Open-Source AI Developer Stack
Self-hosted agent workbenches, model routers and workflow platforms you can run on your own infrastructure.
Running your own AI stack means owning the routing, the credentials and the data. This collection gathers the open-source pieces developers actually deploy: agent workbenches that give an agent its own browser and tools, gateways that route coding clients across many model providers, and workflow platforms for building on top of them. Every project here is self-hostable and links to its upstream repository.
Head-to-head comparisons
Related collections
Frequently asked questions
What belongs in an open-source AI developer stack?
Four layers cover most self-hosted setups: an agent workbench such as OpenBot, a model gateway such as 9Router or OmniRoute, a workflow platform such as Dify or Flowise, and an assistant such as OpenClaw that ties them together.
Are 9Router and OmniRoute really free?
Both are MIT-licensed and self-hosted, so the software costs nothing. The free-token headlines in their READMEs describe third-party free tiers that you connect with your own accounts, and those limits change often. Treat the numbers as project claims and verify them before you rely on them.
Which coding clients can these gateways serve?
Both projects document Claude Code, Codex, Cursor, Cline, Copilot, Antigravity, Gemini, OpenCode and OpenClaw as supported clients, using an OpenAI-compatible endpoint.
Where do I find LLM APIs with free tiers?
The Free LLM APIs collection in this cluster lists providers with recurring free tiers, free model routes and limited developer credits, along with the limits that apply to each.