Diagrid Catalyst 2.0 adds durable execution to 10+ agent frameworks

Diagrid's Catalyst 2.0 addresses a structural gap in the current AI agent ecosystem. Over the past two years, major frameworks—LangGraph, Microsoft Agent Framework, Google Agent Development Kit, and others—have made it significantly easier to build agents, but as co-founder and CEO Mark Fussell noted, they have not made it easier to trust them in production. Most offer only basic checkpointing for recovery, leaving detection and recovery logic to developers, which is both labor-intensive and error-prone.

The release pairs two complementary capabilities: durable execution (agents resume from the exact point of failure) and verifiable execution (each step is cryptographically signed and traced). The verification component draws from Dapr 1.18, released in June, and includes cryptographic history signing, execution lineage propagation, and workflow attestation. This is particularly significant for regulated sectors—healthcare, finance, legal—where compliance teams require a chain of custody for each decision and must prove that execution was complete, traceable, and unaltered. Diagrid's framing positions this as a shift from the first wave of AI (making models intelligent) to making agent systems trustworthy.

The practical leverage is high: because Catalyst 2.0 works as a code package layered on top of existing frameworks, teams avoid costly rewrites and adopt durability and verifiability incrementally. Early adoption by ZEISS Group, a German optics manufacturer, suggests that enterprises see value in a stable foundation that accommodates the rapid iteration of AI frameworks and models.