About
One control plane for the whole LLM stack.
AcruxCore is an LLM-ops platform for engineering teams — prompt versioning, an OpenAI-compatible gateway, tracing, a tool catalog, evaluation, and an audit trail over all of it, wired together so a fix flows from a bad answer back to a live prompt without a redeploy.
Why we built it
Shipping an LLM feature is easy. Keeping it good is not. Prompts live in code and need a deploy to change. Calls fan out to different providers with no shared record of cost or latency. When an answer is wrong, there is no straight line from the bad output back to the prompt that produced it. Teams end up stitching four or five tools together and still fly half-blind.
AcruxCore closes that loop. Land on a session, open its trace, rate the span that missed, and jump to the exact prompt version behind it — edit, save, move the production alias, and the next run picks up the change. No redeploy, no context-switch.
What we believe
- Change without a deploy. Prompts and tools are versioned data, not code. Moving a production alias should take seconds, not a release.
- Your keys, your data. Bring your own provider keys, or self-host the whole platform. We sit in front of the models; we never take them over.
- Drop-in, not rip-and-replace. The gateway is OpenAI-compatible and each part works on its own, so you adopt what you need and keep the rest.
- Everything is measurable. Every call is a trace with spans for model, tokens, latency, and cost — no extra instrumentation to bolt on.
- Open, by default. The platform is open source under the Apache License 2.0 — read the code, self-host it, or send a pull request.
Built for developers
AcruxCore ships first-class SDKs for both TypeScript and Python, plus a plain REST API for everything else. Read the documentation or browse the source on GitHub to see how it fits your stack.