containerized infrastructure:**\n\n| Server | Type | Cost/mo | Agents Hosted | Cost/Agent |\n|--------|------|---------|---------------|------------|\n| Shared | t3.xlarge (4 vCPU, 16 GB) | ~$122 | 10-15 | $8-12 |\n| Shared | t3.2xlarge (8 vCPU, 32 GB) | ~$244 | 25-30 | $8-10 |\n| Dedicated | t3.micro (for personal) | ~$11 | 1 | $11 |\n\n*With 1yr Reserved Instances: costs drop 35-40%.*\n\n### Add-On Revenue\n| Add-On | Price |\n|--------|-------|\n| Extra agent | ₹299/mo |\n| WhatsApp messages (over 1000/mo) | ₹1/message |\n| Premium templates | ₹499-999 one-time |\n| White-label branding | Included in Agency+ |\n| Bundled AI credits (no BYOK needed) | ₹499/mo for 10K messages |\n\n### Revenue Projections\n\n| Milestone | Users | MRR | ARR |\n|-----------|-------|-----|-----|\n| Month 3 | 60 | ₹75K (~$900) | ~$10.8K |\n| Month 6 | 250 | ₹300K (~$3.6K) | ~$43K |\n| Month 12 | 700 | ₹1M (~$12K) | ~$144K |\n| Month 24 | 3,000 | ₹5M (~$60K) | ~$720K |\n\n---\n\n## Target Customers (Revised)\n\n### Phase 1 — Indian Agencies & Freelancers (Month 1-3)\n- Digital marketing agencies deploying chatbots for clients\n- Freelance developers offering \"AI assistant\" as a service\n- **Why they pay:** Managing 5-10 client agents from one place saves hours/week\n\n### Phase 2 — SMBs & Startups (Month 3-6)\n- Small businesses wanting AI for support/sales\n- Startups that need AI agents across departments\n- **Why they pay:** Templates get them started in minutes, not days\n\n### Phase 3 — Global Agencies (Month 6+)\n- International agencies managing AI fleets\n- White-label resellers\n- **Why they pay:** White-label + client billing = new revenue stream for them\n\n---\n\n## Competitive Landscape (Research-Backed)\n\n| Competitor | What | Price | Gap |\n|-----------|------|-------|-----|\n| **Hostinger OpenClaw** | 1-click single agent | $5.99/mo | No multi-agent, no agency tools |\n| **Dante AI** | AI chatbots | $40-400/mo | Chatbot only, not full agent, expensive |\n| **Botpress** | Chatbot builder | $0-495/mo | Visual builder, not OpenClaw, no WhatsApp focus |\n| **Chatfuel** | WhatsApp/IG bots | $69/mo | Simple bots, not AI agents |\n| **Tidio** | Customer service AI | $24-749/mo | Support-focused, not general agents |\n| **ManyChat** | Marketing automation | $15-100/mo | Flow-based, not AI agents |\n| **Railway/Render** | Generic PaaS | $5-85/mo | Not AI-specific, complex |\n\n**FireClaw's unique position:** Only platform for managing multiple OpenClaw agents with agency tools.\n\n---\n\n## Phases (Revised)\n\n### Phase 1: MVP — Agency Dashboard (3 weeks)\n- Landing page targeting Indian agencies\n- Multi-agent deploy from one dashboard\n- 3 pre-built templates (sales bot, support bot, personal assistant)\n- Stripe + Razorpay billing\n- 10-20 beta agencies\n- **Goal:** Prove agencies will pay for multi-agent management\n\n### Phase 2: Templates & Marketplace (3 weeks)\n- 10+ agent templates\n- Skill marketplace (install skills per agent)\n- White-label setup\n- Client billing (agencies charge their clients through us)\n- **Goal:** 100 paying users\n\n### Phase 3: Scale (Ongoing)\n- Multi-region (Mumbai, Singapore, US-East)\n- Agent analytics dashboard\n- Mobile app for monitoring\n- API for programmatic agent management\n- Community template submissions\n- **Goal:** ₹1M MRR\n\n---\n\n## Tech Stack (Revised)\n\n### Frontend (fireclaw.ai)\n- Next.js 16 + TypeScript\n- Tailwind CSS + shadcn/ui\n- Stripe + Razorpay for payments\n- authjs for authentication \n\n### Backend (Provisioning Engine)\n- Node.js API\n- **Docker + Docker Compose** (not bare EC2 per agent)\n- AWS SDK v3 (EC2 for host servers, not per agent)\n- MongoDB Atlas (user data, agent registry)\n- Bull/BullMQ (provisioning queue)\n\n### Agent Infrastructure\n- **Shared host servers** running Docker\n- Each agent = isolated Docker container with OpenClaw\n- Caddy reverse proxy (routing subdomains to containers)\n- Automated health checks + restart\n- Central logging (agent metrics → dashboard)\n\n### Why Docker Over Bare EC2\n| Approach | Cost for 10 agents | Management |\n|----------|-------------------|------------|\n| 10x t3.micro | ~$106/mo | 10 servers to manage |\n| 1x t3.xlarge + Docker | ~$125/mo | 1 server, 10 containers |\n| 2x t3.large + Docker | ~$132/mo | 2 servers, 5 containers each (HA) |\n\nDocker is similar cost but dramatically simpler to manage and scale.\n\n---\n\n## Key Risks & Mitigations\n\n| Risk | Mitigation |\n|------|-----------|\n| Hostinger drops price or adds multi-agent | Move fast, build community moat |\n| OpenClaw changes licensing | We contribute upstream, stay aligned |\n| Low margins on Personal plan | Personal is acquisition funnel → upsell to Business/Agency |\n| Container density limits | Auto-scale to new host servers |\n| India market slow to adopt | Parallel global launch at USD pricing |\n\n---\n\n## Key Metrics\n\n- **Agents deployed** — total across all users\n- **Agents per user** — avg (target: 2.5+)\n- **MRR** — monthly recurring revenue\n- **Provisioning time** — target: <60 seconds\n- **Uptime per agent** — target: 99.5%+\n- **Churn** — target: <5%/month\n- **CAC** — cost to acquire (target: <₹500)\n- **LTV** — lifetime value (target: >₹15,000)\n\n---\n\n## Multi-Provider API Key + Live Model Selection\n\n### Overview\nTwo places users configure AI providers with the same flow:\n1. **FireClaw Deployment Wizard** — during onboarding (Next.js app)\n2. **OpenClaw Dashboard** — topbar key icon popover (Lit UI)\n\n### Flow\n1. User opens provider panel → sees list of supported providers worldwide\n2. User selects a provider → enters their API key\n3. System **fetches live models** from the provider API using that key\n4. Models displayed with auto-select on the latest/best model\n5. User confirms → key + selected model saved to OpenClaw config\n6. Gateway restarts → agent uses the new model\n\n### Supported Providers (Worldwide)\n| Provider | Config Key | Models API | API Key Prefix |\n|----------|-----------|------------|----------------|\n| OpenAI | openai | GET /v1/models | sk- |\n| Anthropic | anthropic | GET /v1/models | sk-ant- |\n| Google Gemini | google | GET /v1beta/models | AI... |\n| Groq | groq | GET /openai/v1/models | gsk_ |\n| Mistral | mistral | GET /v1/models | — |\n| OpenRouter | openrouter | GET /api/v1/models | sk-or- |\n| DeepSeek | deepseek | GET /models | sk- |\n| Together AI | together | GET /v1/models | — |\n| Fireworks AI | fireworks | GET /inference/v1/models | — |\n| Perplexity | perplexity | GET /models | pplx- |\n| Cohere | cohere | GET /v2/models | — |\n| xAI (Grok) | xai | GET /v1/models | xai- |\n| MiniMax | minimax | GET /v1/models | — |\n| Cerebras | cerebras | GET /v1/models | csk- |\n| SambaNova | sambanova | GET /v1/models | — |\n| Qwen (DashScope) | qwen | GET /compatible-mode/v1/models | sk- |\n\n### Suggestion Banner\nShow a recommendation: \"Try MiniMax — powerful models at affordable pricing\" (shown when no key is configured).