AI Transparency Help

Audit LLM call history, review costs, export training data, and understand agent operations.

Browse Calls

Every LLM call made by SOBS — helper, agent, or anomaly — is logged here. Filter by operation type, service, model, or date range.

Review Conversations

Expand any row to see the full conversation: system prompt, user messages, assistant replies, and raw JSON payload.

Export Training Data

Use the Export JSONL button to download conversations as fine-tuning data in OpenAI-compatible JSONL format.

Filter Controls

Control	Description
Operation	Filter by call type: helper, agent, anomaly, guard, and others.
Service	Filter to a specific service associated with the AI call.
Model	Filter to a specific LLM model identifier (e.g. `gpt-4o`).
From / To	Date-time range for the call timestamp.

Cost Estimation

The Cost column shows an estimated USD cost per call based on the model's published token pricing. Costs are estimates only and do not reflect actual billing from your LLM provider.

Cost estimates are based on prompt and completion token counts multiplied by hard-coded per-model rates. They may be inaccurate if your provider uses a different pricing tier.

Markdown Toggle

The MD toggle in each conversation row switches the assistant reply between rendered Markdown and raw text. Use raw mode to copy the exact model output for debugging or fine-tuning.

AI Settings — configure LLM provider, model, and guard settings.
Agent Settings — create and manage autonomous agent rules.
Work Items — GitHub issues created by agent runs.