Skip to content

Jimothy documentation

Jimothy is a desktop app (Electron + Svelte) that turns issues from Jira, Linear or GitHub into shipped code. You describe the work as a pipeline: a graph of steps where each step is an AI coding agent, a shell command, or a human approval gate. Every agent step picks its own harness (Claude Code, Codex, Gemini CLI, Aider, an API, or any CLI you configure), model and prompt. Jimothy runs the pipeline in an isolated workspace, streams every step live, waits for you at approval gates, writes the result back to the issue, and notifies you wherever you are.

Everything runs locally on your machine. There is no Jimothy server: agents run with your user’s permissions, and data lives as plain JSON/JSONL files in the app’s data folder.

Account ──► Trigger ──► Pipeline ──► Run ──► Steps (agent / shell / approval)
(Linear, (filter + (steps, (one │ each agent step runs on a harness
Jira, pipeline) workspace, issue, │ (Claude Code, Codex, API, ...)
GitHub) variables) one ▼
workspace) Write-back (comment, transition)
+ notifications (desktop, Slack, ntfy, ...)
  1. An account is a signed-in connection to Linear, Jira or GitHub. You sign in once and reuse the account everywhere. See Accounts.

  2. A trigger combines an account, a filter (a JQL query, a Linear team and labels, a GitHub repo and labels), and the pipeline to run. Triggers poll the tracker, or receive webhooks, and start one run per matching issue. Schedules (cron) and generic webhooks are triggers too. See Linear, Jira, GitHub, Schedule and Webhooks.

  3. A pipeline is a list of steps plus workspace settings and variables. Steps depend on each other to form a DAG, so independent steps run in parallel; failed reviews can loop back to the implementer. Pipelines can be grouped into projects that share a workspace, commands and variables. See Pipelines overview.

  4. A run is one execution of a pipeline for one issue. Jimothy prepares a workspace for it (a git worktree on a fresh branch, a clone, an empty folder, or an existing folder) and executes the steps. See Runs & statuses and Workspaces.

  5. Steps render their prompt or command from templates (issue fields, earlier step outputs, variables) and run on a harness. See Harnesses.

  6. When the run starts, succeeds or fails, the trigger can write back to the issue (comment, move to a status), and notification channels fan the event out to desktop, Slack, Discord, Teams, ntfy or a webhook. See Write-back and Notifications.

The built-in Feature: Plan → Build → Review → PR template shows most features in six steps:

Step Type Runs on What it shows
Plan Agent Claude Code, opus Read-only planning prompt
Implement Agent Claude Code, sonnet Uses {{steps.plan.output}}, retries once
Run tests Shell your shell {{vars.testCommand | raw}}, loops back to Implement on failure
Cross-model review Agent Codex Pass pattern VERDICT:\s*APPROVE, loops back to Implement
Human approval Approval you Pauses the run until you approve
Open pull request Shell your shell git push + gh pr create

See Built-in templates for the full definitions.