Agent governance

Human-in-the-loop approvals + memory + activity log for AI agents.

Overview

Agents in Atlas request human sign-off before taking consequential actions. AgentApproval gates those calls; AgentMemory persists key facts across runs; the activity log exposes every recent AgentRun for audit. Everything is deterministic - approvals are a state machine, memory is explicit key/value.

Highlights

Approval gate: PENDING APPROVED | REJECTED with optional note
Per-user / per-agent / per-tenant memory scopes
Double-decide guarded - a decided approval cannot be re-decided
Activity log filtered by agent + limit

Important to know

Memory is idempotent on (tenant, scope, key, agent) - upsert replaces.
Approvals survive service restart because they're DB-backed rows, not in-memory queues.

How to use it

Agent emits approval request appears in /settings/agents.
Reviewer approves or rejects with an optional note.
Agent resumes on approval; nothing happens on reject.

Atlas docs · Atlas - the AI-native work platform