EKOSdocs
Docs / Operations Operate

Production Setup

A checklist for running EKOS beyond a laptop.

  • [ ] One writer. One indexing job per workspace, or one console per home — never two consoles over one home, never more than one replica.
  • [ ] Console in production mode (EKOS_CONSOLE_PRODUCTION=true): OIDC, at least one owner, a session secret. Roles granted explicitly; API tokens for agents, never a person's sign-in.
  • [ ] TLS in front of the console or any --tcp / --http MCP server; MCP servers outside the console use --token-file.
  • [ ] Dedicated sources. Do not observe directories holding other projects' virtualenvs, build outputs or data dumps.
  • [ ] Exclusions reviewed — ekos config preview-scan before the first build; ignore-patterns takes exact directory or file names, not globs. Data dumps (one INSERT per row) make commits slow.
  • [ ] Secrets — provider keys and database passwords from the environment, a keychain or the cloud's secret store, never in ekos.toml (it holds only variable names). Add [security] patterns for your own secret shapes.
  • [ ] Live databases — a read-only user; ekos database test confirms writes are refused.
  • [ ] LLM spend — leave [llm-description], [document-semantics] and [semantics] llm-definitions off in automation unless intended; in the console, Allow metered model calls per tenant; [llm] budget-usd refuses a call that could exceed a budget; ekos llm usage shows what was spent.
  • [ ] Health checks — /api/health for the console, ekos doctor --json per project.
  • [ ] Backups — ekos home backup, scheduled, copied off the machine (--to object storage). See Backup & Recovery.
  • [ ] Pin the version and read Upgrades before changing it.