Skip to content

Steerable Framework

An open-source, spec-first agent framework extracted from the DeepPath ecosystem. Build AI agents in Python (server) and TypeScript (browser/Electron) with a single source-of-truth wire protocol, a deterministic harness, a pluggable runtime, and a portable sidecar binary that lets desktop apps share 100% of their agent business logic with their web backend.

Why Steerable

Most agent frameworks are either single-language (lock you out of the desktop / browser) or stitched together at the API level (your web app and your Python server drift apart within weeks). Steerable's spec-first design means:

  • One JSON Schema → Python Pydantic + TypeScript types + runtime validators
  • Drift between languages is detected by CI, not by users
  • Web (TS), API (Py), and desktop (Electron + Py sidecar) all consume the same primitives

Architecture (Tier 1 → Tier 4)

Arrows point from a layer to what it depends on. Dashed arrows mark shared-type relationships and out-of-band channels (Electron spawning the sidecar over stdio).

graph BT
  T4["<b>Tier 4 · UI</b> (TypeScript / React)<br/>@steerable/agent-ui<br/>Hooks: useChatStream · useToolCallStatus · useAgentSession<br/>Components: ChatPanel · MessageList · OrchestrationPlanCard ·<br/>ToolCallRenderer · SSEStreamView<br/>Tailwind preset (dark-mode aware)"]

  T3S["<b>Tier 3 · Sidecar</b> (portable CPython binary)<br/>steerable-sidecar<br/>JSON-RPC over stdio · ready marker · graceful shutdown<br/>agent.chat.stream · tool.invoke · agent.session.* · trace.fetch"]

  T3R["<b>Tier 3 · Runtime</b> (Python only)<br/>steerable-agent-runtime<br/>LLMProvider (OpenAI-compat / Anthropic / Ollama)<br/>ToolRouter · StorageAdapter · TransportAdapter (FastAPI SSE)"]

  T2["<b>Tier 2 · Harness</b> (Python — single source of truth)<br/>steerable-agent-harness<br/>Policy · Budget · Retry · Completion · Tracing · Safety<br/><i>thin TS facade @steerable/agent-harness exists for parity tests</i>"]

  T1["<b>Tier 1 · Protocol</b> (TypeScript + Python, lock-step versions)<br/>@steerable/agent-protocol · steerable-agent-protocol<br/>SSEEvent · ChatMessage · ToolCall · ToolResult · AgentSession<br/>HarnessTrace · TraceSpan · SidecarRequest/Response/Notification"]

  T4 -.->|"shared wire types"| T1
  T4 -->|"spawns + JSON-RPC stdio (Electron)"| T3S
  T3S -->|"embeds"| T2
  T3S -->|"validates against"| T1
  T3R -->|"embeds"| T2
  T3R -->|"validates against"| T1
  T2 -->|"imports types"| T1

  classDef tier fill:#f4f4f5,stroke:#3f3f46,color:#18181b,rx:6,ry:6,padding:12;
  class T1,T2,T3R,T3S,T4 tier;

The two top-tier consumers (a web app + a desktop Electron app) load the same agent-protocol types, so any change to a wire type either flows through CI codegen or is rejected by drift checks before it ships.

What's in v0.1.0

Package Tier Status
@steerable/agent-protocol 1 Released
steerable-agent-protocol 1 Released
@steerable/agent-harness 2 Facade
steerable-agent-harness 2 Released
steerable-agent-runtime 3 Released
steerable-sidecar 3 Released
@steerable/agent-ui 4 Released

Get started in 5 minutes

→ Getting Started

UI Storybook

The full @steerable/agent-ui component library — every variant, every state, with copy-paste hook examples — is published as a live Storybook alongside this site:

→ Browse the Storybook

The Storybook ships an a11y panel (axe-core) and includes per-story snapshots; pull-requests that regress visual or accessibility quality are blocked by the CI gates in .github/workflows/storybook-quality.yml.

Specs

Examples

Migrating from DeepPath internals

→ DeepPath → Steerable migration guide