Docs
Terminology
Terminology
Core terms for Agent Lab, traces, training, and optional advanced platform features.
Agent Lab
| Term | Definition |
|---|---|
| Agent Lab | The dashboard workflow for running local agents, viewing traces, and exporting training data. |
| Local run | An inference call to Ollama via /api/local/run or the Run agent UI. Cost is $0. |
| Trace | A stored record of one AI run: prompt, response, model, tokens, latency, and metadata. |
| Action name | A label on a trace (e.g. agent_lab_run, agent_chat) used to filter training exports. |
| Coach session | A paired local draft + optional paid review. Both runs are traced. |
| Training example | A completed trace with prompt and response, used for JSONL export or Ollama model build. |
| Coach pick | A trace marked trainingPreferred: true after you choose local or coach version. |
| Curated export | JSONL download filtered to coach picks only (curatedOnly=1). |
What a trace contains
| Term | Definition |
|---|---|
| Prompt | The user input for that run (or last user turn in chat). |
| Response | The model output stored on the trace. |
| Execution record | Per-run metrics row linked to a trace (tokens, latency, estimated cost). |
| Ingest | Saving a trace via Agent Lab APIs, coach APIs, or POST /api/traces / SDK. |
Local runs set estimatedCostUsd: 0. Paid coach or cloud provider runs store actual estimated cost when available.
Training
| Term | Definition |
|---|---|
| JSONL export | Download of prompt/response pairs in OpenAI messages or Alpaca instruction format. |
| Modelfile | Ollama format built from traces (FROM, SYSTEM, MESSAGE pairs) for ollama create. |
| Base model | The Ollama model you extend when building from traces (e.g. llama3.2). |
Advanced platform (optional)
These apply when you ingest production traffic or enable compliance features — not required for Agent Lab.
| Term | Definition |
|---|---|
| Governance | Policy checks and risk heuristics on ingested traces. |
| Verification | SHA-256 integrity checks on stored execution records. |
| Audit Vault | Long-term audit event storage and export for compliance workflows. |
| Cost intelligence | Spend and token analytics across providers and workflows. |
See Governance Engine and Verification Layer for details.