Breakglass
Incidents
Settings Connecting…

How Breakglass behaves, and what it's connected to

Every partner has a fallback, and the label says which one is in use. Nothing here can widen what the agent may do: that lives in the catalog and the credential policy.

ProblemA demo can quietly run on fallbacks and look live.
SolutionEach integration shows LIVE, LOCAL, PENDING or OFF, with the reason.
Why it mattersYou always know what was real in a run.
How it's usedCheck the labels before a demo; switch the agent strategy or Fault mode here.

Behaviour

Read-only here: these settings belong to a running backend.

Integrations

Connecting…
CISA KEV feed
At recording (13:38 PDT): LIVE. 1,739 entries, catalog 2026.10.08, last poll unchanged since the previous poll (HTTP 304)
RECORDED
Guild AI
At recording (13:38 PDT): GUILD SESSION. Agent sessions run in Guild; its credential proxy enforces the policy (local loop on any Guild failure).
RECORDED
OpenAI
At recording (13:38 PDT): gpt-6.1-sol. Responses API, strict tool schemas; every call logged to .data/llm-calls.jsonl.
RECORDED
Semgrep
At recording (13:38 PDT): SEMGREP CE (no account). Semgrep Community Edition runs Breakglass's custom rules over the portal source and the reverse-proxy config, for real, on every incident.
RECORDED
ClickHouse
At recording (13:38 PDT): CLICKHOUSE CLOUD. Every edge request, journeys, verification and the A/B.
RECORDED
MongoDB
At recording (13:38 PDT): ATLAS. Inventory, catalog with outcomes, advisories; change stream on new advisories.
RECORDED
ElevenLabs + Twilio
At recording (13:38 PDT): APPROVAL CALLS OFF. Outbound approval call on a denied action (SMS fallback). Off by default: live calls cost money.
RECORDED