How it works

From dead repo
to pull request.

Six stations, one loop. Import your half-done repos, see where each one really stands, send the built-in agent to do the next step, approve its proposed diff, and let tests verify the result before you push. Here is each station, with the controls you'd actually click.

Importing 3 of 14 selected repositories
recipe-appReceiving objects 0% · 12 MB
newsletter-toolReceiving objects 0% · 48 MB
ml-side-questReceiving objects 0% · 310 MB · large

Blobless partial clones — 80–95% less download. Imports run in the background; keep browsing while they land.

How it compares

Honest about the alternatives.

These are good tools — we compare fairly, including the rows we lose. AI IDEs like Cursor and Windsurf are where you write code; app builders like Replit host and deploy for you; boards and runners excel at parallel sessions in one repo; cloud agents keep working while you sleep. Wanju is built around a portfolio of unfinished projects and a built-in proposal gate on the way to done — most people will use it alongside their IDE, not instead of it.

FeatureWanjuVibe Kanban & forks
Portfolio view of every repoScans your whole workspace or GitHub account — staleness, stage, and next step per project. Replit lists all your apps, without state assessment.●—
Guided path to *finished*Release journeys, evidence-backed readiness, and one recommended next step per project.●—
Built-in-agent proposal gateWanju's built-in agent queues diffs in an inbox under a deny-by-default broker. IDEs offer accept/reject per edit in one repo; builders use checkpoints.●◐
Auto test verification after applyWanju runs your test suite after every applied change and attaches the result. IDE and cloud agents may run tests inside their own loop; it isn't enforced on apply.●—
Approve from your phoneTime-limited, end-to-end encrypted sessions, no cloud account. Cursor's background agents and Replit's full mobile app also give phone control — via their clouds.●—
Markdown tasks shared with agentsPlain .md task files in each repo — humans, the app, and agents edit the same state.●—
Choice of AI providerWanju's built-in BYOK agent supports OpenAI, Anthropic, and Gemini. IDEs ship their own agent; other CLIs run in their terminal without integrated approvals.●●
Local-first — code stays on your machineIDE files are local but AI requests route through the vendor's cloud. Cloud agents and builders work in their VMs. Wanju's optional phone approvals relay only end-to-end encrypted diffs.●●
Audit log & emergency lockAppend-only record of every dispatch, edit, approval, push, and MCP call. Builders keep agent history with rollback checkpoints.●—
GitHub multi-repo import & pushPick any number of repos at once; fast partial clones with live progress; push to protected branches. IDEs and builders clone one repo at a time, with full git.●◐
Full IDE: completions, refactors, debuggingTheir home turf. Wanju's editor is for reviewing diffs and quick edits — it complements your IDE rather than replacing it.◐—
Inline AI autocomplete as you typeTab-complete is an IDE feature. Wanju focuses on project-level agent work and Source Control review instead.——
Hosting & one-click deployBuilders host and deploy your app for you; some cloud agents can too. Wanju verifies and pushes to GitHub — where you deploy stays your call.——
Keeps working while your machine is offCloud agents and builders run in their clouds; IDE background agents partially. Wanju is local-first by design, so it runs while your machine does.——
PriceFree forever for the dashboard, readiness report, and agent dispatch. Finish for me is $9/mo, $79/yr, or $149 once — free during the open beta. You bring your own API key, so AI usage is billed by your provider, not by us.FreeFree / freemium

● full · ◐ partial · — not offered. Based on public documentation, July 2026. Spotted an error? Tell us and we'll fix it.

Try the loop on
your own graveyard.

Start on the webDownload desktop