\n\n### Config Patch Format\n```json\n{\n  \"models\": {\n    \"providers\": {\n      \"<provider>\": {\n        \"apiKey\": \"<key>\",\n        \"api\": \"openai-completions\"\n      }\n    }\n  },\n  \"agents\": {\n    \"defaults\": {\n      \"model\": \"<provider>/<model-id>\"\n    }\n  }\n}\n```\n\n### Live Model Fetching\n- Client-side fetch to provider APIs (most support CORS or use OpenAI-compatible endpoints)\n- Fallback: proxy through OpenClaw gateway if CORS blocked\n- Parse response → display model list → auto-select latest\n- Show model metadata: name, context window, pricing tier\n\n---\n\n## Infrastructure Roadmap\n\n### Phase A: Shared VPS Multi-Tenancy (Cost Optimisation)\n\n**Problem:** Current architecture creates one Hetzner VPS per agent (€3.79/agent on CX21).\nStarter plan at $7/agent barely covers infra cost. Margins are too thin.\n\n**Solution:** Shared VPS — pack 3 agents onto one CX32.\n\n#### Architecture\n\n```\nCX32 host (4 vCPU, 8GB, €9.29/mo)  ← \"cluster server\"\n├── nginx (port 443) — routes by subdomain\n├── agent-abc123  (openclaw container, port 18789, 2GB RAM limit)\n├── agent-def456  (openclaw container, port 18790, 2GB RAM limit)\n└── agent-ghi789  (openclaw container, port 18791, 2GB RAM limit)\n```\n\nEach agent gets subdomain: `{agent-id}.agents.fireclaw.app` → nginx → `localhost:1878N`\n\n#### Cost comparison\n\n| Model | Server | Cost/agent |\n|-------|--------|-----------|\n| Current (dedicated) | CX21 per agent | €3.79 |\n| Shared Starter | 3 agents / CX32 | €3.10 |\n| Shared Growth | 5 agents / CX42 | €2.90 |\n\n#### DB changes needed\n\nNew `hosts` collection:\n```ts\n{\n  _id: ObjectId,\n  serverId: number,        // Hetzner server ID\n  serverIp: string,\n  tier: \"shared\" | \"dedicated\",\n  maxSlots: number,        // 3 for CX32, 5 for CX42\n  usedSlots: number,\n  region: string,\n  status: \"active\" | \"draining\" | \"decommissioned\"\n}\n```\n\n`provision.ts` logic:\n1. If plan === \"starter\" → find host with `usedSlots < maxSlots` in same region\n2. If none found → create new CX32, add to `hosts`, set `maxSlots: 3`\n3. Run docker container on host at next available port (18789 + usedSlots)\n4. Increment `usedSlots`\n5. If plan === \"growth\"+ → create dedicated CX21 as before\n\n#### Plan tiers\n\n| Plan | Infrastructure |\n|------|---------------|\n| Starter ($7/agent) | Shared CX32, 2GB RAM cap, 3 agents/server |\n| Growth ($21 for 3) | Dedicated CX21 per agent |\n| Agency ($70 for 10) | Dedicated CX32 per agent |\n\n---\n\n### Phase B: Fireclaw Platform on Hetzner (No Cold Starts)\n\n**Problem:** Vercel cold starts (~500ms) hurt interactive dashboard. Vercel Pro + functions can spike to $100+/mo at scale. `maxDuration=800` is expensive on Vercel.\n\n**Solution:** Self-host Fireclaw platform on Hetzner using Coolify.\n\n#### Architecture\n\n```\nHetzner CX42 (8 vCPU / 16GB / €29/mo)   ← \"platform server\"\n├── Coolify (open-source PaaS, free)\n│   ├── fireclaw-web  (Next.js, PM2, always-on)\n│   ├── mongodb       (replica, backups to R2)\n│   └── caddy         (SSL termination, routing)\n└── GitHub webhook → Coolify → zero-downtime deploy\n```\n\n#### Benefits vs Vercel\n\n| | Vercel | Hetzner + Coolify |\n|---|---|---|\n| Monthly cost | $20+ (spikes) | €29 fixed |\n| Cold starts | 200–800ms | None (PM2 always-on) |\n| Long-running functions | $$ (maxDuration) | Free (it's just Node) |\n| Deploy | Push to main | Push to main (same) |\n| SSL | Automatic | Automatic (Caddy) |\n| Logs | Vercel dashboard | Coolify dashboard |\n\n#### Migration steps\n1. Provision CX42 in Frankfurt (same region as agent VPSes)\n2. Install Coolify: `curl -fsSL https://get.coolify.io | bash`\n3. Add GitHub repo → Coolify auto-detects Next.js\n4. Set all `.env.local` vars in Coolify environment panel\n5. Add DNS A record: `fireclaw.ai → CX42 IP`\n6. Test → cut over → remove Vercel project\n\n#### Storage\n- MongoDB on the same CX42 (with daily dumps to Cloudflare R2)\n- Agent media/files → Cloudflare R2 (free egress, S3-compatible)\n\n---\n\n### Phase C: OpenClaw Upstream PR — Multi-Provider API Key\n\n**Status:** Draft PR openclaw/openclaw#83570 — CLOSED pending fixes.\n\n**Fixes needed before re-opening:**\n\n1. `api` type mapping — most providers (Groq, Mistral, Together, etc.) use the `/v1/chat/completions` endpoint which maps to `openai-completions`, NOT `openai-responses`. Only actual OpenAI should use `openai-responses`. Anthropic stays `anthropic-messages`.\n\n2. Verify Google Gemini — needs `google-vertex` or `openai-completions` depending on endpoint used.\n\n3. End-to-end test on a live OpenClaw instance before re-opening.\n\n**Correct `api` mapping:**\n```\nopenai       → openai-responses      (Responses API)\nanthropic    → anthropic-messages\ngoogle       → openai-completions    (Gemini OpenAI-compat endpoint)\ngroq         → openai-completions\nmistral      → openai-completions\nopenrouter   → openai-completions\ndeepseek     → openai-completions\ntogether     → openai-completions\nfireworks    → openai-completions\nperplexity   → openai-completions\nxai          → openai-completions\nminimax      → openai-completions\nqwen         → openai-completions\ncohere       → openai-completions\nvoyage       → openai-completions    (embeddings only)\nreplicate    → openai-completions\n```\n\n---\n\n### Phase D: MCP Server + CLI\n\n#### MCP Server (`app/api/mcp/route.ts`)\nLets Claude Desktop, Cursor, and any MCP client manage Fireclaw agents programmatically.