CŪ·RĀ·BŌ — LAT. “I WILL TAKE CARE OF IT.”
Curabo runs work through your own tools — from a ticket to a merged, validated change today — with agent seats you can hold to account. Every decision is recorded with its grounds, every rule can be replayed over the record, and trust widens only by your hand.
EARLY ACCESS IS SIGNED IN COHORTS · WE REPLY FROM INFO@CURABO.AI · NO SPAM, ONE THREAD
HOW IT WORKS
Work arrives from Jira, GitHub or Slack. Agent seats classify it, write the spec, implement, and review it through a panel of security, architecture and test lenses. At each risk-weighted gate the work waits for a signature — yours, or a seat you’ve formally delegated — and every step writes a row you can read back.
From your work platform, classified by risk
The spec awaits your hand before a line is written
One dev agent; security, architecture and test verdicts
Evidence, cost and diff on one card. Sign, or veto with a reason
After production: “is this what you needed?”
SOFTWARE DELIVERY IS THE FIRST WORKFLOW · THE SPINE DOESN’T KNOW THE WORK IS CODE
THE LEDGER
Hash-chained decisions with the prompt and policy version on every row, dual attribution on every action, session transparency with takeover. Because the rules are pure, a proposed rule can be replayed over the whole record and asked what it would have done — before anyone is asked to trust it.
SIGNED IS SIGNED — THE INK DRIES
CALIBRATED AUTONOMY
Curabo never promises fewer humans — it earns the curve. Every signature and veto is evidence for what could run alone next. When outcomes degrade, autonomy narrows on its own. It never widens on its own: widening is a person’s decision, made on that evidence.
TARGET · HUMAN-MIN / ITEM: 30 → <5 · UNMEASURED TODAY
YOUR TOOLS, YOUR MODELS
GitHub runs the cycle today; Jira, Slack and Confluence feed it — read freely, write only through gates. Bring your own model keys; a model works a seat only after passing that seat’s certification.
GITHUB TODAY · GITLAB & BITBUCKET NEXT · BRING YOUR OWN KEYS
TEAM COMPOSER
“Mobile fintech app, Go backend, two humans.” The Composer assembles a team from the Role Library — a PM, a stakeholder interface and a reporter beside the engineers. Every seat is a position that outlives the model in it: a job description, a jurisdiction, an employment record, and the right to raise a flag where a person would. Not personas: every role ships with its own evals, re-run after any override you make. Max six active seats — the evidence-backed cap.
ROLES ARE ASSET PACKS, NOT PERSONAS · A SEAT KEEPS ITS RECORD · OVERRIDE ANYTHING, EVALS RE-RUN
Called from refund-calculator.ts:114 on a request path. Agent-drafted upgrade PR awaiting your review: +2 −2, all 137 tests pass in sandbox.
INSIGHT ENGINE
Every per-PR review tool stops at findings; raw alert streams get actioned 2–5% of the time. Curabo runs scheduled, repo-wide analyzers — vulnerable dependencies triaged by reachability, live leaked secrets, flaky tests, hotspot debt — and because it has execution agents, the top of the ladder is a drafted, tested fix awaiting your signature. Capped at a handful per week, ranked by impact. Dismissed with a reason, it never resurfaces unchanged.
MEASURED ON ACTIONED-RATE, NEVER INSIGHT COUNT · NO “TECH DEBT = N DAYS” THEATER
BUILT FOR YOUR AUDITOR, TOO
Ticket → spec → session → diff → verdicts → signatures → outcome, hash-chained, with a standalone verification script.
Every signature stores the hash of the exact diff presented — provable either way. Absences are events too.
Chain checkpoints land in a bucket you own — Object Lock checked against the bucket, never assumed — and GitHub’s own audit log corroborates every bot action.
No per-person performance scores, no behavior-scored routing of work to humans. A product rule, not a policy PDF.
Tell us about your team and stack. We onboard in small cohorts — read-only first on one repo; nothing writes until you sign.
OR WRITE TO US — INFO@CURABO.AI · SOLO FROM $29/MO · SQUADS FROM $750/MO