Diagrid Catalyst 2.0 ships durable execution for AI agents
Diagrid released Catalyst 2.0 on 28 July 2026, a developer infrastructure tool that adds durable execution and verifiable execution features to AI agent frameworks InfoQ. The release targets a concrete operational problem: agents that fail mid-task today lose progress and re-execute completed steps, wasting tokens and time. Catalyst 2.0 eliminates that rework through persistent checkpointing of agent state and cryptographic signing of execution history.
Durable execution cuts token waste and recovery friction
The core feature is durable execution — the ability for an agent framework to pause, persist workflow state to external storage, and resume from the last completed step without re-running succeeded tasks daily.dev. This is particularly valuable for long-running orchestrations, multi-step API chains, and agentic workflows where LLM calls are expensive. Diagrid's own materials frame durable execution as a cost and reliability layer that sits *below* the agent framework itself, meaning it works across LangGraph, Microsoft Agent Framework, Google ADK, and Dapr Agents without requiring framework rewrites.
Cryptographic signing and execution lineage for audit trails
Catalyst 2.0 adds a second layer: verifiable execution. When workflow history signing is enabled for a project, the system cryptographically signs blocks of events for each workflow's execution history as it runs, according to Diagrid's official documentation. This creates a tamper-evident audit trail — each step is timestamped, sequenced, and cryptographically bound to previous steps. The feature is aimed at enterprise deployments where compliance teams demand proof of what an agent actually did, in what order, without gaps or alterations.
Execution lineage tracking surfaces which agent steps called which APIs, which calls succeeded or failed, and what data flowed between them. This addresses a real operational blind spot: today, most agent frameworks log outputs but lose the causal chain — why a decision was made, which LLM prompt produced which action, what state the agent was in at each branch point.
Concrete adoption signal: Uniphar deployment
One documented early adoption came from Uniphar, which deployed Catalyst for workflow orchestration and observability, though specific metrics or timeline were not disclosed TechTarget. This represents the kind of enterprise-grade use case Diagrid is targeting — complex, multi-step workflows where failure recovery and audit trails are non-negotiable.
Developer tool positioning
Catalyst 2.0 is a developer tool, not an end-user agent or AI model. It is infrastructure for building and operating agents at scale. The release is open to enterprises and framework teams integrating Diagrid's APIs into their own stacks. This positions Diagrid as a middleware layer in the emerging AI agent stack — the plumbing that makes frameworks durable and verifiable, regardless of which LLM or orchestration framework sits above it.