AI SRE Agent ChatOps: How AURA Runs in a Slack Incident Channel

Mezmo

Aug 28, 2026

AURA runs as a standalone service and knows nothing about Slack. A separate bridge streams events off its API, which is what makes the chat tool swappable.

Reference code for "aura-bridge" service: https://github.com/mezmo/aura-slack-pd-bridge-example

Learn more about AURA → https://www.mezmo.com/aura
Get started today → https://github.com/mezmo/aura
Get deployment help → https://www.mezmo.com/contact

This is the implementation behind the automated PagerDuty investigation demo. Jeff Vogt walks the whole path an incident takes: a Grafana alert fires PagerDuty, PagerDuty opens a Slack channel and sends a webhook, and a service called AURA Bridge picks it up and calls AURA's API with an investigation prompt.

On the AURA side, the agent runs in the cluster from the Helm chart and its entire configuration lives in the chart values, so merging a change to that file is how the config gets updated. The Kubernetes and Grafana MCP servers are cluster-local and nothing leaves the cluster for them; GitHub and PagerDuty use hosted MCP servers with API tokens. Human-in-the-loop is configured against a list match of tool names, and anything matching waits for approval. Two worker prompts get a closer look: the incident responder resolves an alert down to the individual Grafana rule and reports the data source behind it without interpreting the query, and the metrics analyst enumerates metric and label names before concluding a metric is absent, which separates absent from zero.

The bridge itself runs in three phases. Phase one is fully automated, with status updates from the workers posting to Slack while the investigation runs. Phase two handles follow-up questions, either by mentioning AURA in the channel or replying in a thread, resuming the same investigation. Phase three handles approval requests, posting approve and deny buttons and routing the click back through a permission gate.

The bridge is an example project, not a product. It was built with Claude in about an hour and is on GitHub for reference. It is completely decoupled from AURA, which runs as a standalone service exposing an OpenAI-compatible completions endpoint on the cluster network, and the video proves the decoupling by skipping the bridge and using the AURA CLI directly inside the pod. Exposing that API to a wider SRE team, and the header forwarding that would let team members supply their own tokens, are described as available directions rather than shown. Temporal is named as somewhere to migrate if the bridge ever needed to be mission critical, and is not used here.

This video shows configuration and code rather than an investigation. AURA diagnoses nothing and changes nothing on camera.

For more detail, visit
https://www.mezmo.com/videos/ai-sre-agent-behind-the-scenes-wiring-aura-into-slack

0:00 The alert-to-Slack path

0:37 AURA in the cluster, config in Helm values

0:58 MCP servers and the human-in-the-loop tool list

1:29 Tuning individual worker prompts

2:24 The bridge, phase 1: automated investigation

3:10 Phase 2: follow-up questions in the channel

3:26 Phase 3: approvals in Slack

4:03 What the bridge actually is

4:37 Skipping the bridge with AURA's API and CLI

5:33 Inside the bridge source