OpenAI

Agent Lab: Run local Ollama agents in the dashboard — no SDK required (Quick Start). This guide is for ingesting traces from this provider via the SDK or API.

Use the SDK or API to store OpenAI runs in AITracer — the same Traces view you get from Agent Lab and coach.

Each ingested record typically includes:

• prompt and response text
• model and action name
• token counts and latency
• optional tool-call metadata

Production runs land next to local Ollama traces, so you can search history, compare coach sessions, and export training data from one place.

---

OpenAI integration workflow

Rendering diagram...

---

Standard API requests

After your app calls the Chat Completions or Responses API, call trace.record (or POST /api/traces) with:

• prompt / response (or structured input/output)
• model (e.g. gpt-4o)
• metrics: promptTokens, completionTokens, latencyMs
• actionName or workflow for filtering in the dashboard

See Trace Ingestion API for the full schema.

---

Tool and agent workflows

If OpenAI triggers function calls or multi-step agents, include tool metadata in parameters or nested spans (OpenAI Agents guide: OpenAI Agents Integration).

Useful fields:

• function name and arguments
• retry count
• failure reason
• parent trace or conversation id

---

Coach and training

Traces from OpenAI ingest can be:

• reviewed in Traces alongside Agent Lab runs
• sent through Coach if you copy a production example into a local draft
• exported from Training when marked curated (trainingPreferred)

You do not need governance or verification enabled to use these flows.

---

Optional advanced (production compliance)

If your deployment enables them on ingest:

• Governance Engine — policy and PII signals
• Verification Layer — integrity checks
• Audit Vault — long-term audit exports

Most Agent Lab users can skip this section until they have a concrete compliance requirement.