| name | agentic-jujutsu |
| version | 2.3.2 |
| description | Lock-free version control for multiple AI agents committing concurrently to the same repo, built on Jujutsu (jj). Use when several agents need to commit/branch/rebase simultaneously without lock contention, for conflict-free concurrent worktree isolation, or when you want per-agent operation tracking with learned suggestions for repeated jj workflows. |
Agentic Jujutsu — AI Agent Version Control
Lock-free, self-learning version control designed for multiple AI agents
working simultaneously without conflicts. Wraps the Jujutsu (jj) VCS with a
ReasoningBank trajectory store (learned suggestions), AgentDB operation
tracking, and quantum-resistant integrity fingerprints.
When to use
- Multiple AI agents are modifying the same repo simultaneously and you want to
avoid Git's lock contention and serialised commits.
- You need conflict-free concurrent worktree isolation across agents.
- You want operations tracked per agent, with pattern discovery and learned
suggestions for repeated workflows (deploys, merges, reviews).
- You want fast quantum-resistant (SHA3-512 / HQC-128) integrity checks on
commits or trajectories.
When not to use
- Standard single-actor Git operations with no multi-agent coordination — use
Git directly.
- GitHub PR workflows, issue tracking, or release management — use the
github-code-review / github-release-management skills.
- General agent memory and pattern storage without version control — use
agentdb-memory-patterns.
- CI/CD pipeline automation — use
github-workflow-automation.
Quick start
npx agentic-jujutsu
const { JjWrapper } = require('agentic-jujutsu');
const jj = new JjWrapper();
await jj.status();
await jj.newCommit('Add feature');
await jj.log(10);
const id = jj.startTrajectory('Implement authentication');
await jj.branchCreate('feature/auth');
await jj.newCommit('Add auth');
jj.addToTrajectory();
jj.finalizeTrajectory(0.9, 'Clean implementation');
const suggestion = JSON.parse(jj.getSuggestion('Add logout feature'));
console.log(`Confidence: ${suggestion.confidence}`);
The core loop is: startTrajectory(task) → do work (auto-tracked) →
addToTrajectory() → finalizeTrajectory(score, critique). Later,
getSuggestion(task) returns a learned recommendation. Use honest success
scores (never always 1.0) so the model can learn.
Learning memory (ReasoningBank trajectories, patterns, coordination state) is
stored in centralized RuVector PostgreSQL, retrieved via pgvector HNSW.
Detailed references
references/cookbook.md — worked examples per capability (self-learning,
pattern discovery, multi-agent coordination, quantum security, operation
tracking), advanced multi-agent use cases, best practices, and end-to-end
examples.
references/api.md — full method catalog (Core / ReasoningBank / AgentDB
/ Quantum), validation rules, performance characteristics, RuVector storage
details, troubleshooting, and version history.