daai-console 0.1.0a2

Python SDK for DAAI Console cooperative action governance.

DAAI Console Python SDK

DAAI Console is a governance layer for AI automations. It lets developers intercept risky registered agent actions before execution, route them through policy and approval, and produce audit receipts.

This repository contains the public alpha Python SDK for DAAI Console.

What DAAI Console Is Not

DAAI Console is intentionally narrow in alpha. It is not:

• Magic interception of arbitrary code
• A browser automation framework
• A replacement for MCP
• An OS-wide agent scanner
• A full enterprise security platform

Execution Boundary

DAAI Console governs proposal, policy, approval, status, and receipts.

The developer application owns real business execution, local business logic, executor functions, external integrations, and worker or cron triggers. Place this SDK before risky functions. The SDK does not execute callbacks inside intercept().

Install

From PyPI:

pip install daai-console

Alpha pinning:

pip install daai-console==0.1.0a2

Environment

Set these values in your server-side environment:

export DAAI_API_KEY="..."
export DAAI_WORKSPACE_KEY="..."
export DAAI_BASE_URL="https://your-daai-api.example.com"

Do not expose these values in browser code or frontend bundles.

Minimal Client Example

import os

from daai_console import DaaiClient

client = DaaiClient(
    base_url=os.environ["DAAI_BASE_URL"],
    api_key=os.environ["DAAI_API_KEY"],
    workspace_key=os.environ["DAAI_WORKSPACE_KEY"],
)

payload = {
    "invoice_id": "INV-1025",
    "customer_name": "Acme Finance",
    "amount": 1250,
}

proposal = client.intercept(
    action="send_invoice_reminder",
    payload=payload,
    idempotency_key="invoice-reminder:INV-1025",
)

print(proposal.governance_status.value)
print(proposal.executable)

status = client.status(proposal.action_run_id)

if status.executable:
    try:
        # Your app owns the real business execution.
        provider_result = send_invoice_reminder(payload)
        client.report_executed(
            proposal.action_run_id,
            execution_result={"provider_id": provider_result["id"]},
        )
    except Exception as exc:
        client.report_failed(
            proposal.action_run_id,
            execution_error=f"{type(exc).__name__}: {exc}",
        )
else:
    print("Not executable yet. Wait for approval or policy decision.")

Runtime Helper Example

The runtime helpers make the safe path easier: propose once, persist pending approvals locally, and run only after DAAI Console reports executable=true.

import os

from daai_console import (
    DaaiActionRunner,
    DaaiClient,
    PendingActionManager,
    SQLitePendingStore,
)

client = DaaiClient(
    base_url=os.environ["DAAI_BASE_URL"],
    api_key=os.environ["DAAI_API_KEY"],
    workspace_key=os.environ["DAAI_WORKSPACE_KEY"],
)
store = SQLitePendingStore("daai_pending_actions.sqlite3")
manager = PendingActionManager(client=client, pending_store=store)

manager.propose(
    action="send_invoice_reminder",
    payload={"invoice_id": "INV-1025", "amount": 1250},
    idempotency_key="invoice-reminder:INV-1025",
)


def send_invoice_reminder_executor(payload: dict) -> dict:
    # Keep your existing execution function here.
    return {"message_id": "example-local-result"}


runner = DaaiActionRunner(client=client, pending_store=store)
runner.when_executable(
    action="send_invoice_reminder",
    run=send_invoice_reminder_executor,
)

executed_count = runner.run_pending_once()
print(f"Executed {executed_count} approved action(s).")

Live Alpha Smoke Tests

The unit tests in tests/ are mocked SDK-internal tests. They do not call DAAI Console and should remain safe for normal local runs and CI.

The scripts in examples/live_*.py are opt-in live smoke tests for alpha testers with real DAAI Console credentials. They call the configured API and should be run manually against a staging or alpha workspace.

Before running live scripts, make sure the action is already registered in your DAAI Console workspace. The default expected action is send_invoice_reminder. You can override it with DAAI_TEST_ACTION_NAME.

Set required environment variables:

export DAAI_API_KEY="..."
export DAAI_WORKSPACE_KEY="..."
export DAAI_BASE_URL="https://stage.api.daaihq.com"

DAAI_BASE_URL must point to the API, not the dashboard. For staging, use https://stage.api.daaihq.com, not https://stage.daaihq.com.

Optional environment variables:

export DAAI_TEST_ACTION_NAME="send_invoice_reminder"
export DAAI_PENDING_DB_PATH="./daai_alpha_pending.db"

Run the low-level client smoke test:

python examples/live_client_smoke.py

Run the runtime smoke flow:

python examples/live_runtime_smoke.py propose
python examples/live_runtime_smoke.py list-pending

If the proposal returns pending_approval, approve it from the approval email or dashboard. The SDK does not auto-approve and does not bypass governance. After approval, run:

python examples/live_runtime_smoke.py run-pending

run-pending uses DaaiActionRunner and a fake executor. It only simulates execution, and only runs when DAAI Console reports executable=true.

If the API returns blocked or unknown_action, it usually means the action is not registered in the workspace, the API/workspace key belongs to another workspace, or the action name does not match exactly. Register the action in the dashboard first, then rerun the script with the same action name.

Security Notes

• Store DAAI_API_KEY and DAAI_WORKSPACE_KEY server-side only.
• Do not expose keys in browser or frontend code.
• Approval links should not contain sensitive payload data.
• Local pending stores may contain business payloads; treat them as sensitive.
• The SDK does not execute callbacks inside intercept().
• Rejected and blocked actions should not execute.

Known Limitations

• Python SDK first.
• Cooperative interception only.
• Developers must register and gate actions explicitly.
• No automatic arbitrary-code interception.
• Staging alpha APIs may change.
• No enterprise RBAC in alpha.

Alpha Docs

• Alpha test checklist
• Security model
• Agency adoption prompt
• Known limitations
• Staging alpha guide