What it's for
Letting production monitoring count as evidence, and letting remediation plans open as pull requests.
When you use it
Once, after your agent is in production with observability already running.
What's on the screen
Four cards — LangSmith, Langfuse, OpenTelemetry, GitHub — each showing connection status, last sync time, and the latest snapshot summary. Configure/Edit reveals credential fields; LangSmith and Langfuse have Sync now. Below is your OTLP endpoint, and a list of recent telemetry snapshots.
How it works
LangSmith takes an API key; Langfuse takes public key, secret key, and host. Sync now pulls a snapshot of trace counts. Without credentials it records a clearly-labelled demo snapshot.
OpenTelemetry is push-based: point an OTLP/HTTP JSON exporter at /api/integrations/otel with your x-veriflow-api-key header. Veriflow counts spans and services.
A telemetry snapshot within the last 7 days marks REL-05 (monitoring) and ACC-03 (audit logging) as evidenced on your Controls page.
GitHub takes a fine-grained token with contents and pull-request write, plus owner/repo. It's what turns fix plans into PRs.
Gotchas
- Credentials are encrypted at rest with AES-256-GCM when
VERIFLOW_ENCRYPTION_KEYis set. Set it before storing a GitHub token, which can write to your repo. - Snapshots expire. After 7 days the monitoring evidence stops counting — sync on a schedule.
- Scale plan only.