Launching today

Celmis

Launching today

Self-hosted code intelligence over a symbol graph

1 follower

Self-hosted code intelligence over a symbol graph

1 follower

Visit website

AI Coding Agents

•

AI Infrastructure Tools

•

Observability tools

Most code tools read one repository at a time. Celmis indexes a set of them into one symbol graph, so a question can cross a boundary: ask which service still calls a function whose signature changed, and the answer comes back with file:line citations from a repository you never cloned. Same index, read many ways: PR review that sees cross-repo callers, dependency audit, CycloneDX SBOM, and an MCP server with 18 tools. Self-hosted, docker compose. Your model key, or Ollama locally. AGPL-3.0.

Free

Launch tags:Open Source•Developer Tools•Artificial Intelligence

Launch Team / Built With

ClarifyThe autonomous CRM that helps you sell more

Promoted

Maker

📌

The feature I actually wanted was a loop, not a tool. Something breaks in production at two in the morning. You are not at a desk, you have no checkout, and the laptop is in another room. Right now that means: read the alert, find which repository owns it, wake up enough to open the machine, remember the context, then work. So Celmis takes the alert. Any monitoring system posts to a per-workspace ingest URL — Grafana's unified alerting webhook is parsed as-is. A web push reaches the phone, real VAPID push, so it arrives whether or not a tab is open. The alert already knows which repository it belongs to. You press Fix with Claude, and the embedded session opens holding the repository, the package and both versions rather than an empty chat. It edits the checkout, the runner pushes a branch and opens the pull request. 220 seconds on a real lodash advisory, start to open PR. The rest of the product is that same index read differently: a deterministic dependency audit where no model decides what is vulnerable, a CycloneDX SBOM with an evidence pack whose sha256 manifest a third party can verify without trusting the machine that produced it, questions answered across repository boundaries with file:line citations, and generated documentation. Pull-request review is in there too and it scores decently — 17th of 50 on a public benchmark, under all three judges — but I treat it as a good addition rather than the point. Worth saying why: on that benchmark set zero of the 50 pull requests had a cross-repository caller, so it measured the product with its distinguishing capability switched off. I published that alongside the score, together with an audit of all 79 findings it counted as false positives. 33 of them were real defects the gold set did not contain. Self-hosted, AGPL-3.0, docker compose. Your model provider, or Ollama, vLLM or llama.cpp on your own hardware if nothing may leave the network. Happy to answer anything, including where it breaks.

Report

12d ago

Forum Threads

p/celmis

•

23h ago

Launching Thursday - the one design call I'd want pushback on

Scheduled for Thursday, 00:01 PT. Before it goes live, the decision I'm least sure about: "Fix from here" opens a full agent session rather than filing an issue or opening a draft PR directly. The session edits an isolated checkout, runs, verifies, and only then pushes a branch and opens the pull request a person still merges it. Slower than a one-click auto-PR, on purpose: I wanted a point where a human reads what actually changed before it enters a checkout, not after. Genuinely unsure this is the right tradeoff for everyone. If you've built something in this space where would you make the agent commit faster, and where would you make it slower?

View all