AI Agent Governance
Govern the agent loop,
not the prompt.
A server-enforced state machine for autonomous AI agents: illegal moves get a hard 403, actions are capped per beat, and every transition is logged. Monitoring watches; OpenWeave enforces.
Ask your AI assistant — it reads OpenWeave and explains it. Pick a question, then open or copy:
What is OpenWeave, and how do I use it?
ChatGPT opens prefilled — just hit enter. The rest copy the question to paste into a new chat.
We Govern the Agent Loop
Every AI agent runs in a loop.
An agent loopis the cycle an autonomous agent runs in — wake, read state, act, record, repeat — for hours or days at a time. That loop is where agents succeed or fail, and prompts can't govern it. OpenWeave is the control plane for the agent loop, in production across real projects. It gives every self-running loop the three primitives it can't provide on its own:
The Clock
A heartbeat drives the loop — agents wake, read what changed, and act within a capped blast radius. Continuous autonomy you can reason about.
The Rails
Tickets move through a server-enforced state machine. Illegal transitions are rejected with a hard error — the rules live on the server, not in the prompt.
The Memory
Epics carry a bot-maintained progress log across iterations. Feedback from one loop compounds into the next — and across every agent on the work.
“Others optimize the prompt. We govern the agent loop — and we've shipped it in production.”
The Problem
Monitoring tells you what your agents did. It can't stop what they shouldn't.
A loop runs away — one confused beat rewrites half your backlog while you sleep.
A loop skips steps — an agent jumps straight to "done" without doing the work.
A loop forgets — every iteration starts cold, losing what the last one learned.
Concurrent agents overwrite each other and leave state inconsistent.
The problem isn't the model's output — it's the loop's unconstrained execution.
Monitoring tells you after the fact. By then the loop has already acted.
Differentiation
Govern Actions. Not Prompts.
Others
- –Monitor model outputs
- –Scan prompts for risk
- –Track bias metrics
- –Provide observability dashboards
OpenWeave
- Enforces allowed state transitions
- Rejects illegal state changes at the API layer
- Protects terminal states from corruption
- Logs every execution event immutably
“Others observe. We enforce.”
How It Works
Server-Enforced AI State Machine
Define States
Configure allowed statuses, colors, and terminal states at workspace or project level.
Set Transitions
Define which transitions are legal for bots vs. humans. Backend rejects everything else.
Execute with Authority
Every state change is validated, logged atomically, and protected from concurrent corruption.
Architecture
Deterministic Agent Execution by Design
$ PATCH /api/tickets/SA-42/
> { "status": "COMPLETED" }
← 400 Bad Request
{ "status": "BOT cannot transition from IN_PROGRESS to COMPLETED. Allowed: BLOCKED, IN_TESTING, REVIEW, CANCELLED." }
The backend is the sole authority. Clients cannot mutate state directly. Bots must follow transition rules defined in the state machine. Terminal states cannot be corrupted. OpenWeave is a control plane for execution integrity.
Built For
Built for Multi-Agent Coordination
Organizations deploying internal AI agents
Engineering teams automating with LLMs
Regulated environments requiring audit trails
Companies needing deterministic agent coordination
Teams running multiple autonomous systems
Anyone who needs execution control, not just monitoring
New Category
The control plane for the AI agent loop.
Prompt engineering shaped the answer. Context engineering shaped the input. Loop governance controls what the agent loop is allowed to do.
Execution control infrastructure.
Put your agents under governance.
Define the states agents may enter, and let the server enforce every move. Get a token and start in minutes.