Audit Log

Track every action your AI agent takes.

The audit log records every tool call that passes through Latch, along with the decision made and why.

What Gets Logged

Every tool call is logged with:

Field	Description
Timestamp	When the request occurred
Upstream	Which MCP server was called
Tool	Which tool was invoked
Action Class	How Latch classified the action (READ, WRITE, EXECUTE, etc.)
Decision	What happened (Allowed, Denied, Approval Required, Approved)
Rule	Which rule triggered the decision
Arguments	The tool call arguments (with sensitive values redacted)

Viewing the Audit Log

1. Go to Audit Log in the dashboard
2. Browse recent requests or filter by:
   • Upstream
   • Tool
   • Decision type
   • Date range

Understanding Decisions

In the audit log, each request has one of these decisions:

Decision	Meaning
Allowed	Action proceeded automatically
Denied	Action was blocked
Approval Required	Action is waiting for human approval

Approvals themselves have a separate status (Pending/Approved/Denied/Expired) in the Approvals view.

Automatic Redaction

Latch automatically redacts sensitive values in logged arguments:

• Passwords and secrets → [REDACTED]
• API keys and tokens → [REDACTED]
• Long content (>500 chars) → [TRUNCATED]
• Approval tokens → removed entirely

This keeps your audit log useful for debugging without exposing secrets.

Redacted Fields

Fields matching these patterns are redacted:

• password, passwd, secret
• token, api_key, apiKey
• credential, auth, bearer
• private_key, privateKey

Using the Audit Log

Debugging Agent Behavior

When your agent does something unexpected:

1. Find the request in the audit log
2. Check which tool was called and with what arguments
3. See which rule triggered the decision
4. Adjust rules if needed

Security Review

Periodically review the audit log to:

• Spot unusual patterns (many denied requests, unexpected tools)
• Verify rules are working as intended
• Identify tools that might need stricter policies

Compliance

The audit log provides a complete record for:

• Security audits
• Incident investigation
• Understanding AI agent behavior over time

Filtering and Search

Use filters to find specific requests:

Upstream: github
Decision: Denied
Tool: create_pull_request
Date: Last 7 days

Retention

Audit logs are stored in your Postgres database. Configure retention based on your needs:

• Default: All logs retained indefinitely
• For high-volume deployments, consider periodic cleanup

API Access

You can query the audit log programmatically:

curl -H "Authorization: Bearer latch_..." \
  "https://your-latch/api/audit?limit=100"

See the API Reference for details.