Skip to content
bkuberekPublic

About

A free, fair-source, self-hosted software factory for people who run coding agents in parallel.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

The ropeshift wordmark above separate hemp yarns that twist into one rope, with a blue marker yarn laid through it.

ropeshift

CI Release workflow Latest release Homebrew Go version Platforms License: FSL-1.1-ALv2 Last commit

Ropeshift is a free, fair-source, self-hosted software factory: project management combined with agentic workforce management and operations. It turns planned work into verified outcomes with supervised agents, and it is designed to interrupt a person only when a person is needed. It wraps agent CLIs: this release drives Claude Code, and drivers for Codex and opencode are part of the design. It does not replace them, and it is not a model client.

Website: ropeshift.com

Status

Alpha. Solo mode works: one machine, macOS or Linux, with the Claude Code CLI as the agent runtime, a local ledger, the web UI, and the top terminal view. Install a pre-release with brew install bkuberek/tap/ropeshift or the install script, then follow Get started with the web UI. Expect breaking changes before 1.0.0. The group and cluster modes described below are the design and are not built yet.

Install

Releases cover macOS and Linux on arm64 and amd64. Every channel installs one binary, ropeshift, with ropeshiftd as a link to it and the web UI built in.

Install with Homebrew:

brew install bkuberek/tap/ropeshift

Homebrew may ask you to trust the tap first. If it does, run brew trust bkuberek/tap once.

After brew upgrade, run brew services restart ropeshift, because the upgrade replaces the binary and leaves the running daemon on the old version.

Or install with the install script:

curl -fsSL https://raw.githubusercontent.com/bkuberek/ropeshift/main/install.sh | sh

With no --version, the script installs the newest stable release, or the newest pre-release while no release is stable, and says which it chose. It checks the archive's SHA-256 against the release's checksums file before it extracts anything, and installs to ~/.local/bin. Options go after sh -s --:

curl -fsSL https://raw.githubusercontent.com/bkuberek/ropeshift/main/install.sh | sh -s -- --version v0.1.0-alpha.2 --prefix /usr/local

On Debian, Ubuntu, Fedora, or RHEL, download the .deb or .rpm package for your architecture from the releases page, and install it with sudo dpkg -i or sudo rpm -i.

go install is not an install channel for now: install from a release.

Ropeshift also needs git, tmux for lanes in a terminal pane, and the Claude Code CLI installed and signed in. Then follow Get started with the web UI, from setting up the machine to signing in to the web UI. Releasing lists every channel.

Screenshots

These are the web UI of v0.1.0-alpha.15 and ropeshift top, running locally with sample data. The web UI follows the system's light or dark setting. The web UI guide shows each of its main views.

Needs you puts what waits for a person first, oldest first: work to accept, work its lane could not finish, a parked lane, and questions an orchestrator could not settle, with its recommendation marked.

Needs you, with work to accept and the review's verdict on each criterion Needs you, with work to accept and the review's verdict on each criterion

Decisions lists what members and orchestrators decided or assumed, with the question each answers and what it cites.

Decisions, with a decision and an assumption and their scope, question, and citations Decisions, with a decision and an assumption and their scope, question, and citations

Floor shows each worker and its lanes, with each lane's work order, state, reach, generation, and freshness.

Floor, with one worker and its lanes in the working, ended, parked, and needs-a-decision states Floor, with one worker and its lanes in the working, ended, parked, and needs-a-decision states

Measures shows the fleet mode, its limits, and how many work orders and lanes are in each state, followed by the sources and the product measures.

Measures, with the fleet mode, limits, and counts of work orders and lanes by state Measures, with the fleet mode, limits, and counts of work orders and lanes by state

ropeshift top watches the same fleet from a terminal.

ropeshift top, listing what needs you, the lanes, and the worker

What it is

Ropeshift has four parts. Each is built for solo mode in this release:

Part What it provides in this release
Work ledger The system of record for execution, in a local SQLite ledger per project: claims, leases, outcomes, questions, decisions, and evidence. One installation holds many isolated projects, and a project can span several repositories. Work comes from repository files or is recorded directly, so ropeshift works with no tracker at all
Planning and decisions A planning session with one agent that turns your answers into requirements, documents, milestones, contracts, and work orders, approved as one plan whose branch then lands at the merge authority level you choose; ropeshift adopt to bring an existing project in; a contract review that finds questions before work starts; and a searchable record of every decision and its rationale
Supervised agents One orchestrator per machine pulls work and exercises judgment. A deterministic daemon keeps each Claude Code session healthy, and a fence stops a holder that lost its claim from writing. Every session can hand work to an agent roster you edit with ropeshift agent, each lane runs as the agent its work needs, and the costly agents ask first. A review agent checks each finished change against its contract and gives every criterion a verdict. Accepted work lands at the merge authority level you choose: you merge it, ropeshift merges it after your approval, or ropeshift merges it after the review's approval with every required check green, always at the reviewed commit, through a GitHub pull request with your own gh sign-in or into the base branch with git, and every approval and merge is recorded
Web UI Needs you across every project, with the confirmation cards a delegate prepared; the roadmap with each milestone's work by dependency layer, plans and their versions, plan approval with the groups of an adopted plan, the planning pane beside the plan a session is writing, documents with their requirements traced to criteria, work, and verdicts, decisions, and open questions; work orders, the floor of lanes, and measures; conversations with the project's assistant, an orchestrator, a planning session, or a lane, in a dock on every view and in full on the Conversations view, where the assistant's confirmation cards and a session's approval requests and questions stand inline; each project's repositories, agent roster, settings with the delegate policy and the assistant's idle stops, and audit, which names each delegate and the member it acted for; and the installation's projects, workers, members with their delegate credentials, and usage. Projects are created, imported from a backup upload, backed up as a download, and adopted from an existing folder of clones there, served on loopback, with ropeshift top beside it in a terminal

