Diagrid Catalyst 2.0 brings durable execution to AI agent frameworks

Diagrid released Diagrid Catalyst 2.0, an update to its managed workflow engine that adds durable execution and cryptographic verification for AI agents built with more than 10 frameworks. The launch targets a practical gap in enterprise agent deployments: long-running workflows can fail midstream, forcing teams to rerun work, spend additional tokens or build their own recovery logic.

The company said developers can use Catalyst 2.0 by adding a Diagrid package to agent frameworks they already use, rather than rewriting the agents. Supported frameworks include LangGraph, LangGraph Deep Agents, Microsoft Agent Framework, Google Agent Development Kit, AWS Strands, OpenAI Agents SDK, Claude Managed Agents, CrewAI, Pydantic AI Agents and Dapr Agents.

What is Diagrid Catalyst 2.0?

Catalyst 2.0 is Diagrid’s managed workflow engine for running agent workflows with recovery and verification controls. In Diagrid’s framing, durable execution means a workflow can restart from the point where it failed, while verifiable execution means each step can be signed and traced to its origin.

Diagrid used LangGraph as an example of where it sees the product fitting. The company said LangGraph offers orchestration capabilities but leaves much of the failure detection and recovery work to developers through basic checkpointing. Diagrid claims Catalyst 2.0 can run those workflows through failures without requiring custom recovery code.

The verification features are tied to Dapr 1.18, which was released in June. Diagrid said those capabilities include signed execution history, lineage propagation and workflow attestation. The intended buyers are security and compliance teams that need evidence that an agent completed a process, followed a traceable sequence of actions and was not altered after the fact.

That positioning is narrower, and more concrete, than the usual agent infrastructure pitch. The product is less about making agents smarter and more about making their execution auditable enough for production systems in regulated environments. Diagrid did not disclose pricing, revenue, customer count or headcount for the Catalyst 2.0 release.

Catalyst 2.0 can run in cloud environments, on-premises deployments and fully air-gapped settings, according to Diagrid. The company also claims the product can deliver up to 10 times the performance of open-source Dapr and support millions of concurrent agent workflows. Those are vendor claims; Diagrid did not provide independent benchmark details in the announcement.

Diagrid Chief Executive Mark Fussell said the recent wave of agent frameworks has made agent development easier, while leaving production trust and durability unresolved. Chief Technology Officer Yaron Schneider, who chairs the Agentic AI Foundation’s workflows working group, described the release as part of a shift from improving model intelligence to improving the reliability of agent systems.

ZEISS Group is using Catalyst early. Wendelin Niesl, head of end-to-end core application engineering at the German optics manufacturer, said the product gives ZEISS a stable base for durable AI and traditional workloads while models and frameworks continue to change.

Diagrid was founded in 2021 by the creators of Dapr and KEDA and remains a leading maintainer of Dapr within the Cloud Native Computing Foundation. The company has raised $24.2 million in venture funding, with most of that coming from a 2022 Series A led by Norwest Venture Partners.

This story draws on original reporting from SiliconANGLE.