All measured systems operational
Every figure below is derived from append-only probe and execution records, not from a status field anyone here can edit. Dependencies are probed daily. Days we did not measure are shown grey rather than green, and any service that is failing shows the actual error.
Automation fleet
Counted from execution records. A workflow appears here because it ran, not because it reported that it ran.
Dependencies
90-day probe history, one cell per day
Anthropic
OperationalModel provider for qualification and long-context reasoning
Apollo
OperationalContact and company enrichment
Database and auth
OperationalPostgres, row-level security, and service-role access behind every surface
Email delivery
OperationalTransactional and outreach send path, with delivery webhooks
Email delivery (secondary)
OperationalFallback send path
Firecrawl
OperationalPage extraction for research and enrichment
Hunter
OperationalEmail discovery
OpenAI
OperationalPrimary model provider for drafting and classification
Payments
OperationalCheckout sessions, subscriptions, and invoice records
Semrush
OperationalKeyword and rank tracking data
Failures on this service are an exhausted API credit balance on our account, not a provider outage.
Twilio
OperationalSMS and voice transport
Workflow orchestration
Operationaln8n instance that dispatches scheduled agents
ZeroBounce
OperationalAddress verification. Every cold send is gated on it
What broke recently
Last 90 days. We keep failures on the record permanently.
Our fleet stopped running for days while this page showed green.
In late July our scheduled automations stopped executing. Two clusters of agents went silent on July 25 and July 26 and did not come back. Nothing alerted us, because our own deploy process was overwriting the liveness field that the staleness alarm reads. Refreshing that field daily made the alarm mathematically incapable of firing, so the dashboard stayed green through a fleet-wide outage.
We rebuilt it. Liveness now derives from execution records that no process can overwrite, this page is computed from those same records on every load, and a regression test pins the exact failure so it cannot return. If you want to know why we sell verification rather than autonomy, this is why. An agent that claims it worked is worth nothing; a record that proves it worked is the entire product.
Operational incident log
Automated monitoring records for our internal automation fleet. These track our own operations, not customer-facing service availability.
Scheduled maintenance
None scheduled in the next 30 days. Maintenance windows are announced at least 7 days in advance and held outside North American business hours when possible. Customers on enterprise SLAs are notified by email in addition to this page.
Next window · TBDSubscribe to updates
Pilot offices and integrators get email notifications scoped to the systems they rely on. Ask for status updates and we will add you to the notification list for the services you depend on.
Subscribe by emailThe numbers we publish, then live by.
These are commitments and targets, not measurements. The measured numbers are above. Per-tenant SLAs are formalized in vendor agreements; the platform-wide commitments below apply to every customer on day 1.
Measured monthly, excluding scheduled maintenance windows
Retry budget with exponential backoff before dead-letter
Production outage affecting an office's ability to take or dispatch calls
Advance notice published to /security#subprocessors before any change takes effect
The full security and operational posture.
For the underlying controls, encryption, RLS, audit logging, RBAC, AI governance, SOC 2 roadmap, subprocessors, incident response policy, see the Trust Center.