These parts of the design are not built yet: reading work from an issue tracker and writing results back to it, workers on more than one machine, and drivers for Codex and opencode.

How work flows

  1. Work arrives from repository files or directly in the ledger. Reading it from a tracker is part of the design.
  2. A person or a planning agent writes a plan and its contracts.
  3. Contract review collects every question it can find, in one batch. Work becomes ready only when it passes a definition of ready.
  4. An orchestrator claims ready work and starts a lane: one writer in one worktree.
  5. When a lane has a question, it climbs a ladder: existing decisions, open questions, a decision within policy, a recorded assumption, and only then a person.
  6. A lane that truly cannot proceed parks. It saves its work, stops its process so it uses no quota, and resumes when the answer arrives.
  7. The lane ends with a typed outcome, artifacts, and check results, and a member accepts the work or sends it back for rework. Writing the result back to a tracker is part of the design.

The north-star metric is human interruptions per completed work order. ropeshift measures prints it beside the measures that keep it honest.

Ways to run it

Team size changes configuration, not concepts, and every mode is designed to use the same versioned API. Solo mode is the one built in this release:

Mode How it runs Store Status
Solo One binary on one machine, with no Docker, no account, and no network connection of its own (the agent CLI reaches its provider) SQLite Built
Self-hosted group Docker Compose on a LAN or tailnet, with workers that dial out, so laptops open no inbound ports Postgres Design, not built yet
Cluster A Helm chart and Terraform modules in your own cloud Postgres Design, not built yet

Agent sign-in

Each agent CLI is installed and signed in on each worker, through its vendor's own flow. Ropeshift never reads, stores, relays, or proxies agent credentials. A worker that uses a subscription is bound to the person who signed in, and shared workers use API keys.

Claude subscription accounts

Ropeshift runs the unmodified claude binary that you installed and signed in to, so a lane can run on your Claude subscription. It is an independent project, and it is not affiliated with or endorsed by Anthropic, OpenAI, or any other vendor.

Anthropic's Consumer Terms restrict automated use of subscription accounts, and its usage limits assume ordinary, individual usage. Running the unmodified binary does not by itself resolve that, and autonomous operation moves further from ordinary use. Running automated sessions carries account risk that you accept.

A subscription worker starts at attended, where it claims work only inside a run window a person opened, and goes further only after a member acknowledges the stated risk for that worker. No vendor has approved this use. Onboarding states the risk in full.

Documentation

Learn ropeshift

  • Get started with the web UI: install ropeshift, set up the machine, start the service, sign in to the web UI, and find your way around it.

Concepts

  • How ropeshift works: the plan, the lanes, the review, the question ladder, and where a person comes in, in plain words.
  • Vision: the problem, the principles, and the non-goals.
  • Project structure: the recommended layout of a project, the repository file format, and what stays in your repositories.
  • Authorization: who may approve what, and the limit of same-user isolation.
  • Product measures: the north star and how each measure is computed.

Reference

  • Onboarding: the detail behind every step of the guides, and what to know before a lane runs.
  • Set up a project: from a fresh machine to a project with an approved plan and merged work, with each choice explained.
  • Projects of several repositories: one plan across repositories, the manifest, dependencies, read-only reads, and merge order.
  • Your first plan, end to end: a planning session, plan approval, lanes, reviews, and acceptance, in the web UI and on the command line.
  • Adopt an existing project: bring in a project that already has code, documents, and branches, and review it as one plan.
  • Connect a GitHub repository: review through pull requests, the forge binding, merge authority, and required checks.
  • Web UI: each main view of the web UI, what it shows, and what a member does there.
  • Command line: every command, the daemon's flags, and the exit codes.
  • API: the routes of the installation and of each project.
  • Qualification: how each Claude Code version is checked before ropeshift relies on it.
  • Notifications: what a notice says, and where it goes.
  • Telemetry: what is recorded, where it stays, and when it leaves the machine.
  • Versioning: what a version promises, and what ropeshift version reports.
  • Releasing: what a release holds, and each install channel.
  • Troubleshooting: what to do when the daemon refuses to start on a database.

License and contributions

Ropeshift is fair source under the Functional Source License, FSL-1.1-ALv2. See LICENSE.md. Each version also becomes available under Apache-2.0 two years after its release. Outside code contributions are not open yet.

The developer documentation covers building, testing, and releasing ropeshift, and capturing its screenshots.

About

A free, fair-source, self-hosted software factory for people who run coding agents in parallel.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Sponsor this project

Packages

Contributors

Languages