About
About Statewave
Statewave is an open-source memory runtime for AI agents and LLM applications. It records raw events as immutable episodes, compiles them into typed semantic and episodic memories with provenance, and returns ranked, token-bounded context bundles that drop straight into a prompt.
We built Statewave because the problem most LLM apps spend engineering time on isn’t which model to call — it’s remembering what was already said, decided, or learned. Prompt stuffing breaks at scale, naive RAG returns embedding-nearest instead of decision-relevant, and chat-history replay drowns agents in noise. A memory layer with structure, ranking, and provenance is infrastructure, not a feature, and it deserves to be open and self-hostable.
Our principles
What we’re committed to
Open source under Apache 2.0
Server, SDKs, connectors, and this website are public on GitHub. Apache-2.0 is permissive and includes an explicit patent grant, so teams can use, fork, modify, distribute, and ship commercial products on Statewave without a separate agreement.
Self-hosted, no managed cloud
Statewave runs on your infrastructure. Episodes, compiled memories, and the embeddings that index them stay in your Postgres. There is no Statewave-hosted backend the SDK secretly phones home to — the demo on the homepage talks to a Statewave instance we operate transparently.
Provenance first
Every compiled memory carries the IDs of the episodes it was derived from, with confidence scores and validity windows. Context bundles can be traced back to the raw events that produced them, so an agent’s answer is always explainable to a human.
Framework-neutral
Compilation goes through LiteLLM (100+ providers, including locally-hosted Ollama / vLLM). Retrieval and context assembly are provider-agnostic. Statewave is an HTTP service plus two thin SDKs — nothing on the agent side is locked to a specific LLM, framework, or vendor.
Proven, not promised
Every claim on this site is backed by automated evals and benchmarks that run in CI. The figures below come from the statewave-memory-benchmarks harness and the public reference repos linked below — anyone can clone them and reproduce every number.
708
Unit tests
56
Eval assertions
8
Support eval criteria
73%
Fewer tokens vs raw history (reference demo)
Live community counts
374
GitHub stars · 38 forks
34.9k
Docker pulls
v1.5.0
statewave on PyPI · 85/mo
v1.5.0
@statewavedev/sdk on npm · 321/mo
Counts last refreshed 2026-10-09.
Open-source ecosystem
What we ship
Statewave is split across focused repositories so teams can adopt only what they need. All are public on GitHub under Apache-2.0.
statewave
Core server — REST API, compiler, ranking, storage layer.
statewave-py
Python SDK.
statewave-ts
TypeScript SDK.
statewave-docs
Architecture, API contracts, ADRs, deployment guides.
statewave-examples
Runnable examples and end-to-end agent flows.
statewave-connectors
GitHub, Jira, Slack, Notion, Discord, Zendesk, Intercom, Freshdesk, Gmail, n8n, Zapier, databases, Markdown/ADRs, MCP, and more — a suite of modular connector packages.
statewave-admin
Read-only operator console for inspecting subjects, episodes, memories, and bundles in a running instance.
statewave-memory-benchmarks
Open evaluation harness — every published proof figure is reproducible by running these scripts.
statewave-web
This marketing site.
Contact
Get in touch
Bugs, feature requests, integration questions
Open an issue on GitHub. That’s where the project is built in the open.
Commercial use, enterprise support, procurement
licensing@statewave.ai. Apache-2.0 is permissive — you don’t need a contract to use Statewave commercially. We offer optional SLA, indemnity, architecture review, and managed hosting on request.
Press, partnerships, brand assets
press@statewave.ai. Logos and a one-page fact sheet are on the press page.
Security disclosures
security@statewave.ai. Please don’t open public issues for vulnerabilities.
— the Statewave team
FAQ
Frequently asked questions
How is Statewave licensed, and can I use it commercially?
Everything — the server, both SDKs, the connectors, the admin console, the benchmark harness, and this website — is public on GitHub under Apache-2.0. That license is permissive and carries an explicit patent grant, so you can use, fork, modify, distribute, and ship commercial products on Statewave without signing anything. Optional SLA, indemnity, architecture review, and managed hosting are available on request at licensing@statewave.ai.
Is there a hosted Statewave I can sign up for?
No. Statewave runs on your infrastructure, and there is no Statewave-hosted backend the SDKs quietly phone home to — episodes, compiled memories, and embeddings stay in your Postgres. The one exception is transparent: the live demo on this site talks to an instance we operate ourselves, so you can try the runtime without installing it. Managed hosting exists only as an optional commercial arrangement, never as the default path.
How much adoption does the project have?
Public counts, refreshed out-of-band by a script and committed to the repo so this page is never a live guess: 374 GitHub stars and 38 forks, 34.9k Docker pulls, statewave 1.5.0 on PyPI (85 downloads a month), @statewavedev/sdk 1.5.0 on npm (321 a month), counted on 2026-10-09. The engineering figures beside them come from CI — 708 unit tests, 56 eval assertions, and 8 support eval criteria — and every published proof figure is reproducible by cloning the statewave-memory-benchmarks harness and running it yourself.
Which repository should I start with?
statewave is the core server — REST API, compiler, ranking, and storage. statewave-py and statewave-ts are the SDKs; statewave-docs holds the architecture notes, API contracts, and ADRs; statewave-examples has runnable end-to-end agent flows. statewave-connectors ships GitHub, Jira, Slack, Notion, Discord, Zendesk, Intercom, Freshdesk, Gmail, n8n, Zapier, database, Markdown/ADR, and MCP packages. Adopt only the pieces you need — they are separate packages on purpose.
Answers last checked against the Statewave docs and repositories on .