Skip to content

Laneward

Runs several coding agents on one repository at the same time, each in its own git worktree, and keeps them from colliding. Approved work continues after you close your editor.

Install it Read what it does not protect you from

It does not plan work and it does not write code. You decide what needs doing; Laneward records each task, decides when it is safe to start, spawns the agent, checks what it touched, and shows the result on one screen.

Read this before you install, not after

Laneward listens on 127.0.0.1 with no authentication. It is built for one trusted single-user machine. Do not expose it.

Each lane's worktree receives a copy of the driven repository's .env, so an agent can read your real secret values. Only DATABASE_URL is rewritten, to a database created for that lane. Redaction is not implemented. See Safety and limits.

The three moving parts

The hub is a long-running web service. It owns every record (plans, lanes, messages, approvals, notifications), answers HTTP on 127.0.0.1:8787, and serves the dashboard. It decides nothing on its own: it answers questions and stores answers.

The conductor is a loop that does the work the hub only records. Every five seconds it asks the hub what it may start, spawns an agent for each lane it is allowed to run, scores what the agent touched, and reports back. It exists so work continues while no editor session is open.

The agent is whatever coding agent you declared: Codex, Claude Code, or a raw command of your own. It reads a brief on stdin, works inside the worktree it is given, and signals with its exit code. Laneward ships no default agent, and refuses the first lane until you declare one.

A lane

A lane is one bounded task given to one agent. It owns:

  • a set of file paths nobody else may touch while it is unfinished,
  • a worktree, at <repo>-worktrees/<lane_id>,
  • a branch, lane/<lane_id>,
  • usually a database of its own, created by copying the driven repository's connection and renaming the database.

Two lanes whose paths overlap cannot run at the same time, and the hub refuses to register the second one at all while the first is unfinished. That refusal is the whole point of the system.

How a lane ends

How a lane ends A lane is registered as pending, becomes running once every gate passes, and from running reaches one of three ends: completed, failed, or waiting_approval. A retryable failure sends it back to pending for attempts 1 and 2, and a resolved approval does the same. registered pending running completed failed waiting_approval every gate passes exit 0, evidence clean, checks pass Git boundary violation, or attempt 3 exit 10, or an escalation marker retryable failure, attempts 1 and 2 the approval is resolved

completed means the agent exited 0, touched only what it owned, and the checks the driven repository declares passed. It does not mean the work is correct, reviewed, committed, or merged. Commit and merge stay manual by design.

What it will not do for you

  • Plan work, split it into lanes, or write briefs.
  • Commit, merge, push, or open pull requests. All four are deliberately manual.
  • Redact secrets from the .env it copies into a lane worktree.
  • Prevent a second conductor from running against the same database.
  • Authenticate anything, or serve anywhere but 127.0.0.1.

Project status

Development has stopped, and the reason is money rather than interest. What works was measured rather than assumed, and the runs are written up in the evidence notes, including what each one failed to establish. Issues and pull requests may go unanswered. The licence is MIT: fork it, take it somewhere, no permission needed.

If you are looking for something to run in production, this is not it. If you are looking for a working design to read, take apart, or continue, that is exactly what is here, and this guide is how you get it running.

Where to go next

  • Install and first run

    A hub, a conductor and a database, from a clone, on this machine.

    Install

  • Configuration

    Declare an agent, fill the model tiers, and read every variable there is.

    Configure

  • Your first lane

    One real task end to end: brief, lane, agent, verdict, work in your hands.

    Drive a lane

  • Writing briefs

    The contract the agent actually reads, and how to make it score correctly.

    Write a brief

  • Troubleshooting

    Why a lane will not start, and what every refusal in the system means.

    Diagnose it

  • Safety and limits

    No authentication, a copied .env, and what has actually been verified.

    Read the limits

The glossary is the vocabulary this project uses in a specific sense, and the architecture series is the design underneath, including the decision log every D-0NN reference points at.