HolmesGPT + Dapr Workflow
HolmesGPT is an open-source agent for SRE and on-call work — alert triage, log analysis, and incident investigation across your Kubernetes and observability stack. Make your HolmesGPT investigations production-ready by running them as durable Dapr workflows on Catalyst, so a long-running investigation survives process crashes, deploys, and tool-call timeouts.
HolmesGPT is supported through the same diagridio/python-ai durable-execution package as the other Python frameworks. Like every agent framework on Catalyst, it requires a workflow state store; the Conversation API is optional.
Why Dapr Workflow for HolmesGPT?
| Challenge | Dapr Solution |
|---|---|
| Long investigations lost when a process crashes | Durable workflows checkpoint every step |
| Chat memory lost on restart | State store persists agent memory across restarts |
| Custom retry logic for flaky tool calls | Built-in retries with exponential backoff |
| No visibility into what the agent did | Built-in observability traces every tool call |
Get started
Install the HolmesGPT extra in its own environment. It ships strict dependency pins that conflict with the other agent extras, so it is intentionally not part of diagrid[all]:
pip install "diagrid[holmesgpt]"
Then follow the runnable HolmesGPT example in the python-ai repository, which wraps a HolmesGPT investigation in a Dapr workflow and runs it on Catalyst:
See also
- Develop AI agents — framework requirements table and the other supported frameworks
- AI agents — the durable execution and identity model agents run under
- Durable workflows — the underlying primitive agents are built on
