Diagrid

Ensures AI agents automatically recover from crashes, outages, and restarts without losing state or progress, guaranteeing workflows are always carried to completion from the last completed step.

Signs every step of an agent's execution into a tamper-proof record, providing provenance (where actions originated), attestation (who executed each step), and a cryptographic audit log to prove exactly what an autonomous agent did.

Supports integration with popular AI agent frameworks such as LangGraph, CrewAI, Google ADK, Strands, OpenAI Agents, Microsoft Agent Framework, and Spring AI, allowing developers to make their existing agents durable and verifiable without code rewrites.

Enforces security at the execution layer by allowing explicit control over access between agents, applications, and MCP servers, with a 'deny by default' policy to govern connections.

Built by the creators of Dapr and KEDA, two graduated CNCF projects, ensuring a multi-cloud and lock-in-free architecture based on widely adopted open-source software.

Provides capabilities for end-to-end distributed tracing and tools to observe failed agent runs, workflow retry storms, and inspect workflow steps.