YAA\IF Platform

Observability

The Third Eye: every request correlated across seven layers

Trust is the promise. The Third Eye is the evidence. Every request, whether it arrived from a Teams channel, a portal, or an API, is correlated end-to-end and reconstructable when someone asks what the agents actually did over the weekend.

Every request travels through seven layers (Channel, Agent, Skill, Tool, Model, Memory, Response). The Third Eye stitches the layers into one structured, queryable trace, with cost attribution and audit kept separate from telemetry.

YAAIF Third Eye Request R1 · 158cd785…6dfa
channel: teams healthy 2 events · 1 turn 1.84s
tokensin 703·out 33·total 736
  1. L1 Channel Teams inbound transport · 1 req · 24 events
  2. L2 Agent Default Agent orchestration · 2 turns · healthy path
  3. L3 Skill equipment-details-read capability routing · 1 selected · 5 events
  4. L4 Tool sap_get_equipment_data execution · 1 started · 2 I/O captured
  5. L5 Model openai / gpt-5.4 LLM context · 2 calls · 2 contexts · 1 patch
  6. L6 Memory Episodic agent memory · 3 events · 3 active · 0 evicted
  7. L7 Response Completion response stage · 1 completed · 0 failures
flow intake LLM turn 1 tool loop · skill_search LLM turn 2 response · completion

Representative Third Eye view. Select any request to highlight its path across every layer. Switch between Flow (chronological reasoning) and Layers (system correlation) without leaving the trace.

What the Third Eye gives you

Evidence your platform, finance, and audit teams can each use

Request overlay across 7 layers

One correlated trace per request: who asked, which agent orchestrated it, which skill was selected, every tool called, every model invocation, every memory event, every response. Re-construct what happened without piecing logs together.

Cost transparency by user, agent, and workflow

Tokens, model spend, and credits are attributed to the tenant, user, and workflow that incurred them. No surprise invoices and no untracked AI spend.

Audit kept separate from telemetry

Operational logs are for debugging and stay fail-open. Approvals, policy decisions, and external-system access keep their own append-only audit trails for regulators and internal review.

Drops into your SRE stack

Built on open standards (OpenTelemetry traces, Prometheus metrics, standard log shipping) so traces, metrics, and logs flow into the dashboards and alerting your platform team already runs.

Session → request → turn → tool loop

The Third Eye preserves the real hierarchy: a session holds many requests, each request its turns, each turn its tool loops and model calls. Streaming progress for in-flight work, durable history for the questions that come weeks later.

Tenant-scoped from ingestion to query

Multi-tenant isolation enforced at the data layer, not just the UI. One tenant's runs, sessions, and costs never surface in another's views or alerts.

Sensitive payloads are consent-gated. The platform captures what the model saw and what it returned, but viewing those payloads requires a configurable consent grant. Debugging power does not become a privacy backdoor.

More on the platform

Continue exploring how YAAIF runs

Trust

The control posture that holds up to audit and regulators.

How it works

The five-layer architecture every request flows through.

Deploy

Bare-metal Linux, your cloud, or managed SaaS. Reversible.

Next step

See how the Third Eye fits your existing SRE and audit stack

A 30-minute briefing walks through OpenTelemetry export, audit separation, and consent gating against your platform team's existing dashboards.

Book a 30-min briefing Re-read the trust posture