{"repo": "bensonlee5/openral", "files": {".env.example": {"de448f8f4eef590ab86625416cb4df969156dc92": {"t": 1786068545.4065588, "n": 2473, "h": "fb969184ac300c1e", "c": "# OpenRAL — example reasoner LLM configurations\n#\n# Pick ONE of the three blocks below, copy to `.env` (gitignored), and\n# `source .env` before `ros2 run openral_reasoner_ros reasoner_node` or\n# any sim run that activates the S2 reasoner. `openral doctor` will report\n# what's missing if any required variable is unset.\n#\n# CLAUDE.md §1.4 — the open core has no default LLM. Explicit beats\n# implicit; you must pick one here.\n\n# ── Option 1: paid baseline (Anthropic Haiku 4.5) ────────────────────\n# Recommended default. ~$1/$5 per Mtok, ~0.74 s TTFT, native tool use.\n# One-time: `uv add anthropic --package openral-reasoner`\n#\n# export OPENRAL_REASONER_LLM_PROVIDER=anthropic\n# export OPENRAL_REASONER_LLM_MODEL=claude-haiku-4-5\n# export OPENRAL_REASONER_LLM_API_KEY=sk-ant-...\n\n# ── Option 2: free baseline (OpenRouter free models) ─────────────────\n# DeepSeek v3 / Llama-4-Maverick / Qwen3-235B all pass tool-calling\n# tests. Use `openrouter/free` as the model id for the auto-router.\n# One-time: `uv add openai --package openral-reasoner`\n#\n# export OPENRAL_REASONER_LLM_PROVIDER=openrouter\n# export OPENRAL_REASONER_LLM_MODEL=deepseek/deepseek-chat-v3:free\n# export OPENRAL_REASONER_LLM_API_KEY=sk-or-...\n\n# ── Option 3: local baseline (Ollama + Qwen3 8B) ─────────────────────\n# Setup: `just bootstrap-ollama`\n# One-time: `uv add openai --package openral-reasoner`\n#\n# export OPENRAL_REASONER_LLM_PROVIDER=openai-compatible\n# export OPENRAL_REASONER_LLM_MODEL=qwen3:8b\n# export OPENRAL_REASONER_LLM_BASE_URL=http://localhost:11434/v1\n\n# ── Named vendor presets (Gemini / xAI / DeepSeek) ───────────────────\n# Thin auth-required shortcuts over the OpenAI-compatible client that\n# pre-fill the vendor base URL (issue #74). See the table in\n# packages/openral_reasoner_ros/README.md.\n# One-time: `uv add openai --package openral-reasoner`\n#\n# export OPENRAL_REASONER_LLM_PROVIDER=gemini    # or: xai / deepseek\n# export OPENRAL_REASONER_LLM_MODEL=gemini-2.5-flash\n# export OPENRAL_REASONER_LLM_API_KEY=...\n\n# ── Local vLLM preset (self-hosted) ──────────────────────────────────\n# Convenience preset for a local `vllm serve <model>` (loopback :8000,\n# no auth). Equivalent to openai-compatible + BASE_URL, minus the typing.\n# One-time: `uv add openai --package openral-reasoner`\n#\n# export OPENRAL_REASONER_LLM_PROVIDER=vllm\n# export OPENRAL_REASONER_LLM_MODEL=qwen2.5-7b-instruct\n# export OPENRAL_REASONER_LLM_BASE_URL=http://localhost:8000/v1  # only if non-default\n"}}}}