Skip to content

Roadmap

Planned

  • Read-only production data for the skill: e.g. query results from a read replica or BI tool, through a guarded tool with query checks, row caps and timeouts, so RCAs can confirm hypotheses against real data.
  • More chat providers: Slack and Microsoft Teams, behind the existing provider interface.
  • Configurable reply templates, beyond the disclaimer setting.
  • Packaging: a service unit / container image and a pyproject.toml.
  • Triage evaluation: replay stored history through triage to compare models and tune rules.
  • Mitigation: possibly, later, carefully scoped automatic mitigation for well-understood alerts. Today NIMI only suggests.

Known limitations

  • Local checkouts may differ from what is deployed; the skill records the commit it analyzed.
  • The skill can't see production logs, metrics or databases; it reasons from the alert, history, code and runbooks.
  • One process, one host: fine for a team's alert volume, not built for high availability.
  • Google Chat is the only provider so far.