Investigation Settings
Give Agent SRE standing notes about your environment with Context.md
Investigation settings hold the configuration that applies to every Investigation Agent SRE runs. Open them with the gear icon at the right end of the toolbar on the Investigations tab. The dialog has a navigation on the left and one section on the right; today the live section is Context.md.
Context.md
Context.md is a Markdown document, written by your team, that Agent SRE reads at the start of every investigation, before it looks at any telemetry. It is the place for what an experienced on-call engineer knows about your environment and a newcomer does not:
- Naming and tagging conventions — how services, namespaces and labels are named, and what each label means.
- Architecture — where traffic enters, which services sit in front of which, shared dependencies.
- Known noise — restarts, batch jobs and spikes that happen by design and are not findings.
- What to check first — for a given kind of alert, the dependency or component to look at before anything else.
Agent SRE treats these notes as standing guidance from the team. They override its generic assumptions about how a cluster is laid out, but never the evidence it gathers during the run: a note says what usually happens, telemetry shows what is happening now.
Investigations only: Context.md is handed to investigation runs. It is not read by Conversations or by the follow-up chat on an investigation.
Writing it
The editor accepts plain Markdown; headings and bullet lists are enough. Keep it short and specific. A document that describes the whole platform dilutes the notes that matter. The character limit is shown under the editor.
An example that covers the four areas above:
# context.md
## Naming and tagging
- Services are named <team>-<app>; the `env` label is prod | staging.
## Architecture
- All traffic enters through the `edge-gateway` namespace.
## Known noise
- `batch-*` pods restart nightly at 02:00 UTC by design.
## Check first
- For payment alerts, check the `payments-db` connection pool before anything else.Saving needs write access to the AI Investigations module. Anyone with read access can open the dialog and read the document.
Versions
Every save creates a new version; nothing is edited in place. The footer shows the current version and when it was saved. Click History to see every version with its date, author and size:
- Pick a version to load it into the editor. A banner says which version you are viewing, and the Save button becomes Restore version N.
- Restoring saves that content as the newest version, so the history stays a straight line and nothing is lost.
- Saving content identical to the current version creates no new version.
Each investigation keeps track of the version of Context.md it was given, so a past report can always be explained even after the document has changed.
Coming later
The other entries in the navigation are placeholders for the sections that follow: Investigation memories is marked Soon, Rate limits has become Token Limits, and Integrations takes you to the Integrations page where Confluence, Jira and custom MCP servers are connected.