Skip to content
LatentOpsRequest accessContact

Runtime Control.Before the next step.

Review AI actions with your team's policies. Bring in a person when it matters, and keep a record of the decision.

Request access Read the docs

Runtime Control is currently a platform preview.

Proposed action

Publish this update to production.

deploy.release
Context
Production workspace
Policy
A person must approve production changes.
Needs approval

The runtime pauses the action and asks the release owner to review it.

The decision stays with the action.

An illustrative workflow. No tools are executed here.

A pause at
the right moment.

An agent can propose an action in seconds. Put the review between that proposal and the tool that changes something.

  1. 01

    The agent proposes.

    Send the action, tool arguments and relevant context for review.

  2. 02

    Your policy decides.

    Return a decision to continue, warn, block or request a person's approval.

  3. 03

    Your runtime responds.

    Enforce the decision before execution, and keep the review in your workspace.

A shared view.
A clear decision.

When more agents and people get involved, a local check is only part of the picture. Runtime Control gives the team somewhere to review the work together.

Open the workspace

Set the rules for your workspace.

Decide which tools, environments and resources need extra checks. Use the same policy across the agent workflows you connect.

Bring the right person into the decision.

Route sensitive actions for approval. Give the reviewer the proposed action and its context, so they can make an informed call.

Keep the reasons with the result.

Review past checks, approvals and incidents in one workspace. Follow what was proposed, what was decided and why.

Keep execution in your application.

Your runtime enforces the returned decision before calling the tool. Keep permissions, credentials and sandboxing in place.

Start with
one tool call.

Add a review at the point where your agent is about to use a tool. Connect a workspace API key, send the proposed action, then handle the returned decision.

Explore the integrations
bash
curl -X POST https://api.latentops.space/v1/runtime/review \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $LATENTOPS_API_KEY" \
  -d '{
    "prompt": "Check what changed in this project.",
    "tool_name": "run_shell",
    "tool_args": {"command": "git status --short"},
    "policy": "balanced",
    "log_event": true
  }'

This request reviews a command. Your application decides whether to execute it.

The work behind
the product.

Our DARC research asks how an action's context should affect the decision to let it proceed. Read the preliminary evaluation, including its test conditions and limitations.

Read the research

What should your
agent check first?

Tell us about one workflow you'd like to bring into review.

Talk to LatentOps

Need a local command-review tool? Meet Ordin