top | item 47047509 Kakveda v1.0.3: SDK-first governance layer for AI agents 1 points| prateekdalal | 13 days ago |github.com 1 comment order hn newest prateekdalal|13 days 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 KakvedaAgentagent = KakvedaAgent()agent.execute( prompt="delete user records", tool_name="db_admin", execute_fn=real_function )The SDK handles:Pre-flight policy checksEvent publishingTrace ingestionDashboard registrationHeartbeat monitoringFail-closed behaviorCircuit breaker logicLegacy 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.
prateekdalal|13 days 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 KakvedaAgentagent = KakvedaAgent()agent.execute( prompt="delete user records", tool_name="db_admin", execute_fn=real_function )The SDK handles:Pre-flight policy checksEvent publishingTrace ingestionDashboard registrationHeartbeat monitoringFail-closed behaviorCircuit breaker logicLegacy 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.
prateekdalal|13 days ago
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.