Archived: 06b AI SRE with hosted LibreChat
Archived reference for the former optional hosted LibreChat module.
Archived module
Module 06b is no longer part of the workshop. Do not provision LibreChat for the current learner path. Continue directly from 06 AI SRE to 07 Test, fail, and fix. The material below remains available only as a historical reference.
Historical starting point
When this module was active, learners stayed on build-workshop-v1 and budgeted about
15 minutes.
Prerequisites: Module 05 is complete, your instructor has provided an HTTPS LibreChat
URL and account, and you have a fine-grained GitHub token with read-only access to
ClickHouse/ClickHouse_Demos. Self-paced learners can deploy LibreChat with its official
Railway one-click cloud guide before
starting this module.
Cloud-hosted only
Do not run LibreChat, MongoDB, or another database on your workshop machine. LibreChat
and its persistence run in the hosted instance. Its OAuth callbacks must use that
instance's HTTPS URL, never localhost.
Why
Module 06 put the agent in your editor. Here a chat-native SRE agent investigates two remote planes at once: ClickStack telemetry and the public workshop source on GitHub.
Step 1 — Sign in to hosted LibreChat
Open the HTTPS URL supplied by your instructor and sign in with your assigned account. Self-paced learners use the public Railway URL created during cloud deployment.
The page address must begin with https://. If it points to localhost, stop and use
the hosted URL.
Step 2 — Connect the remote MCP servers
Open MCP Settings. Your instructor may already have shared these server definitions; otherwise add them in the UI:
| Name | Type | URL | Authentication |
|---|---|---|---|
clickstack | Streamable HTTP | https://mcp.clickhouse.cloud/clickstack | Browser OAuth |
github | Streamable HTTP | https://api.githubcopilot.com/mcp/ | Your read-only PAT |
For clickstack, first copy the UUID from ClickStack Team Settings → Data →
Connections → Default → Service ID into LibreChat's per-user
CLICKSTACK_SERVICE_ID field. Then choose Connect and complete ClickHouse Cloud
OAuth. The callback returns to the hosted LibreChat HTTPS URL.
For github, provide your fine-grained PAT only in LibreChat's per-user credential
dialog. Restrict it to ClickHouse/ClickHouse_Demos with Contents: read. Enable the
read-only/repository toolset if the server form exposes those controls.
Both servers must show Connected before continuing.
Step 3 — Create the SRE agent
Open Agent Builder and create an agent:
- Name it
NYC Taxi Ops SRE. - Paste the Instructions block from
workshops/build_workshop/app/librechat/sre-agent-definition.md. - Select the strong tool-capable model provided by the hosted instance.
- Add the
clickstackandgithubMCP tools. - Save the agent and ask
What can you investigate?
The answer should mention ClickStack telemetry and the
ClickHouse/ClickHouse_Demos/workshops/build_workshop/app source.
Step 4 — Break something, then investigate
Inject the map fault and rebuild only the local workshop app:
cd "$(git rev-parse --show-toplevel)/workshops/build_workshop/app"
git checkout fault/01-map-not-loading
docker compose --env-file .env.workshop \
-f docker-compose.workshop.yml -f docker-compose.otel.yml up -d --buildOpen the app so the browser SDK records the fault. In hosted LibreChat, ask:
Users report UI errors. Use ClickStack telemetry and the GitHub source to find the root cause. Is this a code bug?The agent should connect the nyc-taxi-frontend error to
frontend/src/ui/ZoneMap.tsx and show the evidence it used.
Completion check
- LibreChat is reached through an instructor-provided or self-paced HTTPS cloud URL.
- The agent has connected remote
clickstackandgithubMCP tools. - Its diagnosis cites both telemetry and a source file on the selected fault branch.
The active workshop continues at 07 Test, fail, and fix.