\n\nTools to expose:\n- `deploy_agent(template, plan, name)` → starts provisioning, returns agent ID\n- `list_agents()` → all agents for authenticated user\n- `get_agent_status(id)` → status, logs, domain\n- `update_agent_prompt(id, prompt)` → live system prompt push\n- `delete_agent(id)` → destroys VPS and removes DB record\n\nAuth: Bearer token from user profile API keys table.\n\n#### CLI (`packages/cli/`)\nnpm package `fireclaw-cli` for paid users.\n\n```bash\nnpx fireclaw login           # OAuth → stores API key in ~/.fireclaw\nnpx fireclaw deploy          # interactive: pick template → name → plan → deploy\nnpx fireclaw list            # table of all agents + status\nnpx fireclaw logs <id>       # stream provision + runtime logs\nnpx fireclaw ssh <id>        # proxy SSH to agent VPS (Growth+ plan)\nnpx fireclaw delete <id>     # destroy agent\n```\n\nDB addition needed: `api_keys` collection per user (hashed, last-used timestamp).\n"}, "57514ddc060621fa045526ffd3773760e5508761": {"t": 1785771362.8613694, "n": 15041, "h": "79c2b6ccfa7d6631", "c": "# 🔥 FireClaw — The Plan (Research-Backed v2)\n\n## What Is FireClaw?\n\n~~One-click deploy your personal AI assistant.~~ ← Hostinger already does this at $5.99/mo.\n\n**FireClaw is the multi-agent management platform for OpenClaw.**\n\nDeploy, manage, and monitor multiple AI agents from one dashboard. Built for agencies, teams, and power users who need more than a single agent.\n\nThink: **Vercel for AI agents** — not another Hostinger.\n\n---\n\n## Why Not Just Use Hostinger?\n\nHostinger offers 1-click OpenClaw at **$5.99/mo** with 100K+ agents deployed. They win on:\n- Price ($5.99 vs anything we could charge for single agent)\n- Brand trust (4M+ users)\n- Infrastructure (global data centers)\n\n**But Hostinger can't do:**\n- ❌ Manage 10 agents from one dashboard\n- ❌ White-label agents for your clients\n- ❌ Pre-built agent templates (sales bot, support bot)\n- ❌ Cross-agent analytics\n- ❌ Agency billing (charge your clients through your dashboard)\n- ❌ Agent marketplace / skill store\n- ❌ India-first pricing in ₹\n\n**FireClaw's wedge: Multi-agent management, agency tools, and templates.**\n\n---\n\n## The Pitch (30 seconds)\n\n> \"Managing AI agents for your clients? Deploy 10 OpenClaw agents in 5 minutes, manage them all from one dashboard, white-label them, and charge your clients — starting at ₹1,999/month.\"\n\n---\n\n## Revenue Model (Research-Backed)\n\n### Pricing (India-First)\n\n| Plan | Agents | Price/mo | Infra Cost | Margin |\n|------|--------|----------|------------|--------|\n| **Personal** | 1 | ₹499 (~$6) | ~$8 (shared) | ~25% |\n| **Business** | 3 | ₹1,999 (~$24) | ~$20 (shared) | ~17% |\n| **Agency** | 10 | ₹4,999 (~$60) | ~$50 (shared) | ~17% |\n| **Enterprise** | Unlimited | Custom | Custom | 30%+ |\n\n**Why these prices work:**\n- Personal at ₹499 matches Hostinger's $5.99 in value\n- Business/Agency have NO equivalent competitor (Hostinger = 1 agent/plan)\n- Margins improve with density (more agents per server)\n\n### Infrastructure Strategy (NOT 1 EC2 per agent)\n\nInstead of dedicated instances, use **shared containerized infrastructure:**\n\n| Server | Type | Cost/mo | Agents Hosted | Cost/Agent |\n|--------|------|---------|---------------|------------|\n| Shared | t3.xlarge (4 vCPU, 16 GB) | ~$122 | 10-15 | $8-12 |\n| Shared | t3.2xlarge (8 vCPU, 32 GB) | ~$244 | 25-30 | $8-10 |\n| Dedicated | t3.micro (for personal) | ~$11 | 1 | $11 |\n\n*With 1yr Reserved Instances: costs drop 35-40%.*\n\n### Add-On Revenue\n| Add-On | Price |\n|--------|-------|\n| Extra agent | ₹299/mo |\n| WhatsApp messages (over 1000/mo) | ₹1/message |\n| Premium templates | ₹499-999 one-time |\n| White-label branding | Included in Agency+ |\n| Bundled AI credits (no BYOK needed) | ₹499/mo for 10K messages |\n\n### Revenue Projections\n\n| Milestone | Users | MRR | ARR |\n|-----------|-------|-----|-----|\n| Month 3 | 60 | ₹75K (~$900) | ~$10.8K |\n| Month 6 | 250 | ₹300K (~$3.6K) | ~$43K |\n| Month 12 | 700 | ₹1M (~$12K) | ~$144K |\n| Month 24 | 3,000 | ₹5M (~$60K) | ~$720K |\n\n---\n\n## Target Customers (Revised)\n\n### Phase 1 — Indian Agencies & Freelancers (Month 1-3)\n- Digital marketing agencies deploying chatbots for clients\n- Freelance developers offering \"AI assistant\" as a service\n- **Why they pay:** Managing 5-10 client agents from one place saves hours/week\n\n### Phase 2 — SMBs & Startups (Month 3-6)\n- Small businesses wanting AI for support/sales\n- Startups that need AI agents across departments\n- **Why they pay:** Templates get them started in minutes, not days\n\n### Phase 3 — Global Agencies (Month 6+)\n- International agencies managing AI fleets\n- White-label resellers\n- **Why they pay:** White-label + client billing = new revenue stream for them\n\n---\n\n## Competitive Landscape (Research-Backed)\n\n| Competitor | What | Price | Gap |\n|-----------|------|-------|-----|\n| **Hostinger OpenClaw** | 1-click single agent | $5.99/mo | No multi-agent, no agency tools |\n| **Dante AI** | AI chatbots | $40-400/mo | Chatbot only, not full agent, expensive |\n| **Botpress** | Chatbot builder | $0-495/mo | Visual builder, not OpenClaw, no WhatsApp focus |\n| **Chatfuel** | WhatsApp/IG bots | $69/mo | Simple bots, not AI agents |\n| **Tidio** | Customer service AI | $24-749/mo | Support-focused, not general agents |\n| **ManyChat** | Marketing automation | $15-100/mo | Flow-based, not AI agents |\n| **Railway/Render** | Generic PaaS | $5-85/mo | Not AI-specific, complex |\n\n**FireClaw's unique position:** Only platform for managing multiple OpenClaw agents with agency tools.