Start building
Everything you need to integrate Statewave into your AI application. Running locally in under 2 minutes.
Zero to memory in one command
One line boots a local Statewave, wires it into your MCP clients, and seeds your repo — self-hosted, offline, no account.
Tear it down with npx @statewavedev/statewave --down.
Configures automatically
Which MCP clients should I set up? 1. Claude Code ✓ detected 2. Claude Desktop ✓ detected 3. Cursor ✓ detected 4. VS Code (GitHub Copilot) ✓ detected 5. Codex CLI ✓ detected Enter = detected, 'a' = all, 'n' = none: a Optional — an LLM API key sharpens the memory: Paste an LLM API key, or press Enter to skip: Starting Statewave (api + admin + db) via docker compose… ✔ Container statewave-quickstart-db-1 Healthy ✔ Container statewave-quickstart-api-1 Started ✔ Container statewave-quickstart-admin-1 Started ✓ Server healthy at http://localhost:8100. ✓ Configured Claude Code (server id: statewave) ✓ Configured Cursor · VS Code (Copilot) · Codex CLI Seeding repo:your-project from this repo… ✓ ingested 139/139 episodes — compiled: yes Admin console: http://localhost:8080
Everything you need to build with Statewave
Documentation, SDKs, examples, architecture and deployment guides.
Python SDK
Sync + async clients, Pydantic models, retry with backoff. pip install statewave.
TypeScript SDK
Fetch-based client with full type definitions. npm install @statewavedev/sdk.
API v1 Contract
Full endpoint reference with examples, scoring model and webhook specs.
Architecture Overview
System design, component architecture, middleware stack and compilation pipeline.
Deployment Guide
Docker Compose, Fly.io and Railway. Production configuration and health checks.
Examples
Quickstart, support agent, coding agent, LLM loop, eval suite and benchmarks.
GitHub Repository
Source code, issues and discussions. Apache-2.0 licensed.
OpenRouter Proxy
Give OpenRouter calls persistent memory without touching your client: change the base URL, add one header.
FAQ
Frequently asked questions
How long does it take to get Statewave running locally?
Under two minutes with one command: npx @statewavedev/statewave boots a local runtime, wires it into Claude Code, Claude Desktop, Cursor, VS Code Copilot, and Codex CLI, and seeds your repo — self-hosted, offline, no account. Tear it down with --down. The Docker Compose route takes about five minutes: docker compose up -d brings the API up on port 8100 alongside Postgres with pgvector, and curl localhost:8100/healthz reports ok once it is ready.
Which SDKs and clients are supported?
The Python SDK (pip install statewave, 1.5.0) ships sync and async clients, Pydantic models, and retry with backoff. The TypeScript SDK (npm install @statewavedev/sdk, 1.5.0) is fetch-based with full type definitions. Anything else talks to the documented v1 REST contract directly, and the shipped MCP server exposes the same runtime to any MCP-compatible client.
Do I need a GPU or a separate vector database?
Neither. Storage is Postgres plus the pgvector extension — usually already in your stack — and retrieval is a single ranked query, not a call out to a second system. The heuristic compiler needs no model at all, so a fully local deployment is CPU-only. Embeddings are optional: without them task relevance falls back to word overlap (0–5 instead of 0–8), and the rest of the scoring model is unchanged.
Where are the runnable examples?
statewave-examples carries the quickstart, a support agent, a coding agent, a bare LLM loop, the eval suite, and the benchmarks. The smallest useful integration is two calls — get_context before the model call, create_episode after it — and that is the whole surface for a chat app. The repo also has a React chat panel that renders every compiled fact grouped by kind, scored, and sourced back to its episode.
Answers last checked against the Statewave docs and repositories on .