PROMI · Master Orchestrator

The Operating System
for Your AI Worker Team

PROMI doesn't just route tasks. It coordinates AI workers, enforces quality, manages schedules, detects failures, and keeps your entire AI operation running — without manual oversight.

Docs companion

Read the PROMI orchestration guide.

The docs version explains the lifecycle behind goals, handoffs, verification, review gates, and promotion.

What PROMI Does

🔀

Task Routing

Analyses each task and routes it to the worker with the right skills. Backend work goes to Sophi. UI work goes to Marv. Documentation goes to Sage. No manual assignment needed.

🤝

Autonomous Handoff

When Sophi finishes an API, PROMI automatically hands the work to Marv for the frontend — with full context. Files changed, decisions made, what to do next. Zero meetings.

🛡️

Quality Gates

Two-stage code review: spec compliance first, code quality second. Verification gate: workers must prove work is done with evidence, not just claim it. "Done" means "proven done".

Circuit Breaker

Detects patterns of failure and automatically halts work before errors cascade. If a worker hits 3 failures on the same problem, PROMI stops and escalates — no runaway AI.

📋

Sprint Management

Plans sprints, assigns tasks, tracks progress, manages dependencies. The task board shows what every worker is working on, what's blocked, and what's next.

🏛️

Steering Floor Enforcement

Your organisation's inviolable policies are enforced on every worker, every decision. Data residency, encryption, access controls — impossible to bypass, not just difficult.

The Telepathic Link

PROMI connects to every AI worker through Cortex — the shared memory system. Every decision, every lesson, every event is visible to all workers in real time. This is not prompt chaining. This is autonomous collaboration.

You describe
Plain English
Keith specs
Requirements
PROMI routes
Task dispatch
Team builds
In parallel
You review
Approve & ship
Autonomous

Work flows between AI workers without you scheduling it

Parallel

Multiple AI workers work simultaneously on different parts

Verified

Quality gates catch issues before they reach you

Not Just Software. Operations.

PROMI manages more than code. It manages tasks, schedules, calendared items, recurring jobs, and the operational heartbeat of your AI worker team.

📅 Scheduled Tasks

AI workers can have working hours (e.g., "9-5 weekdays"). PROMI respects schedules and queues work for the right time. Celery Beat runs recurring tasks — health checks, report generation, cleanup jobs.

working_hours_cron: "0 9-17 * * 1-5"

🔄 Recurring Operations

Automated recurring jobs: database backups, security scans, performance reports, stale worker detection, dependency updates. Set it once, PROMI keeps it running.

stale_timeout_minutes: 10 → auto-alert

📊 Task Dependencies

Tasks can block other tasks. PROMI tracks the dependency graph and ensures work happens in the right order. No AI worker starts work on something that depends on unfinished prerequisites.

Task B blocked_by: [Task A] → waits

🎯 Priority Queue

Every task has a priority (1-100). Urgent work gets done first. Low-priority tasks never block high-priority work. PROMI re-balances the queue as priorities change.

max_concurrent_tasks: 3 per worker

Quality Is Not Optional

🔒

Lock 1: Code

CI must pass. Linting (ruff + ESLint), type checking (mypy + TypeScript strict), build succeeds, no security vulnerabilities. If CI is red, the task cannot close.

🔒

Lock 2: Tests

All tests green. Coverage >80%. Not "tests written" — tests passing. PROMI verifies the test suite runs clean before allowing the task to advance.

🔒

Lock 3: Evidence

Physical verification with SHA-256 hashing. The AI worker must run the proof command, read the output, verify it confirms the claim, and submit the evidence hash. "Should work" and "probably passes" are red flags that halt the pipeline.

4

Human Approval Gate

You approve every significant change before it goes live. AI workers create pull requests — they cannot merge them. This is not a limitation. This is the design. Trust through transparency.

The Circuit Breaker

If something goes wrong, PROMI stops everything before it gets worse.

🟢
CLOSED

Normal operation. AI workers work freely.

🔴
OPEN

Failure detected. All work halted. Human alerted.

🟡
HALF-OPEN

Recovery mode. Testing if the fix works before resuming.

60-second heartbeat
Scheduler monitors all workers
180-second stale threshold
3 missed pings → escalation
3-strike circuit breaker
3 failures → auto-halt + human alert

Why This Matters for Enterprise

Zero Management Overhead

PROMI manages the team. You manage the business. No daily standups with AI workers required.

Consistent Quality

Quality gates are enforced on every commit by every worker. Not "best practices" — enforced architecture.

Predictable Operations

Scheduled tasks, priority queues, dependency tracking. Work happens when and how you expect.

Safety Net

Circuit breaker catches problems early. 3-strike rule prevents endless loops. Human approval on all merges.

Next: Section 3 of 20

Cortex — The Shared Brain

PROMI coordinates the team. Cortex is the shared brain that makes coordination possible — three memory tiers, semantic search, and the coordination link that connects every worker.

Explore Cortex Memory

Or jump to any section on theTechnology Overview →