ropeshift
A free, fair-source, self-hosted software factory for people who run coding agents in parallel.
It is built to turn planned work into verified outcomes with a fleet of supervised agents, and to interrupt you only when a person is needed.
In development, with no release yet.
The problem
- Starting several agent sessions is easy. Keeping them productive for hours is not.
- Two sessions can take the same work, and nobody notices until a branch is damaged.
- Agents stop to ask questions, often the same ones, and sessions sit idle waiting for a person.
- Agents do not share one view of the goals, requirements, and past decisions.
- There is no single place to see what is planned, running, stuck, or finished.
How work flows
Ropeshift is designed so that work moves through these seven steps. None of them is released yet.
Rope yards laid one colored yarn into a rope to mark it. Here that yarn is you, and it touches the lane only at the last rung.
- Few interruptions
- The measure of success is human interruptions per completed work order.
- Never ask twice
- Answered questions are meant to be reused, and an open question shared, not repeated.
- No assumed roles
- Questions are organized by category, so one person can answer all of them, and on a team any member can answer any of them.
- Work arrives from a tracker, from repository files, or directly in the ledger.
- A person or a planning agent writes the plan and its contracts.
- Contract review collects every question it can find, in one batch, before work starts.
- An orchestrator claims ready work and starts a lane: one writer in one git worktree.
- A question climbs a ladder, and a person is the last rung.
- Existing decisions
- Open questions
- A decision within policy
- A recorded assumption
- A person
- A lane that truly cannot proceed parks. It saves its work, stops so it uses no quota, and resumes when the answer arrives.
- The lane ends with a typed outcome, artifacts, and check results, written back to the tracker.
The four parts
- Work ledger
- The system of record for execution: claims, outcomes, questions, decisions, and evidence. Your tracker stays the system of record for what people plan, and ropeshift is designed to work with no tracker at all.
- Planning and decisions
- Plans and versioned contracts, a contract review built to find questions before work starts, and a searchable record of every decision and why it was made.
- Supervised agent fleet
- One orchestrator per machine pulls work and makes decisions within policy. A deterministic daemon is built to keep each session healthy, and the design means no two machines can own the same work.
- Web UI
- The roadmap, plans, decisions, and open questions, and what is in progress, on hold, stuck, and completed, with a terminal view beside it.
Works with your tools
These integrations are planned. Team size changes configuration, not concepts, and no particular tracker or process is required.
- Work
- GitHub Issues and Projects, Jira, Linear, Notion, YouTrack, Odoo Projects, and repository files.
- Knowledge
- Confluence, Notion, and repository documents.
- Code
- GitHub and Bitbucket pull requests.
- Agents
- Claude Code, Codex, opencode, and other agent CLIs.
Ropeshift wraps agent CLIs. It does not replace them, and it is not a model client. Each CLI signs in through its own vendor's flow on your machine, and ropeshift is designed never to read, store, or relay agent credentials.
Ways to run it
Ropeshift is designed to run in three modes.
| Mode | How it runs |
|---|---|
| Solo | One binary on one machine, with no Docker and no network requirement. |
| Group | Docker Compose on a LAN or tailnet, with workers that dial out, so laptops open no inbound ports. |
| Cluster | A Helm chart and Terraform modules in your own cloud. |
Follow it, read it, run it
All of that will happen in the ropeshift repository, the one home for this project.