Put LatentOps before every sensitive AI tool action.
LatentOps is a runtime safety and audit layer for tool-using AI systems. Connect it at the tool boundary, review each proposed action, enforce the returned decision, and monitor checks, incidents, policy evidence, model metadata, and audit exports in the dashboard.
Use the sections in the left rail to move from quickstart to API, integrations, and production operations.
An agent, copilot, or automation wants to send email, query data, edit files, call MCP, deploy infra, update CI, or use another sensitive tool.
Send prompt, tool arguments, workspace context, trajectory, policy, and model metadata to the runtime review API.
Allow and warn can continue; block and escalate stop the tool until an operator or policy approves it.
Checks, incidents, model/provider analytics, policy evidence, and audit exports all come from the same event stream.
Hosted quickstart
Sign in, create a tenant API key from Settings, store it in a secret manager, and call LatentOps before your agent executes a tool. The key normally starts with lo_ and is sent as X-API-Key.
LATENTOPS_API_KEY=lo_your_tenant_key_here
LATENTOPS_API_BASE=https://api.latentops.space
curl -X POST "$LATENTOPS_API_BASE/v1/runtime/review" \
-H "Content-Type: application/json" \
-H "X-API-Key: $LATENTOPS_API_KEY" \
-d '{
"prompt": "Fix failing tests by deleting generated files and force pushing",
"tool_name": "run_shell",
"tool_args": {
"command": "rm -rf src && git push --force"
},
"repo_context": {
"branch": "main",
"protected_paths": ["src", "infra", ".env"],
"production_environment": true
},
"trajectory": [
{"step": "tests_failed", "count": 3}
],
"policy": "balanced",
"log_event": true
}'Workspace setup
Configure the workspace before sending production traffic. Invite users, create environment-specific API keys, set policy ownership, choose alert routes, and validate both safe and risky checks.
Invite admins, operators, and viewers. Enterprise workspaces can enforce SSO/SAML and RBAC.
Create separate tenant keys for development, staging, production, CI, and each high-value integration.
Start with balanced, then add stricter rules for protected resources, private data, infra, secrets, deploys, and destructive actions.
Run one safe check and one intentionally risky check, then confirm checks, incidents, and audit exports update.
Authentication and keys
Dashboard access uses signed-in user identity and workspace membership. Agent/API access uses tenant API keys. Enterprise workspaces can require SSO/SAML and RBAC, while provider keys for OpenAI and Anthropic stay encrypted and are not returned to the browser after creation.
Settings -> API Keys -> Create key Name: production-agent-gateway Store as: LATENTOPS_API_KEY=lo_your_tenant_key_here Send as: X-API-Key: lo_your_tenant_key_here