\n\n---\n\n## Phases (Revised)\n\n### Phase 1: MVP — Agency Dashboard (3 weeks)\n- Landing page targeting Indian agencies\n- Multi-agent deploy from one dashboard\n- 3 pre-built templates (sales bot, support bot, personal assistant)\n- Stripe + Razorpay billing\n- 10-20 beta agencies\n- **Goal:** Prove agencies will pay for multi-agent management\n\n### Phase 2: Templates & Marketplace (3 weeks)\n- 10+ agent templates\n- Skill marketplace (install skills per agent)\n- White-label setup\n- Client billing (agencies charge their clients through us)\n- **Goal:** 100 paying users\n\n### Phase 3: Scale (Ongoing)\n- Multi-region (Mumbai, Singapore, US-East)\n- Agent analytics dashboard\n- Mobile app for monitoring\n- API for programmatic agent management\n- Community template submissions\n- **Goal:** ₹1M MRR\n\n---\n\n## Tech Stack (Revised)\n\n### Frontend (fireclaw.ai)\n- Next.js 16 + TypeScript\n- Tailwind CSS + shadcn/ui\n- Stripe + Razorpay for payments\n- authjs for authentication \n\n### Backend (Provisioning Engine)\n- Node.js API\n- **Docker + Docker Compose** (not bare EC2 per agent)\n- AWS SDK v3 (EC2 for host servers, not per agent)\n- MongoDB Atlas (user data, agent registry)\n- Bull/BullMQ (provisioning queue)\n\n### Agent Infrastructure\n- **Shared host servers** running Docker\n- Each agent = isolated Docker container with OpenClaw\n- Caddy reverse proxy (routing subdomains to containers)\n- Automated health checks + restart\n- Central logging (agent metrics → dashboard)\n\n### Why Docker Over Bare EC2\n| Approach | Cost for 10 agents | Management |\n|----------|-------------------|------------|\n| 10x t3.micro | ~$106/mo | 10 servers to manage |\n| 1x t3.xlarge + Docker | ~$125/mo | 1 server, 10 containers |\n| 2x t3.large + Docker | ~$132/mo | 2 servers, 5 containers each (HA) |\n\nDocker is similar cost but dramatically simpler to manage and scale.\n\n---\n\n## Key Risks & Mitigations\n\n| Risk | Mitigation |\n|------|-----------|\n| Hostinger drops price or adds multi-agent | Move fast, build community moat |\n| OpenClaw changes licensing | We contribute upstream, stay aligned |\n| Low margins on Personal plan | Personal is acquisition funnel → upsell to Business/Agency |\n| Container density limits | Auto-scale to new host servers |\n| India market slow to adopt | Parallel global launch at USD pricing |\n\n---\n\n## Key Metrics\n\n- **Agents deployed** — total across all users\n- **Agents per user** — avg (target: 2.5+)\n- **MRR** — monthly recurring revenue\n- **Provisioning time** — target: <60 seconds\n- **Uptime per agent** — target: 99.5%+\n- **Churn** — target: <5%/month\n- **CAC** — cost to acquire (target: <₹500)\n- **LTV** — lifetime value (target: >₹15,000)\n\n---\n\n## Multi-Provider API Key + Live Model Selection\n\n### Overview\nTwo places users configure AI providers with the same flow:\n1. **FireClaw Deployment Wizard** — during onboarding (Next.js app)\n2. **OpenClaw Dashboard** — topbar key icon popover (Lit UI)\n\n### Flow\n1. User opens provider panel → sees list of supported providers worldwide\n2. User selects a provider → enters their API key\n3. System **fetches live models** from the provider API using that key\n4. Models displayed with auto-select on the latest/best model\n5. User confirms → key + selected model saved to OpenClaw config\n6. Gateway restarts → agent uses the new model\n\n### Supported Providers (Worldwide)\n| Provider | Config Key | Models API | API Key Prefix |\n|----------|-----------|------------|----------------|\n| OpenAI | openai | GET /v1/models | sk- |\n| Anthropic | anthropic | GET /v1/models | sk-ant- |\n| Google Gemini | google | GET /v1beta/models | AI... |\n| Groq | groq | GET /openai/v1/models | gsk_ |\n| Mistral | mistral | GET /v1/models | — |\n| OpenRouter | openrouter | GET /api/v1/models | sk-or- |\n| DeepSeek | deepseek | GET /models | sk- |\n| Together AI | together | GET /v1/models | — |\n| Fireworks AI | fireworks | GET /inference/v1/models | — |\n| Perplexity | perplexity | GET /models | pplx- |\n| Cohere | cohere | GET /v2/models | — |\n| xAI (Grok) | xai | GET /v1/models | xai- |\n| MiniMax | minimax | GET /v1/models | — |\n| Cerebras | cerebras | GET /v1/models | csk- |\n| SambaNova | sambanova | GET /v1/models | — |\n| Qwen (DashScope) | qwen | GET /compatible-mode/v1/models | sk- |\n\n### Suggestion Banner\nShow a recommendation: \"Try MiniMax — powerful models at affordable pricing\" (shown when no key is configured).\n\n### Config Patch Format\n```json\n{\n  \"models\": {\n    \"providers\": {\n      \"<provider>\": {\n        \"apiKey\": \"<key>\",\n        \"api\": \"openai-completions\"\n      }\n    }\n  },\n  \"agents\": {\n    \"defaults\": {\n      \"model\": \"<provider>/<model-id>\"\n    }\n  }\n}\n```\n\n### Live Model Fetching\n- Client-side fetch to provider APIs (most support CORS or use OpenAI-compatible endpoints)\n- Fallback: proxy through OpenClaw gateway if CORS blocked\n- Parse response → display model list → auto-select latest\n- Show model metadata: name, context window, pricing tier\n\n---\n\n## Infrastructure Roadmap\n\n### Phase A: Shared VPS Multi-Tenancy (Cost Optimisation)\n\n**Problem:** Current architecture creates one Hetzner VPS per agent (€3.79/agent on CX21).\nStarter plan at $7/agent barely covers infra cost. Margins are too thin.\n\n**Solution:** Shared VPS — pack 3 agents onto one CX32.\n\n#### Architecture\n\n```\nCX32 host (4 vCPU, 8GB, €9.29/mo)  ← \"cluster server\"\n├── nginx (port 443) — routes by subdomain\n├── agent-abc123  (openclaw container, port 18789, 2GB RAM limit)\n├── agent-def456  (openclaw container, port 18790, 2GB RAM limit)\n└── agent-ghi789  (openclaw container, port 18791, 2GB RAM limit)\n```\n\nEach agent gets subdomain: `{agent-id}.agents.fireclaw.app` → nginx → `localhost:1878N`\n\n#### Cost comparison\n\n| Model | Server | Cost/agent |\n|-------|--------|-----------|\n| Current (dedicated) | CX21 per agent | €3.79 |\n| Shared Starter | 3 agents / CX32 | €3.10 |\n| Shared Growth | 5 agents / CX42 | €2.90 |\n\n#### DB changes needed\n\nNew `hosts` collection:\n```ts\n{\n  _id: ObjectId,\n  serverId: number,        // Hetzner server ID\n  serverIp: string,\n  tier: \"shared\" | \"dedicated\",\n  maxSlots: number,        // 3 for CX32, 5 for CX42\n  usedSlots: number,\n  region: string,\n  status: \"active\" | \"draining\" | \"decommissioned\"\n}\n```\n\n`provision.ts` logic:\n1. If plan === \"starter\" → find host with `usedSlots < maxSlots` in same region\n2. If none found → create new CX32, add to `hosts`, set `maxSlots: 3`\n3. Run docker container on host at next available port (18789 + usedSlots)\n4. Increment `usedSlots`\n5. If plan === \"growth\"+ → create dedicated CX21 as before\n\n#### Plan tiers\n\n| Plan | Infrastructure |\n|------|---------------|\n| Starter ($7/agent) | Shared CX32, 2GB RAM cap, 3 agents/server |\n| Growth ($21 for 3) | Dedicated CX21 per agent |\n| Agency ($70 for 10) | Dedicated CX32 per agent |\n\n---\n\n### Phase B: Fireclaw Platform on Hetzner (No Cold Starts)\n\n**Problem:** Vercel cold starts (~500ms) hurt interactive dashboard. Vercel Pro + functions can spike to $100+/mo at scale. `maxDuration=800` is expensive on Vercel.