AI SRE Agent with Temporal, ClickHouse, and Codex: AURA
1,133 requests failed on one bad commit. The patch and the regression test are already written by the time anyone is asked to read the exact diff.
- 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 demo runs AURA as one step inside a Temporal workflow, alongside Codex. A GET request against a product catalog service goes from success to HTTP 500, and ClickHouse records the version, commit, trace ID, and exact error for every request. By the time AURA investigates, all 1,133 requests on that version have failed.
AURA reaches ClickHouse through MCP, and in this demo those tools are read-only. It compares the healthy and failing versions, traces one error signature to price formatting in the product handler, and returns the affected commit, supporting query results, a reproduction, and the exact query the workflow should use after deployment. It cannot change the service.
Temporal passes that investigation to Codex. Codex works in an isolated Git work tree, a separate copy of the repository, fixes the formatter, adds a regression test, and runs the test suite. Four tests pass. Temporal then stops. Nothing can deploy until a person reviews the exact diff and approves the rollout.
Approval resumes the same workflow. Temporal deploys the reviewed patch, then AURA checks only the new ClickHouse events and finds 12 successful requests, no failures, and no return of the original error. Temporal preserves the full workflow history.
0:00 How AURA, Codex, and Temporal fit together
0:44 A product catalog regression in ClickHouse
1:27 Codex patches and tests in an isolated work tree
1:48 Approval, deploy, and verification
#AURA #Temporal #AISRE
Transcript
How AURA, Codex, and Temporal fit together
0:00 This demo shows how you can run AURA in a Temporal workflow alongside Codex. You'll see AURA investigate a failing catalog service with ClickHouse evidence, Codex prepare and test the patch, and Temporal pause for human approval before deployment. After the release, AURA checks the live service to verify the fix.
The catalog service writes each request to ClickHouse, a database built for fast analytics. AURA is an agent for site reliability work. It reaches ClickHouse through MCP, a standard way to give an agent tools. In this demo, those tools are read-only. Codex works on the code. Temporal coordinates each handoff and keeps the workflow state.
A product catalog regression in ClickHouse
0:44 The run begins with a real product catalog regression. A GET request for a product goes from success to HTTP 500. ClickHouse records the version, commit, trace ID, and exact error. By the time AURA investigates, all 1,133 requests on that version have failed.
AURA can read the evidence, but it cannot change the service. It compares the healthy and failing versions. One error signature points to price formatting in the product handler. Its response includes the affected commit, supporting query results, a reproduction, and the exact query the workflow should use after deployment.
Codex patches and tests in an isolated work tree
1:27 Temporal passes that investigation to Codex. Codex works in an isolated git worktree, which is a separate copy of the repository. It fixes the formatter, adds a regression test, and runs the test suite. Four tests pass. Temporal then stops. Nothing can deploy until a person reviews this exact diff and approves the rollout.
Approval, deploy, and verification
1:48 Approval resumes the same workflow. Temporal deploys the reviewed patch, then AURA checks only the new ClickHouse events. It finds 12 successful requests, no failures, and no return of the original error.
AURA found the cause in ClickHouse. Codex prepared and tested the fix. A person approved the exact change before deployment. AURA then verified the recovery against live evidence. Temporal preserves the full workflow history.
More videos
