How-to guides¶
For engineers who know what they want done: each guide is one task, with the steps to finish it on your own cluster. The guides assume you know Kubernetes and Helm, and link to explanation pages for the why.
Install and connect¶
- Install with Helm: the chart, the Secrets and a first answer in a thread.
- Configure Slack: the Slack app, its tokens and the ChatConnection.
- Create the Binding identity: choose a user ID or issue and store a bearer token.
- Write a Binding: your agent in your channel.
Agents¶
- Connect a kagent agent: kagent 1.0 or 0.9.x.
- Connect a generic A2A agent: any agent that streams A2A over JSON-RPC.
- Connect an agent on Google AX: an A2A agent running in an AX Task.
- Let the agent ask questions: answers from the thread.
Alerts¶
- Investigate Alertmanager alerts: an investigation in each alert's thread, with a trigger written in full.
- Start investigations from any bot's messages: a trigger written in full for another tool's bot.
- Tune triggers with a dry run: replay a channel's history before going live.
Recipes¶
- Start from a recipe: a worked example with its Agent Kourier resources, its checks and, for one of
them, its agent. Two recipes:
sre-triage, Alertmanager alerts to a read-only kagent agent, andcustom-bot-trigger, a CI bot's failure posts to an agent of your own, with a hand-written trigger.
Security¶
- Put a verifying front door before kagent: verify the Binding's token and strip identity headers.
- Rotate the Binding token: by hand or on a schedule.
Operate¶
- Run two replicas (high availability): Postgres, two replicas on separate nodes, a budget for drains, and a failover drill.
- Export metrics and traces: Prometheus, OpenTelemetry and alerts worth having.
- Upgrade: a new build, a rollback, the move across the rename, and the trigger changes (presets removed, limits required, fields renamed).
- Troubleshoot an install: from what you see to the cause and the fix.