\n\n**Solution:** Self-host Fireclaw platform on Hetzner using Coolify.\n\n#### Architecture\n\n```\nHetzner CX42 (8 vCPU / 16GB / €29/mo)   ← \"platform server\"\n├── Coolify (open-source PaaS, free)\n│   ├── fireclaw-web  (Next.js, PM2, always-on)\n│   ├── mongodb       (replica, backups to R2)\n│   └── caddy         (SSL termination, routing)\n└── GitHub webhook → Coolify → zero-downtime deploy\n```\n\n#### Benefits vs Vercel\n\n| | Vercel | Hetzner + Coolify |\n|---|---|---|\n| Monthly cost | $20+ (spikes) | €29 fixed |\n| Cold starts | 200–800ms | None (PM2 always-on) |\n| Long-running functions | $$ (maxDuration) | Free (it's just Node) |\n| Deploy | Push to main | Push to main (same) |\n| SSL | Automatic | Automatic (Caddy) |\n| Logs | Vercel dashboard | Coolify dashboard |\n\n#### Migration steps\n1. Provision CX42 in Frankfurt (same region as agent VPSes)\n2. Install Coolify: `curl -fsSL https://get.coolify.io | bash`\n3. Add GitHub repo → Coolify auto-detects Next.js\n4. Set all `.env.local` vars in Coolify environment panel\n5. Add DNS A record: `fireclaw.ai → CX42 IP`\n6. Test → cut over → remove Vercel project\n\n#### Storage\n- MongoDB on the same CX42 (with daily dumps to Cloudflare R2)\n- Agent media/files → Cloudflare R2 (free egress, S3-compatible)\n\n---\n\n### Phase C: OpenClaw Upstream PR — Multi-Provider API Key\n\n**Status:** Draft PR openclaw/openclaw#83570 — CLOSED pending fixes.\n\n**Fixes needed before re-opening:**\n\n1. `api` type mapping — most providers (Groq, Mistral, Together, etc.) use the `/v1/chat/completions` endpoint which maps to `openai-completions`, NOT `openai-responses`. Only actual OpenAI should use `openai-responses`. Anthropic stays `anthropic-messages`.\n\n2. Verify Google Gemini — needs `google-vertex` or `openai-completions` depending on endpoint used.\n\n3. End-to-end test on a live OpenClaw instance before re-opening.\n\n**Correct `api` mapping:**\n```\nopenai       → openai-responses      (Responses API)\nanthropic    → anthropic-messages\ngoogle       → openai-completions    (Gemini OpenAI-compat endpoint)\ngroq         → openai-completions\nmistral      → openai-completions\nopenrouter   → openai-completions\ndeepseek     → openai-completions\ntogether     → openai-completions\nfireworks    → openai-completions\nperplexity   → openai-completions\nxai          → openai-completions\nminimax      → openai-completions\nqwen         → openai-completions\ncohere       → openai-completions\nvoyage       → openai-completions    (embeddings only)\nreplicate    → openai-completions\n```\n\n---\n\n### Phase D: MCP Server + CLI\n\n#### MCP Server (`app/api/mcp/route.ts`)\nLets Claude Desktop, Cursor, and any MCP client manage Fireclaw agents programmatically.\n\nTools to expose:\n- `deploy_agent(template, plan, name)` → starts provisioning, returns agent ID\n- `list_agents()` → all agents for authenticated user\n- `get_agent_status(id)` → status, logs, domain\n- `update_agent_prompt(id, prompt)` → live system prompt push\n- `delete_agent(id)` → destroys VPS and removes DB record\n\nAuth: Bearer token from user profile API keys table.\n\n#### CLI (`packages/cli/`)\nnpm package `fireclaw-cli` for paid users.\n\n```bash\nnpx fireclaw login           # OAuth → stores API key in ~/.fireclaw\nnpx fireclaw deploy          # interactive: pick template → name → plan → deploy\nnpx fireclaw list            # table of all agents + status\nnpx fireclaw logs <id>       # stream provision + runtime logs\nnpx fireclaw ssh <id>        # proxy SSH to agent VPS (Growth+ plan)\nnpx fireclaw delete <id>     # destroy agent\n```\n\nDB addition needed: `api_keys` collection per user (hashed, last-used timestamp).\n"}, "306c9e01df72cf25a975e39f2b5f17e3e19ccb19": {"t": 1785771364.2203586, "n": 9371, "h": "b1d53a4700a6d12c", "c": "# 🔥 FireClaw — The Plan (Research-Backed v2)\n\n## What Is FireClaw?\n\n~~One-click deploy your personal AI assistant.~~ ← Hostinger already does this at $5.99/mo.\n\n**FireClaw is the multi-agent management platform for OpenClaw.**\n\nDeploy, manage, and monitor multiple AI agents from one dashboard. Built for agencies, teams, and power users who need more than a single agent.\n\nThink: **Vercel for AI agents** — not another Hostinger.\n\n---\n\n## Why Not Just Use Hostinger?\n\nHostinger offers 1-click OpenClaw at **$5.99/mo** with 100K+ agents deployed. They win on:\n- Price ($5.99 vs anything we could charge for single agent)\n- Brand trust (4M+ users)\n- Infrastructure (global data centers)\n\n**But Hostinger can't do:**\n- ❌ Manage 10 agents from one dashboard\n- ❌ White-label agents for your clients\n- ❌ Pre-built agent templates (sales bot, support bot)\n- ❌ Cross-agent analytics\n- ❌ Agency billing (charge your clients through your dashboard)\n- ❌ Agent marketplace / skill store\n- ❌ India-first pricing in ₹\n\n**FireClaw's wedge: Multi-agent management, agency tools, and templates.**\n\n---\n\n## The Pitch (30 seconds)\n\n> \"Managing AI agents for your clients? Deploy 10 OpenClaw agents in 5 minutes, manage them all from one dashboard, white-label them, and charge your clients — starting at ₹1,999/month.\"\n\n---\n\n## Revenue Model (Research-Backed)\n\n### Pricing (India-First)\n\n| Plan | Agents | Price/mo | Infra Cost | Margin |\n|------|--------|----------|------------|--------|\n| **Personal** | 1 | ₹499 (~$6) | ~$8 (shared) | ~25% |\n| **Business** | 3 | ₹1,999 (~$24) | ~$20 (shared) | ~17% |\n| **Agency** | 10 | ₹4,999 (~$60) | ~$50 (shared) | ~17% |\n| **Enterprise** | Unlimited | Custom | Custom | 30%+ |\n\n**Why these prices work:**\n- Personal at ₹499 matches Hostinger's $5.99 in value\n- Business/Agency have NO equivalent competitor (Hostinger = 1 agent/plan)\n- Margins improve with density (more agents per server)\n\n### Infrastructure Strategy (NOT 1 EC2 per agent)\n\nInstead of dedicated instances, use **shared containerized infrastructure:**\n\n| Server | Type | Cost/mo | Agents Hosted | Cost/Agent |\n|--------|------|---------|---------------|------------|\n| Shared | t3.xlarge (4 vCPU, 16 GB) | ~$122 | 10-15 | $8-12 |\n| Shared | t3.2xlarge (8 vCPU, 32 GB) | ~$244 | 25-30 | $8-10 |\n| Dedicated | t3.micro (for personal) | ~$11 | 1 | $11 |\n\n*With 1yr Reserved Instances: costs drop 35-40%.