llmthing Built by Rytter

Planned · the design is written

Agents do the work. Checks decide.

Coding agents leave sessions everywhere and decide for themselves when they're done. llmthing finds any session and lets checks decide.

How it works Built by Rytter

$ pharaoh run workflow.yaml
BUILD codex implemented the task · 3 commits
VERIFY cargo 2 of 148 tests failed
BUILD codex retry 1 of 3 · with verify.output
VERIFY cargo 148 of 148 passed · clippy clean
REVIEW claude approved · 1 note left in the report
ACCEPT verify.passed · review.approved
DONE report and branch handed off

Features

Two tools, one job each.

  • Find any session

    Every agent's sessions on the machine, indexed with their repository, worktree and branch.

  • Resume where it lived

    Back into the right directory, on the right branch, with the right agent, however long ago.

  • The loop in a file

    Build, verify, review and accept are stages in a file, with a limit on retries.

How it works

Checks decide, not the chat.

  1. 1

    Build

    The agent builds, and a retry goes back carrying the failing output, so it fixes what broke.

  2. 2

    Verify

    Verification is commands with exit codes, so passing is decided by the tests, not the agent.

  3. 3

    Review

    A reviewer can reject, and a run that is blocked or out of retries stops and asks you.

Who built it

Built by Rytter.

I'm Jakob Rytter. I build software and run it, from the server to the screen, and llmthing is how I mean to run my agents.

rytter.it All the things