1 line
2.9 KiB
JSON
1 line
2.9 KiB
JSON
{"repo": "axclogo/axcagentengine", "files": {"llms.txt": {"c342ff9f9a3d0c8e1a51d7d39dd830009cc2c9fc": {"t": 1785674435.2527225, "n": 2712, "h": "5262faf6f08d8f7c", "c": "# AxcAgentEngine\n\n## What is this?\n\nAxcAgentEngine is a Python agent execution engine with Plan-Observe-Replan\n(POR), ReAct execution, tool calling, and a plugin architecture. It runs on top\nof OpenAI-compatible Chat Completions providers through a provider protocol.\n\n## Key Features\n\n- POR execution engine: plan, schedule by dependency, observe, and replan.\n- ReAct executor for direct iterative tool use.\n- Plugin architecture with YAML-driven loading and phase/priority ordering.\n- OpenAI-compatible HTTP client plus provider protocol.\n- OpenAI Chat Completions compatible HTTP API subset.\n- Parallel read-only tools and serial write/risky tools.\n- Tool outputs must return `ToolOutput`.\n- Provider-owned tool-name mapping for model-safe function names.\n- Context compression, memory, knowledge, MCP, hooks, tracing, safety, and HITL plugins.\n- Sidecar multi-agent orchestration, simulation, evaluation, and AgentOps helpers.\n\n## Open Source Files\n\n- License: `LICENSE` (Apache-2.0)\n- Security policy: `SECURITY.md`\n- Contributing guide: `CONTRIBUTING.md`\n- API docs: `docs/API.md`\n- Plugin docs: `docs/PLUGIN_DEVELOPMENT.md`\n- Security model: `docs/SECURITY_MODEL.md`\n- Release checklist: `docs/RELEASE_CHECKLIST.md`\n- Examples: `examples/README.md`\n\n## Install\n\n```bash\npip install axc-agent-engine\n```\n\n## Quick Start\n\n```python\nfrom axc_agent_engine import Engine, LLMConfig\n\nengine = Engine(\n default_llm=LLMConfig(\n base_url=\"https://api.openai.com/v1\",\n api_key=\"sk-xxx\",\n model=\"gpt-4o\",\n ),\n)\n\nagent = engine.load_agent(\"./agents/my_agent.yaml\")\nresult = await agent.chat(\"hello\")\n```\n\n## API Compatibility\n\nThe HTTP API is an OpenAI Chat Completions compatible subset:\n\n- `POST /v1/chat/completions`\n- `GET /v1/agents`\n- `GET /v1/capabilities`\n\nRequest-level OpenAI `tools`, `tool_choice`, and `n > 1` are intentionally\nunsupported. Tools are loaded from Agent YAML and plugins so the engine can\nenforce capability policy, risk metadata, plugin hooks, workspace rules, and\naudit events. Clients should call `/v1/capabilities` before assuming API parity.\n\n## Architecture Notes\n\n- Engine core: `Executor`, `LLMCaller`, planning/POR, context, events, and recovery.\n- Extension boundary: plugins implement hooks and return `ToolDefinition` objects.\n- Tool boundary: every tool returns `ToolOutput`; large results can be externalized.\n- Provider boundary: model I/O and model-safe function-name mapping live in providers.\n- Runtime policy: workspace, command execution, capability checks, and audit sinks are injectable.\n\n## Validation Before Release\n\n```bash\npython -m pytest\npython -m compileall -q axc_agent_engine\ngit diff --check\npython -m build\ntwine check dist/*\n```\n"}}}} |