*\n\n### Add-On Revenue\n| Add-On | Price |\n|--------|-------|\n| Extra agent | ₹299/mo |\n| WhatsApp messages (over 1000/mo) | ₹1/message |\n| Premium templates | ₹499-999 one-time |\n| White-label branding | Included in Agency+ |\n| Bundled AI credits (no BYOK needed) | ₹499/mo for 10K messages |\n\n### Revenue Projections\n\n| Milestone | Users | MRR | ARR |\n|-----------|-------|-----|-----|\n| Month 3 | 60 | ₹75K (~$900) | ~$10.8K |\n| Month 6 | 250 | ₹300K (~$3.6K) | ~$43K |\n| Month 12 | 700 | ₹1M (~$12K) | ~$144K |\n| Month 24 | 3,000 | ₹5M (~$60K) | ~$720K |\n\n---\n\n## Target Customers (Revised)\n\n### Phase 1 — Indian Agencies & Freelancers (Month 1-3)\n- Digital marketing agencies deploying chatbots for clients\n- Freelance developers offering \"AI assistant\" as a service\n- **Why they pay:** Managing 5-10 client agents from one place saves hours/week\n\n### Phase 2 — SMBs & Startups (Month 3-6)\n- Small businesses wanting AI for support/sales\n- Startups that need AI agents across departments\n- **Why they pay:** Templates get them started in minutes, not days\n\n### Phase 3 — Global Agencies (Month 6+)\n- International agencies managing AI fleets\n- White-label resellers\n- **Why they pay:** White-label + client billing = new revenue stream for them\n\n---\n\n## Competitive Landscape (Research-Backed)\n\n| Competitor | What | Price | Gap |\n|-----------|------|-------|-----|\n| **Hostinger OpenClaw** | 1-click single agent | $5.99/mo | No multi-agent, no agency tools |\n| **Dante AI** | AI chatbots | $40-400/mo | Chatbot only, not full agent, expensive |\n| **Botpress** | Chatbot builder | $0-495/mo | Visual builder, not OpenClaw, no WhatsApp focus |\n| **Chatfuel** | WhatsApp/IG bots | $69/mo | Simple bots, not AI agents |\n| **Tidio** | Customer service AI | $24-749/mo | Support-focused, not general agents |\n| **ManyChat** | Marketing automation | $15-100/mo | Flow-based, not AI agents |\n| **Railway/Render** | Generic PaaS | $5-85/mo | Not AI-specific, complex |\n\n**FireClaw's unique position:** Only platform for managing multiple OpenClaw agents with agency tools.\n\n---\n\n## Phases (Revised)\n\n### Phase 1: MVP — Agency Dashboard (3 weeks)\n- Landing page targeting Indian agencies\n- Multi-agent deploy from one dashboard\n- 3 pre-built templates (sales bot, support bot, personal assistant)\n- Stripe + Razorpay billing\n- 10-20 beta agencies\n- **Goal:** Prove agencies will pay for multi-agent management\n\n### Phase 2: Templates & Marketplace (3 weeks)\n- 10+ agent templates\n- Skill marketplace (install skills per agent)\n- White-label setup\n- Client billing (agencies charge their clients through us)\n- **Goal:** 100 paying users\n\n### Phase 3: Scale (Ongoing)\n- Multi-region (Mumbai, Singapore, US-East)\n- Agent analytics dashboard\n- Mobile app for monitoring\n- API for programmatic agent management\n- Community template submissions\n- **Goal:** ₹1M MRR\n\n---\n\n## Tech Stack (Revised)\n\n### Frontend (fireclaw.ai)\n- Next.js 16 + TypeScript\n- Tailwind CSS + shadcn/ui\n- Stripe + Razorpay for payments\n- authjs for authentication \n\n### Backend (Provisioning Engine)\n- Node.js API\n- **Docker + Docker Compose** (not bare EC2 per agent)\n- AWS SDK v3 (EC2 for host servers, not per agent)\n- MongoDB Atlas (user data, agent registry)\n- Bull/BullMQ (provisioning queue)\n\n### Agent Infrastructure\n- **Shared host servers** running Docker\n- Each agent = isolated Docker container with OpenClaw\n- Caddy reverse proxy (routing subdomains to containers)\n- Automated health checks + restart\n- Central logging (agent metrics → dashboard)\n\n### Why Docker Over Bare EC2\n| Approach | Cost for 10 agents | Management |\n|----------|-------------------|------------|\n| 10x t3.micro | ~$106/mo | 10 servers to manage |\n| 1x t3.xlarge + Docker | ~$125/mo | 1 server, 10 containers |\n| 2x t3.large + Docker | ~$132/mo | 2 servers, 5 containers each (HA) |\n\nDocker is similar cost but dramatically simpler to manage and scale.\n\n---\n\n## Key Risks & Mitigations\n\n| Risk | Mitigation |\n|------|-----------|\n| Hostinger drops price or adds multi-agent | Move fast, build community moat |\n| OpenClaw changes licensing | We contribute upstream, stay aligned |\n| Low margins on Personal plan | Personal is acquisition funnel → upsell to Business/Agency |\n| Container density limits | Auto-scale to new host servers |\n| India market slow to adopt | Parallel global launch at USD pricing |\n\n---\n\n## Key Metrics\n\n- **Agents deployed** — total across all users\n- **Agents per user** — avg (target: 2.5+)\n- **MRR** — monthly recurring revenue\n- **Provisioning time** — target: <60 seconds\n- **Uptime per agent** — target: 99.5%+\n- **Churn** — target: <5%/month\n- **CAC** — cost to acquire (target: <₹500)\n- **LTV** — lifetime value (target: >₹15,000)\n\n---\n\n## Multi-Provider API Key + Live Model Selection\n\n### Overview\nTwo places users configure AI providers with the same flow:\n1. **FireClaw Deployment Wizard** — during onboarding (Next.js app)\n2. **OpenClaw Dashboard** — topbar key icon popover (Lit UI)\n\n### Flow\n1. User opens provider panel → sees list of supported providers worldwide\n2. User selects a provider → enters their API key\n3. System **fetches live models** from the provider API using that key\n4. Models displayed with auto-select on the latest/best model\n5. User confirms → key + selected model saved to OpenClaw config\n6. Gateway restarts → agent uses the new model\n\n### Supported Providers (Worldwide)\n| Provider | Config Key | Models API | API Key Prefix |\n|----------|-----------|------------|----------------|\n| OpenAI | openai | GET /v1/models | sk- |\n| Anthropic | anthropic | GET /v1/models | sk-ant- |\n| Google Gemini | google | GET /v1beta/models | AI... |\n| Groq | groq | GET /openai/v1/models | gsk_ |\n| Mistral | mistral | GET /v1/models | — |\n| OpenRouter | openrouter | GET /api/v1/models | sk-or- |\n| DeepSeek | deepseek | GET /models | sk- |\n| Together AI | together | GET /v1/models | — |\n| Fireworks AI | fireworks | GET /inference/v1/models | — |\n| Perplexity | perplexity | GET /models | pplx- |\n| Cohere | cohere | GET /v2/models | — |\n| xAI (Grok) | xai | GET /v1/models | xai- |\n| MiniMax | minimax | GET /v1/models | — |\n| Cerebras | cerebras | GET /v1/models | csk- |\n| SambaNova | sambanova | GET /v1/models | — |\n| Qwen (DashScope) | qwen | GET /compatible-mode/v1/models | sk- |\n\n### Suggestion Banner\nShow a recommendation: \"Try MiniMax — powerful models at affordable pricing\" (shown when no key is configured).