Skip to content
Parsemend
Start free
source code stays put

Local CLI

The fix run, on your machine.

Run npx parsemend fix <issue> in the repository. The CLI pulls the context bundle for that issue, executes the fix run locally against your own model credentials, and opens the draft pull request. Your source never leaves the machine — Parsemend only ever sees the error events you already send it.

The four questions

Where the run executes
Your machine
Who pays for tokens
You, directly to your provider
Model choice
Anthropic · OpenAI · Gemini · Ollama
Source seen by Parsemend
Never

How the data flows

The CLI pulls the context bundle for one issue and runs the whole loop on your machine. Your source never crosses the network except to the provider you chose.

Your infrastructure
source codefix rundraft PR
Parsemend
error eventsfix queue
Model provider
model call

Fixing an error from the command line

terminal
$ npx parsemend fix ISSUE-4821  pulled context bundle (14 frames, 22 breadcrumbs)  root cause 0.94 auth retry races the session refresh  plan 0.91 (+4 -1)  draft PR opened, awaiting your review
  1. 01

    Install nothing

    Run npx parsemend fix <issue-id> from the root of the repository. There is no global install.

  2. 02

    Authenticate once

    The CLI reads a project token from PARSEMEND_TOKEN, and your model credentials from the provider's usual environment variable.

  3. 03

    Pull the context bundle

    The CLI fetches the stack trace, in-app frames and breadcrumbs for that issue. It reads the surrounding source from your working copy, not from Parsemend.

  4. 04

    Watch the run

    Root cause, then plan, then patch. Each step prints its confidence. Below the project threshold the run pauses and asks you.

  5. 05

    Open the draft PR

    The CLI pushes a branch and opens a draft pull request with the root cause in the description. It never merges.

What Parsemend contributes

  • Context bundles

    The stack trace, the in-app frames, the surrounding source, the breadcrumbs and the recent commits that touched those lines — assembled into one payload an agent can act on without crawling the repo.

  • Fix-run state machine

    Root cause, then solution, then coding, then change description. Each step is a checkpoint. A run that dies at step three resumes at step three rather than starting over.

  • Confidence gating

    Every step emits a confidence score. Below the project's threshold the run pauses and waits for a human instead of confidently shipping a wrong patch.

  • Draft PR post-back

    The patch arrives as a draft pull request against the right branch, with the root cause and the reasoning in the description. It never merges itself.

Who this is for

Anyone who cannot send source code to a third party, and anyone who wants to watch the run happen.

Compare all four modes or see what each tier includes.