AI AGENT ACTION AUDIT LAYER
Add action logs, approval gates, execution status, and client-ready receipts to your AI automations with a simple Python SDK.
Registered actions only. Deterministic policies. Approval gates where needed.
Proposed action
run_8c4fInput summary
Follow up on overdue invoice INV-1025 for ABC Pty Ltd before a finance workflow sends email.
Policy decision
Policy gatePolicy
Require approval above $500
Decision
Human approval required
Approver
client@example.com
Action logged
Policy evaluated
Approval requested
Awaiting execution
DAAI gives consultants and clients a shared record of what the agent proposed, what policy decided, what humans approved, and what actually executed.
Define the action your agent may propose, such as sending an invoice reminder or marking an invoice paid.
Place DAAI before the real business function so the action can be logged, evaluated, and controlled.
DAAI creates an action run with input details, source context, reasoning, policy decision, and status.
Low-risk actions can be logged automatically. Risky actions can require approval before execution.
After execution, report the result back to DAAI and keep a clear receipt for client review.
Built for consultants and agencies who need action visibility, execution status, approval gates where needed, and receipts clients can review.
Capture every proposed agent action with actor, payload summary, reasoning, policy decision, and status.
Require human approval for sensitive or high-risk actions before the connected automation executes them.
IF
amount > $500
THEN
require approval
Coming soon: shareable client access links for action history and receipts.
Client view
Action history and receipts
Coming soon: shareable client access links for action history and receipts.
Add DAAI before risky functions in your existing Python automation without rebuilding the whole workflow.
ticket = manager.propose(
action="send_invoice_reminder",
payload=payload,
)
if ticket.executable:
send_invoice_reminder(payload)Generate clear governance receipts that explain what happened and why.
Track whether an action was allowed, blocked, approved, rejected, executed, or failed.
allowed
logged by DAAI
approved
ready for worker
failed
reported by app
Track what the agent proposed, what policy decided, what needed approval, what was blocked, and what actually executed.
Workspace
24
action runs
7
pending approval
14
executed
3
blocked
| Action | Actor | Policy | Status | Execution | Time |
|---|---|---|---|---|---|
| send_invoice_reminder | finance_agent | approval > $500 | Pending approval | Not executed | 2m ago |
| update_invoice_status | finance_agent | always allow | Allowed | Executed | 11m ago |
| mark_invoice_paid | finance_agent | always approval | Approved | Awaiting report | 18m ago |
| unknown_action | agent | not registered | Blocked | Not executed | 32m ago |
Python SDK
Place the SDK before risky functions so the proposal is logged, evaluated, and controlled before your automation performs the real action.
1import os2from daai import DaaiClient, PendingActionManager, SQLitePendingStore3 4client = DaaiClient(5 api_key=os.environ["DAAI_API_KEY"],6 workspace_key=os.environ["DAAI_WORKSPACE_KEY"],7 base_url=os.environ.get("DAAI_BASE_URL", "https://beta.api.daaihq.com"),8)9 10store = SQLitePendingStore("daai_pending.db")11manager = PendingActionManager(client=client, pending_store=store)12 13ticket = manager.propose(14 action="send_invoice_reminder",15 payload={16 "actor": "finance_agent",17 "invoice_id": "INV-1025",18 "customer_name": "ABC Pty Ltd",19 "amount": 1250,20 "reasoning": "Invoice is overdue and needs a polite follow-up.",21 },22)23 24if ticket.executable:25 send_invoice_reminder(ticket.payload)DAAI logs registered proposals first. Approval gates are applied only where policy says a human decision is needed.
Registered action and payload
Action run records context and reasoning
Allowed, pending approval, or blocked
Human decision before connected automation executes
Decision and execution result recorded
Start with the limits enforced by the backend today. Paid tiers will be available after a successful beta test.
Paid tiers will be available after a successful beta test.
Beta access is intentionally limited during launch to protect system reliability.
$0/month
For proving the audit, approval, execution, and receipt path with real clients.
These limits are enforced in the backend during beta.
After beta
For consultants auditing and controlling a small set of client automations after launch validation.
Paid tiers will be available after a successful beta test.
After beta
Built for small agencies standardizing action visibility and controls after the beta proves reliability.
Paid tiers will be available after a successful beta test.
Leave an email and a short note. We will follow up when paid tiers are ready after beta.
Start with one client workspace. Register one risky action. Log the proposal, control execution when needed, and keep the receipt.