\n\n### Config Patch Format\n```json\n{\n  \"models\": {\n    \"providers\": {\n      \"<provider>\": {\n        \"apiKey\": \"<key>\",\n        \"api\": \"openai-completions\"\n      }\n    }\n  },\n  \"agents\": {\n    \"defaults\": {\n      \"model\": \"<provider>/<model-id>\"\n    }\n  }\n}\n```\n\n### Live Model Fetching\n- Client-side fetch to provider APIs (most support CORS or use OpenAI-compatible endpoints)\n- Fallback: proxy through OpenClaw gateway if CORS blocked\n- Parse response → display model list → auto-select latest\n- Show model metadata: name, context window, pricing tier\n"}, "27ed3a183a5563b0ff93e1650f63ae4392339b17": {"t": 1785771365.4783561, "n": 7106, "h": "2fd2fb089bc96659", "c": "# 🔥 FireClaw — The Plan (Research-Backed v2)\n\n## What Is FireClaw?\n\n~~One-click deploy your personal AI assistant.~~ ← Hostinger already does this at $5.99/mo.\n\n**FireClaw is the multi-agent management platform for OpenClaw.**\n\nDeploy, manage, and monitor multiple AI agents from one dashboard. Built for agencies, teams, and power users who need more than a single agent.\n\nThink: **Vercel for AI agents** — not another Hostinger.\n\n---\n\n## Why Not Just Use Hostinger?\n\nHostinger offers 1-click OpenClaw at **$5.99/mo** with 100K+ agents deployed. They win on:\n- Price ($5.99 vs anything we could charge for single agent)\n- Brand trust (4M+ users)\n- Infrastructure (global data centers)\n\n**But Hostinger can't do:**\n- ❌ Manage 10 agents from one dashboard\n- ❌ White-label agents for your clients\n- ❌ Pre-built agent templates (sales bot, support bot)\n- ❌ Cross-agent analytics\n- ❌ Agency billing (charge your clients through your dashboard)\n- ❌ Agent marketplace / skill store\n- ❌ India-first pricing in ₹\n\n**FireClaw's wedge: Multi-agent management, agency tools, and templates.**\n\n---\n\n## The Pitch (30 seconds)\n\n> \"Managing AI agents for your clients? Deploy 10 OpenClaw agents in 5 minutes, manage them all from one dashboard, white-label them, and charge your clients — starting at ₹1,999/month.\"\n\n---\n\n## Revenue Model (Research-Backed)\n\n### Pricing (India-First)\n\n| Plan | Agents | Price/mo | Infra Cost | Margin |\n|------|--------|----------|------------|--------|\n| **Personal** | 1 | ₹499 (~$6) | ~$8 (shared) | ~25% |\n| **Business** | 3 | ₹1,999 (~$24) | ~$20 (shared) | ~17% |\n| **Agency** | 10 | ₹4,999 (~$60) | ~$50 (shared) | ~17% |\n| **Enterprise** | Unlimited | Custom | Custom | 30%+ |\n\n**Why these prices work:**\n- Personal at ₹499 matches Hostinger's $5.99 in value\n- Business/Agency have NO equivalent competitor (Hostinger = 1 agent/plan)\n- Margins improve with density (more agents per server)\n\n### Infrastructure Strategy (NOT 1 EC2 per agent)\n\nInstead of dedicated instances, use **shared containerized infrastructure:**\n\n| Server | Type | Cost/mo | Agents Hosted | Cost/Agent |\n|--------|------|---------|---------------|------------|\n| Shared | t3.xlarge (4 vCPU, 16 GB) | ~$122 | 10-15 | $8-12 |\n| Shared | t3.2xlarge (8 vCPU, 32 GB) | ~$244 | 25-30 | $8-10 |\n| Dedicated | t3.micro (for personal) | ~$11 | 1 | $11 |\n\n*With 1yr Reserved Instances: costs drop 35-40%.*\n\n### Add-On Revenue\n| Add-On | Price |\n|--------|-------|\n| Extra agent | ₹299/mo |\n| WhatsApp messages (over 1000/mo) | ₹1/message |\n| Premium templates | ₹499-999 one-time |\n| White-label branding | Included in Agency+ |\n| Bundled AI credits (no BYOK needed) | ₹499/mo for 10K messages |\n\n### Revenue Projections\n\n| Milestone | Users | MRR | ARR |\n|-----------|-------|-----|-----|\n| Month 3 | 60 | ₹75K (~$900) | ~$10.8K |\n| Month 6 | 250 | ₹300K (~$3.6K) | ~$43K |\n| Month 12 | 700 | ₹1M (~$12K) | ~$144K |\n| Month 24 | 3,000 | ₹5M (~$60K) | ~$720K |\n\n---\n\n## Target Customers (Revised)\n\n### Phase 1 — Indian Agencies & Freelancers (Month 1-3)\n- Digital marketing agencies deploying chatbots for clients\n- Freelance developers offering \"AI assistant\" as a service\n- **Why they pay:** Managing 5-10 client agents from one place saves hours/week\n\n### Phase 2 — SMBs & Startups (Month 3-6)\n- Small businesses wanting AI for support/sales\n- Startups that need AI agents across departments\n- **Why they pay:** Templates get them started in minutes, not days\n\n### Phase 3 — Global Agencies (Month 6+)\n- International agencies managing AI fleets\n- White-label resellers\n- **Why they pay:** White-label + client billing = new revenue stream for them\n\n---\n\n## Competitive Landscape (Research-Backed)\n\n| Competitor | What | Price | Gap |\n|-----------|------|-------|-----|\n| **Hostinger OpenClaw** | 1-click single agent | $5.99/mo | No multi-agent, no agency tools |\n| **Dante AI** | AI chatbots | $40-400/mo | Chatbot only, not full agent, expensive |\n| **Botpress** | Chatbot builder | $0-495/mo | Visual builder, not OpenClaw, no WhatsApp focus |\n| **Chatfuel** | WhatsApp/IG bots | $69/mo | Simple bots, not AI agents |\n| **Tidio** | Customer service AI | $24-749/mo | Support-focused, not general agents |\n| **ManyChat** | Marketing automation | $15-100/mo | Flow-based, not AI agents |\n| **Railway/Render** | Generic PaaS | $5-85/mo | Not AI-specific, complex |\n\n**FireClaw's unique position:** Only platform for managing multiple OpenClaw agents with agency tools.