v0.9.2 · public beta

A coding agent that lives in your terminal.

Somcode reads your whole repository, prints the plan in plain text, edits the files, and runs your test suite before it tells you it is done.

$ curl -fsSL https://somcode.fmys.dev/install | sh

Free while in public beta · macOS and Linux · what is included

somcode - acme/billing-api

$ somcode

◇ Somcode v0.9.2

◇ Indexing 1,284 files ... done (0.8s)

> the retry helper swallows the original error, fix that

◇ Reading src/webhooks/retry.ts

◇ Reading src/webhooks/handler.ts

◇ Plan

1. preserve the cause when re-throwing

2. record the attempt number on the wrapped error

3. add a regression test for the 5xx path

Apply this plan? [y/N] y

◇ Editing src/webhooks/retry.ts +14 -6

- throw new Error('retry failed');

+ throw new Error('retry failed', { cause: err });

◇ Editing src/webhooks/retry.test.ts +31 -0

◇ Running npm test

✓ 142 passed, 0 failed (3.4s)

◇ Done. 2 files changed, 45 insertions, 6 deletions.

Built for repositories you already care about

Not a chat window that pastes snippets at you. An agent that works inside the project, with the same constraints you have.

[]

Repo-aware, not file-aware

Somcode indexes symbols and imports across the project, so it opens the files that actually matter rather than the ones you remembered to mention.

Runs your tests, not its own

Every change is checked against the test command you already use. If the suite is red, Somcode says so instead of declaring victory.

Plan first, then apply

It prints the plan and waits. You approve, it edits. No surprise rewrites of files you never asked it to touch.

Local-first by design

The index, the history and the diffs stay on your machine. Only the model calls leave it, and only with the context each call needs.

Bring your own model

OpenAI, Anthropic, or any OpenAI-compatible endpoint including a local Ollama. Switching is one line in the config.

Scriptable

somcode -p "..." runs non-interactively with a plain exit code, so the same agent works in CI and in a pre-commit hook.

Three steps

Run it in the repo

somcode in the project root. It indexes the codebase once and keeps the index fresh as you work.

Describe the change

Plain English, the way you would describe it to a colleague. No prompt engineering, no file list to assemble by hand.

Approve and let it run

Read the plan, approve it, and watch the edits and the test run. Review the diff like you would review anyone else's pull request.

Works with NodePythonGoRustJavaRuby And any editorany CI

Questions

Does it send my code anywhere?

The repository index, the session history and the diffs never leave your machine. Each model call sends only the context that call needs, to the provider you configured.

Which models does it work with?

OpenAI, Anthropic, or any OpenAI-compatible endpoint, which includes a local Ollama server. You supply the key or the endpoint; there is no Somcode account.

Is it really free?

Yes, for now. Somcode is in public beta and we are not charging for it. If that ever changes, the local CLI stays free and anything paid would be a hosted add-on.

Can I use it at work?

Yes. Commercial use is fine, there is no seat count and no per-repository limit.

Why is the installer invite-gated?

The beta channel is deliberately small while we work through platform-specific bugs. Send a note to hello@somcode.fmys.dev and we will send the current build.

Windows?

Linux and macOS today. Windows support is in progress and will land during the beta.