Repository navigation
duobrain v0.1.0
duobrain v0.1.0
duobrain gives exactly two people and their existing AI tools a shared project memory backed by Git.
This first release includes a local engine, read-only dashboard, and AI collaboration guides under the MIT license.
Included
- Shared plans, roles, scoped sessions, overlap checks, and handoffs on an isolated
duobrain/statebranch. - Information requests with source evidence and human feedback requests with attributable responses.
- Source-preserving wiki records, search, captured-method comparison, and on-demand wiki refinement.
- Resumable AI-led onboarding and participant nicknames.
- A dashboard organized into task tabs, in English or Korean from the browser language, with relative times,
plain-language event labels, whose turn each open request is waiting on, and refresh while the page is visible. - AI setup through the product's
AGENTS.md:initwrites a managed block (imported byCLAUDE.md),
guideresolves local paths, andagents-syncandupdatekeep the block and the tool current. - A short English AI guide with on-demand references, and compact
status --brief/overlap --briefoutput. - Commit work records:
capture-extract,commit-noteand theduobrain-commitskill store how and why
each commit was made onduobrain/state, linked from aDuobrain-Recordcommit trailer.
Run
Requires Node.js 22+ and Git; no npm install is needed.
git clone --branch v0.1.0 https://github.com/slowspurt/duobrain.git
cd duobrain
npm link # optional: puts `duobrain` on the PATH; nothing is installed
node bin/duobrain.js --version
node src/dashboard/run.jsOpen http://127.0.0.1:4173 for sample records. To connect a real two-person project, follow
Getting started.
Keep the tool checkout separate from the product repository. Later versions install with duobrain update.
Verification and limits
The engine, wiki, dashboard, capture and tool suite contains 104 tests. Release preparation also runs
the concurrent-work, plan-refinement, and method-refinement scenarios against temporary Git repositories.
CI is configured for Node.js 22 and 24 on Linux and macOS; configuration alone is not a passing CI run.
Actual acceptance with two people on separate machines and their existing AI tools remains pending.
There is no background AI service, automatic partner wake-up, real-time lock, or scheduled wiki refinement.
The dashboard reads locally synchronized records; pending delivery requires sync retries.
Commit work records are written by the user's AI from the local Claude Code transcript; summaries can
contain mistakes and should be checked before committing. Other agents write the summary themselves.
Distribution is through GitHub source archives and Git checkout. No npm package is published.