\n\n---\n\n## Phases (Revised)\n\n### Phase 1: MVP — Agency Dashboard (3 weeks)\n- Landing page targeting Indian agencies\n- Multi-agent deploy from one dashboard\n- 3 pre-built templates (sales bot, support bot, personal assistant)\n- Stripe + Razorpay billing\n- 10-20 beta agencies\n- **Goal:** Prove agencies will pay for multi-agent management\n\n### Phase 2: Templates & Marketplace (3 weeks)\n- 10+ agent templates\n- Skill marketplace (install skills per agent)\n- White-label setup\n- Client billing (agencies charge their clients through us)\n- **Goal:** 100 paying users\n\n### Phase 3: Scale (Ongoing)\n- Multi-region (Mumbai, Singapore, US-East)\n- Agent analytics dashboard\n- Mobile app for monitoring\n- API for programmatic agent management\n- Community template submissions\n- **Goal:** ₹1M MRR\n\n---\n\n## Tech Stack (Revised)\n\n### Frontend (fireclaw.ai)\n- Next.js 16 + TypeScript\n- Tailwind CSS + shadcn/ui\n- Stripe + Razorpay for payments\n- authjs for authentication \n\n### Backend (Provisioning Engine)\n- Node.js API\n- **Docker + Docker Compose** (not bare EC2 per agent)\n- AWS SDK v3 (EC2 for host servers, not per agent)\n- MongoDB Atlas (user data, agent registry)\n- Bull/BullMQ (provisioning queue)\n\n### Agent Infrastructure\n- **Shared host servers** running Docker\n- Each agent = isolated Docker container with OpenClaw\n- Caddy reverse proxy (routing subdomains to containers)\n- Automated health checks + restart\n- Central logging (agent metrics → dashboard)\n\n### Why Docker Over Bare EC2\n| Approach | Cost for 10 agents | Management |\n|----------|-------------------|------------|\n| 10x t3.micro | ~$106/mo | 10 servers to manage |\n| 1x t3.xlarge + Docker | ~$125/mo | 1 server, 10 containers |\n| 2x t3.large + Docker | ~$132/mo | 2 servers, 5 containers each (HA) |\n\nDocker is similar cost but dramatically simpler to manage and scale.\n\n---\n\n## Key Risks & Mitigations\n\n| Risk | Mitigation |\n|------|-----------|\n| Hostinger drops price or adds multi-agent | Move fast, build community moat |\n| OpenClaw changes licensing | We contribute upstream, stay aligned |\n| Low margins on Personal plan | Personal is acquisition funnel → upsell to Business/Agency |\n| Container density limits | Auto-scale to new host servers |\n| India market slow to adopt | Parallel global launch at USD pricing |\n\n---\n\n## Key Metrics\n\n- **Agents deployed** — total across all users\n- **Agents per user** — avg (target: 2.5+)\n- **MRR** — monthly recurring revenue\n- **Provisioning time** — target: <60 seconds\n- **Uptime per agent** — target: 99.5%+\n- **Churn** — target: <5%/month\n- **CAC** — cost to acquire (target: <₹500)\n- **LTV** — lifetime value (target: >₹15,000)\n"}, "4a6857917955d6a123d2e3f2e05d350c1d7a8dfb": {"t": 1785771366.8019862, "n": 3882, "h": "51b97254df2947a4", "c": "# 🔥 FireClaw — The Plan\n\n## What Is FireClaw?\n\n**One-click deploy your personal AI assistant.**\n\nFireClaw is a managed platform that lets anyone deploy their own OpenClaw AI agent in under 2 minutes. WhatsApp-first, always-on, fully isolated.\n\nThink: **Hostinger for AI agents.**\n\n---\n\n## The Pitch (30 seconds)\n\n> \"Want an AI assistant on WhatsApp that remembers everything, runs 24/7, and is 100% yours? Click deploy, scan a QR code, and it's live. $29/month.\"\n\n---\n\n## How It Works (User Flow)\n\n```\n1. User visits fireclaw.ai\n2. Signs up (email or Google)\n3. Clicks \"Deploy New Agent\"\n4. Picks a plan (Starter $29 / Pro $79 / Agency $149)\n5. Enters: Agent name, WhatsApp number\n6. Pays (Stripe/Razorpay)\n7. We auto-provision an EC2 instance in ~90 seconds\n8. User gets a dashboard URL + QR code\n9. Scans QR → WhatsApp connected\n10. Done. AI agent is live.\n```\n\n---\n\n## Revenue Model\n\n| Plan | Server | Our Cost | Price | Margin |\n|------|--------|----------|-------|--------|\n| Starter | t3.micro (1 vCPU, 1GB) | ~$8/mo | $29/mo | 72% |\n| Pro | t3.medium (2 vCPU, 4GB) | ~$33/mo | $79/mo | 58% |\n| Agency | t3.large (2 vCPU, 8GB) | ~$65/mo | $149/mo | 56% |\n\n**AI costs:** Users bring their own API key (OpenAI/Anthropic/GitHub Copilot) — or we offer a bundled plan with usage-based AI ($0.01/message).\n\n**100 clients on Starter = $2,900/mo revenue, $2,100 gross profit.**\n\n---\n\n## Target Customers\n\n### Phase 1 — Founders & Solopreneurs\n- Want a personal AI assistant on WhatsApp\n- Don't know how to self-host\n- Will pay $29-79/mo for convenience\n\n### Phase 2 — Agencies & SMBs\n- Want AI agents for each client\n- Customer support bots, lead gen bots\n- Deploy + manage multiple agents from one dashboard\n\n### Phase 3 — Enterprises\n- Dedicated instances, SLAs\n- Custom domains, branding\n- White-label option\n\n---\n\n## Competitive Landscape\n\n| Competitor | What They Do | Gap |\n|-----------|--------------|-----|\n| Hostinger | 1-click WordPress hosting | No AI agents |\n| Railway/Render | Deploy any app | Not AI-specific, complex |\n| DigitalOcean App Platform | 1-click apps | No OpenClaw, no WhatsApp |\n| Twilio | Communication APIs | Build-it-yourself |\n| OpenClaw (self-hosted) | The AI agent itself | Requires technical setup |\n\n**FireClaw's wedge:** We make OpenClaw deployment as easy as creating a Gmail account.\n\n---\n\n## Phases\n\n### Phase 1: MVP (2-3 weeks)\n- Landing page with waitlist + early access\n- Manual-ish provisioning (automated script, human review)\n- 10-20 beta users\n- Prove willingness to pay\n\n### Phase 2: Full Automation (2-3 weeks)\n- Auto-provisioning API (EC2 + setup script + DNS)\n- Self-service dashboard (deploy, manage, monitor)\n- Stripe billing integration\n- WhatsApp QR linking in dashboard\n\n### Phase 3: Scale (ongoing)\n- Multi-region (Mumbai, Singapore, US-East, EU)\n- Agent templates (job apply bot, support bot, research bot)\n- White-label for agencies\n- Mobile app for managing agents\n- Marketplace for skills/plugins\n\n---\n\n## Tech Stack\n\n### Frontend (fireclaw.ai)\n- Next.js 16 + TypeScript\n- Tailwind CSS + shadcn/ui\n- Stripe for payments\n- Better-Auth for authentication\n\n### Backend (Provisioning Engine)\n- Node.js API routes\n- AWS SDK v3 (EC2, Route53, IAM)\n- MongoDB (user data, instance tracking)\n- Bull/BullMQ (job queue for provisioning)\n\n### Per-Client Instance\n- Ubuntu 22.04 on EC2\n- Node.js 22 + OpenClaw (npm)\n- Caddy (reverse proxy + auto HTTPS)\n- Systemd (auto-restart)\n\n### Infrastructure\n- AWS EC2 (client instances)\n- Vercel (fireclaw.ai dashboard)\n- MongoDB Atlas (SaaS database)\n- Cloudflare or Route53 (DNS + subdomains)\n- Stripe (billing)\n\n---\n\n## Key Metrics to Track\n\n- **Provisioning time** — target: <90 seconds\n- **Uptime per instance** — target: 99.5%+\n- **MRR** — monthly recurring revenue\n- **Churn** — % users who cancel per month\n- **CAC** — cost to acquire one customer\n- **LTV** — lifetime value per customer\n"}}}}