10xGraph by 10xScale: graph engineering for production AI agents.
Website · Docs · Quickstart · Changelog
10xGraph is an open-source Python framework for multi-agent AI systems. Other frameworks give you the graph. 10xGraph gives you the graph and the production server around it, under the MIT license. Formerly Agentflow.
- The production server, in the box.
10xgraph-apiserves your compiled graph over REST, SSE, WebSocket and realtime audio, with JWT or custom auth, scoped authorization on every endpoint, owner-only threads and rate limiting, and writes your Docker Compose and Kubernetes files. No paid tier. - Correct under failure. A tool that already ran is not run again after a crash or retry: no double charge, no duplicate email. State writes are versioned, and node and tool timeouts stop a hung tool from holding a worker.
- Built to scale. Active threads are cached in Redis, durable history lives in PostgreSQL, and events stream to Kafka, Redis, RabbitMQ or OpenTelemetry.
- One stack, backend to frontend. A typed TypeScript client, and remote tools that run in the user's browser.
- You own it. Self-hosted, any model (OpenAI, Google Gemini, Anthropic), no LangChain dependency.
10xGraph is pre-1.0: pin your versions and read the changelog when you upgrade.
| Package | Install | Repository |
|---|---|---|
| Core engine | pip install 10xgraph |
10xGraph/10xGraph |
| Production server and CLI | pip install 10xgraph-api |
10xGraph/10xgraph-api |
| TypeScript client | npm install 10xgraph-client |
10xGraph/10xgraph-client |
| Website and docs | 10xgraph.com | 10xGraph/10xgraph-docs |
Your first pull request can be small: a typed module, an example, a fixed doc. Start with the contributing guide, meet the people who build 10xGraph at 10xgraph.com/maintainers, and ask questions in Discussions. Report security issues privately through the Security tab of the affected repository.
MIT licensed. Built and run in production by 10xScale.