1 pointby prateekdalal5 hours ago1 comment
  • prateekdalal5 hours ago
    We’ve released v1.0.2 and v1.0.3 of Kakveda (open source AI governance runtime).

    This release shifts integration from manual middleware to an SDK-first model.

    Key change: Instead of calling /warn and /publish manually, agents now integrate via: from kakveda_sdk import KakvedaAgent

    agent = KakvedaAgent()

    agent.execute( prompt="delete user records", tool_name="db_admin", execute_fn=real_function )

    The SDK handles:

    Pre-flight policy checks

    Event publishing

    Trace ingestion

    Dashboard registration

    Heartbeat monitoring

    Fail-closed behavior

    Circuit breaker logic

    Legacy manual helpers were removed to reduce integration friction.

    The goal is to treat LLMs as suggestion engines while keeping execution inside a deterministic governance layer.

    Would appreciate feedback from folks running multi-agent systems in production.