Open-source portfolioSany Alam / Independent project / 2026

DAAI Console

A cooperative governance layer for registered Python agent actions.

Before an automation takes a consequential action, ask for permission. Evaluate a deterministic policy, pause for human approval, and record what happened. The developer's application owns the actual execution.

View source Read the engineering case study

The original hosted beta is discontinued. This page showcases the implementation; it is not a live governance service.

A proposal becomes an action story.

Verified local demo
Verified demo output: an allowed action executes locally, a pending action waits, approval permits execution, rejected and blocked actions do not run, and reused or invalid tokens are refused.
Actual SDK and FastAPI logic. Synthetic invoice data, in-memory persistence, and a simulated executor. No email, payment, or external business action.
  1. 01

    Propose

    The Python SDK submits a registered action and payload before execution.

  2. 02

    Govern

    A deterministic rule allows, blocks, or requests an expiring human decision.

  3. 03

    Execute locally

    The application's executor runs only when the API explicitly permits it.

  4. 04

    Report & record

    Execution results and the governance receipt complete the action story.

Small boundary. Deliberate choices.

Implementation
Deterministic runtime policy
Four explicit policy types. No LLM calls decide whether a runtime action may execute.
Governance and execution are separate
An approval authorizes execution; it does not claim success. Both states remain visible.
Explicit trust boundary
Hashed keys and expiring decision tokens protect access. Applications must cooperate with the gate.
Replay with payload identity
Workspace-scoped idempotency returns the existing run and refuses conflicting payloads.

Built with

Python · FastAPI · PostgreSQL · Supabase Auth · Next.js · Tailwind

DAAI does not intercept arbitrary code or enforce an OS-level sandbox. External side effects and exactly-once execution remain the developer's responsibility. The full dashboard needs a separate Supabase setup; the isolated demo is the verified starting point.

Inspect the work.

Source, schema migrations, regression tests, and a reproducible local walkthrough. MIT-licensed project source, with the original deployment history kept private.

Console source Run the local demo Companion Python SDK MIT license