> ## Documentation Index
> Fetch the complete documentation index at: https://docs.nablrco.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Airlock
> Run concurrent sessions without stepping on each other
Run concurrent sessions without stepping on each other.
## The Problem
Multiple teammates - or multiple agent sessions - working the same repo at once used to share everything: the same checkout, the same branch, the same state files. Two sessions running in parallel could silently overwrite each other's active-story state, or get blocked by a gate that was actually counting another session's commits. No crash, no error - just wrong answers and false-positive blocks.
## What It Does
Isolates every concurrent session in its own git worktree, then merges finished work back into `dev` automatically:
* Each session gets a dedicated worktree on a story-scoped branch, with its own state - no shared active-story pointer, no cross-session interference
* Commit-atomicity gates only count that session's own commits since it branched from `dev` - never commits landed by other concurrent sessions
* Finished stories merge back as a single squash commit on `dev`, gated on the story's own closure checks passing
* Worktree and branch are cleaned up automatically on a clean merge
* Optional PR mode pushes the branch and opens a PR instead of merging locally, for teams that want human review at this step too
## Try It
```
activate nablr
Start story STORY-123
```
nablr creates (or attaches to) a dedicated worktree for the story the first time you initialize it - no extra setup. Run as many stories in parallel, in as many sessions, as you need.
When the story is done and its closure checks pass:
```
Merge story STORY-123 back to dev
```
For a PR instead of a local merge:
```
Merge story STORY-123 back to dev via PR
```
## Expected Output
* A dedicated worktree per active story, isolated from other concurrent sessions
* A single squash commit on `dev` per merged story, regardless of how many WIP commits it took to get there
* Automatic worktree and branch cleanup after a clean merge
* Conflicting-file list and a clean abort (never a stuck mid-conflict checkout) if the merge can't proceed
## Who This Is For
* Teams running multiple engineers or multiple agent sessions against the same repo at once
* Engineering leads who've hit silent state collisions or false-positive gate blocks under concurrent AI-assisted development
* Teams wanting PR-gated review even for agent-driven merges
Recommended ceiling is 3-5 concurrent sessions per repo. Past that, nablr logs a soft warning but does not block - very high concurrency increases the odds of two sessions racing to clean up the same worktree at once.