Every plan, build, bug and fix — captured as your agent ships, closed by your own commits. Nothing slips, nothing's lost.
You've hit /clear. The agent forgot the three bugs you found yesterday. The plan lives in a chat you can't scroll back to. The build history lives nowhere.
Vibstr is the memory layer. The agent ships; Vibstr keeps the record.
The value isn't the loop. It's that you only do step 1 and 3.
One-click from Settings → PROJECT → Connect Claude Code. Paste the master snippet into your CLAUDE.md once. You never paste anything per-build again.
# Vibstr workflow
When the user asks for a build:
1. POST to vstr-plan-create
with the plan body +
derive_items: true
2. Read the returned
item_numbers from the response
3. Code against them
4. Commit with "closes #NN"
references
5. Run the build-report curlClaude reads your CLAUDE.md on every session start. The loop is now in muscle memory — Claude's, not yours.
Tell Claude "fix the search ranking." Claude writes the plan, creates the items in Vibstr via the API, codes the change. You watch.
vstr-plan-create
→ 3 items derived
#342 Server-side ranker
#343 Cascade re-roll on
hero swap
#344 Vitest sweep green
before pushThe items are in Vibstr. You didn't file them. You didn't write the plan.
Claude writes the commit message. You push.
git commit -m "Add server-side
ranker -- closes #342 #343 #344
Six sources, $1 floor, pantry-led
ranking."Items flip to done. Build records the version, the commit hash, the prompt, an auto-generated changelog. Any .md files touched by the commit — specs, plans, release notes — auto-link to the build's Documents tab. You filed nothing by hand.
The closes #NN parser is deterministic. It runs first, before any AI touches anything, and it costs zero AI credits. The AI semantic matcher only catches what your commit message didn't name explicitly.
We use AI as little as possible. That's the point.
Bulk-ship items by hand. Parse plans with zero AI. Post-commit script for bash, PowerShell, or GitHub Actions. Manual-first parity is non-negotiable.
No lock-in. Never was.
We never see your AI bill. We never mark up compute. Pay us for the workflow, pay Anthropic for the tokens.
No sprints, no story points, no points-vs-hours holy war. Labels carry the rest.
VivifyScrum service ends June 30, 2026. The migration importer is live — upload your sprint columns export, map columns to status, hit Import. Five minutes, tested against real 51-item sprints.
Vibstr won't make you better at AI coding.
It's the record layer — the memory and the ledger. The agent still drifts, still forgets mid-session. What changes is that the durable record of your plans, builds and bugs survives outside the context window.
It's a system of record for AI-driven dev. Not a fix for the agent's amnesia.
Set it up in five minutes. Or read exactly how it works first — the curl, the schema, the whole loop. We don't bury any of it.