See what the brain did
Every assistant turn, extraction and background job leaves a record in Jackdaw. These screens show what ran, in what order, how long it took and what it cost.
| Screen | Where | Use it to |
|---|---|---|
| Traces | menu, System | Open one turn or job step by step: the prompt, each tool call and result, the model, tokens, cost and time. Sort by cost or duration to find expensive or slow turns. |
| Debug | menu, System | Health dashboards (tabs below). |
| Audit log | Settings | See who did what and when: logins, agents, MCP clients, members and clients. |
| Config | Settings | Compare your agents, skills, tools and workers with what the release ships. Shows what an update added and what you changed. |
Debug tabs
| Tab | Shows |
|---|---|
| Overview | Traffic, spend, embedding cache and failures at a glance. |
| Spend | Token cost by model and by agent, last 7 days. |
| Context | For each turn: the question, the context sent to the model and the reply. |
| Journey | Each thing that happened (a message, a file, an email) and what the brain did in response, live. |
| Integrity | Live brain activity shows each new item’s summary, embedding, chunks and facts as it lands. Corpus audit scans everything stored for broken records. |
| Sanity check | Configuration checks, each failure with its fix. |
| Topics, Digests, Facts | What the brain has learned and rolled up. |
| Agents, Telegram | Agent activity, and paired Telegram chats. |
| Tool validation | Tool calls the argument checker flagged. |
Common questions
- Why did the assistant answer that? Open the turn in Traces, or check Debug > Context for what it was given.
- Why is an item not found in search? Open Debug > Integrity and check the item got a summary and an embedding.
- What is costing money? Debug > Spend, then sort Traces by cost.
- Did the update land? Settings > Config lists anything missing.
- Who changed this? Settings > Audit log, filtered by action and date.
Container logs on the box are covered in Troubleshooting.