One observability layer for every agent in the org.
Many BUs, many runtimes, many vendors. Trefur reads what every team already emits — built on OpenTelemetry, self-hostable on your network, fan-out friendly with the backends you already run.
What you are trying to do.
- →Get one trace view across every business unit, framework, model vendor, and region.
- →Adopt agent observability without re-platforming the dozen pipelines already in flight.
- →Keep sensitive content inside the network. Redact what leaves.
- →Roll up cost and reliability per BU, per environment, per agent — for executives, not just engineers.
- →Onboard new agent workloads without each team picking a new vendor.
What hurts today.
Twelve teams, twelve tracing setups
Different frameworks, different vendors, different log shapes. Asking a cross-BU question turns into a six-week project with three vendors on the call.
Security asks where the data goes
Your agents see customer data. Your observability vendor sees the prompts. Security wants to know the path each byte takes. You cannot draw it.
Procurement renegotiates every quarter
Each team's pet observability vendor wants a contract. You consolidate, you split again, you consolidate. The agent layer is the next round.
Executives cannot see the AI program
Cost is in one report, reliability in another, customer outcomes in a third. The CTO update slide is a guess.
What Trefur gives you.
One canonical trace, every BU
OpenTelemetry GenAI semantic conventions on the wire. Trefur reads what every team already emits — no framework lock-in, no replatform.
Self-hostable collector
Single Go binary in your network. Local batching, on-disk buffering, redaction. Routes to Trefur and your existing backends in parallel.
PII redaction at the collector
Configurable redaction before anything leaves your perimeter. Tunable per environment, per agent, per field.
SAML SSO, SCIM, audit logs
Standard enterprise auth and provisioning. Audit logs for every access. Quiet-hours and role-based routing for oncall.
Rolls up for executives
Cost per BU, reliability per BU, throughput per BU. The slide builds itself; the numbers are real.
Vendor-neutral by design
OpenAI, Anthropic, Bedrock, Azure OpenAI, Gemini, Groq, MiniMax. Swap any one for another without losing trace continuity.
An enterprise rollout with Trefur.
- Q1
Procurement approves Trefur for two pilot BUs. Collector deployed inside each VPC, fan-out to existing Datadog and Honeycomb stays in place.
- Q1
Security signs off after reviewing the redaction config and the collector's on-disk encryption. Customer PII never leaves the perimeter unredacted.
- Q2
Six more BUs onboarded. The CIO sees one dashboard: 240k agent runs/day, cost per run trending down 22%, retry rate steady.
- Q2
Oncall consolidated. Every BU's agent alerts route through one PagerDuty schedule with team-level escalation. Slack channels for non-paging.
- Q3
Procurement cuts two redundant vendor contracts. Trefur Enterprise replaces them at lower TCO; the existing backends stay because Trefur fans out to them.
Stand up one agent observability layer across the org.
Enterprise tier adds SAML SSO, SCIM provisioning, dedicated support, and contract-level SLAs.