User guideAI Teammates
Operator flags
| Variable | Effect |
|---|---|
OWLPANE_AI_AGENTS=off | Kill switch for auto agent dispatch (Ask still works if LLM configured). |
OWLPANE_AI_AGENT_MAX_RUNS_PER_HOUR | Per-tenant hourly cap for auto runs (default 60). |
OWLPANE_AI_APPROVE_EXECUTE=true | Approved queue actions run mutators. |
OWLPANE_AI_APPROVE_K8S_EXECUTE=1 | Restart/scale approvals with payload.cluster queue kubernetes_ops (agent applies). |
OWLPANE_AI_AGENT_HANDOFF=1 | Triage/correlation warn/critical → chain run to hired Runbook teammate. |
OWLPANE_AI_ESCALATION_BRIDGE=1 | Critical briefings ping immediate escalation channels. |
OWLPANE_AI_SLACK_APPROVE=1 | Post approve/reject buttons to Slack for pending proposals. |
OWLPANE_AI_SLACK_APPROVE_RBAC=1 | Slack approve/reject only for users linked via /v1/ai/slack-approve/links. |
OWLPANE_AI_RUNBOOK_VECTOR=1 | Reserved for pgvector RAG (stub; FTS remains default). |
Helm (owlpane-api chart): aiApproveExecute, aiApproveK8sExecute, aiAgentHandoff, aiEscalationBridge, aiSlackApprove, aiSlackApproveRbac, pdfWorker.url. Hetzner staging enables approve/handoff/bridge and PDF worker URL in values-hetzner.yaml. See SEO2LS/seo-infra/helm-charts/owlpane-api/STAGING-AI-TEAMMATES.md.
Postgres: migrations 089_ai_runbook_rag, 090_slack_approve_links on each environment.
Engineering e2e: npm run test:e2e:ai-teammates in owlpane/api (requires npm run build, npm run e2e:api, npm run seed:users).
Tenant BYOK and egress acknowledgment: Settings → AI in the console.
Last updated