steerable-agent-runtime 0.1.0

Steerable agent runtime: LLM, tool, storage, and transport adapters.

steerable-agent-runtime

Tier 3 runtime for the Steerable framework.

Provides four orthogonal pluggable adapters:

• LLMProvider — chat-completion / streaming / tool-call abstraction, with reference implementations for OpenAI-compatible servers (covers OpenAI, Ollama, vLLM, SiliconFlow, etc.) and Anthropic native.
• ToolRouter — in-process tool registry. Auto-classifies tools into ToolModes using steerable_agent_harness.policy, supports per-tool permission overrides, and dispatches ToolCall → ToolResult.
• StorageAdapter — persistence interface for AgentSession, ChatMessage, ChatAgent, and HarnessTrace + spans + events. Reference impls: in-memory (default for sidecar/dev) and SQLAlchemy (for hosted backends).
• TransportAdapter — wire format. FastAPISseTransport exports SSE for hosted setups; StdioJsonRpcTransport powers the steerable-sidecar.

The runtime is Python only by design — frontends never depend on it directly. Browsers/Electron consume runtime output via either the SSE transport (over HTTP) or the stdio JSON-RPC transport (sidecar pattern).

Install

pip install steerable-agent-runtime[all]

Selectively install just the bits you need:

pip install "steerable-agent-runtime[openai]"
pip install "steerable-agent-runtime[anthropic]"
pip install "steerable-agent-runtime[sqlalchemy]"
pip install "steerable-agent-runtime[fastapi]"