INXM / Open Source
Built in the open.
Orchestrator runs against the systems that already run the enterprise — and a lot of that plumbing is infrastructure everyone needs. Where we can, we build it in the open. Here is what we've published on github.com/inxm-ai.
Our projects
Python · FastAPI · GPL-3.0
Enterprise MCP Bridge
Most open source MCP servers are built for local development and simple demos — fine on a laptop, short of what production needs. Enterprise MCP Bridge is a FastAPI wrapper that exposes any Model Context Protocol server over plain HTTP/JSON and adds the enterprise layer around it: OAuth2 token exchange with automatic token injection and refresh, group-based data access, concurrent isolated multi-user sessions with pluggable storage backends, auto-generated OpenAPI docs, and SSE streaming for long-running operations. Point it at a local, containerized, or remote MCP server and it becomes a service your organization can actually deploy.
Rust · Desktop app · Publishing soon
INXM Local
A local-first desktop app for compiled-AI workflows — the Compiled AI method on your own machine. The LLM is the compiler, not the runtime: you describe intent in chat, the compiler produces a typed, validated plan, and a deterministic executor runs it. No AI improvisation in the execution path. Plans have persistent conversations, typed reusable inputs, human-in-the-loop approval steps, and a repair loop that hands failed runs back to the compiler. It manages your MCP servers in the UI and exposes its own local MCP server, so coding agents can compile and run INXM workflows too.
Forks we maintain
Alongside our own projects, we maintain forks of open source tools we build on and contribute back to.
Rust · Library + CLI
fork-llm
A Rust library and CLI for unifying and orchestrating multiple LLM, agent, and voice backends — OpenAI, Claude, Gemini, Ollama, ElevenLabs, and more — behind a single, extensible API. Build, chain, evaluate, and serve multi-step AI workflows including speech-to-text, text-to-speech, completions, vision, and reasoning.
TypeScript · MCP
forked-timetracker-mcp
A conversational-first time tracking platform built on the Model Context Protocol — a working example of what MCP-native tooling looks like when the interface is a conversation instead of a form.
Why we publish
INXM's product is trust: deterministic Plans, real audit trails, data that stays in your perimeter. Publishing our infrastructure in the open is part of that — you can read the code that sits between your systems and ours. Issues and pull requests are welcome on every repository.
Follow us on GitHub →