# Share agent traces across a team

What each agent was asked to do, the steps and tool calls it took, how it ended and what went wrong. The agent writes its own summary with add-only access, and the team reads it.

## An example

**You:** Save this trace: the refactor agent changed 14 files to move auth into middleware, ran tests twice, the second run passed. It skipped the docs folder.

**ADoor:** Added 1 entry to Agent runs and 6 to Steps

**Agent runs**

| Agent | Task | Outcome | Steps |
| --- | --- | --- | --- |
| refactor-bot | Auth into middleware | Tests pass | 6 |
| triage-bot | Label 40 issues | 3 need review | 4 |

## What you can ask later

- “What did the refactor agent do yesterday, and which files did it touch?”
- “Which runs failed this week, and at which step?”
- “How often does triage-bot need human review?”
- “Find every run that called the deploy tool.”
- “Chart agent runs per day by outcome.”

## Who can see what

- **Each agent** writes its own traces with add-only access, and can’t read other agents’ traces unless you allow it.
- **Engineers** read every trace.
- **A review agent** reads failed runs and adds comments, without editing the traces.
- Every agent gets its own access, never more than the person who connected it.

## How it compares with LangSmith, Langfuse or logs

LangSmith and Langfuse win at capturing every model call automatically, with token counts, latency and a trace viewer.

ADoor holds the summarized record a team reads: what was attempted, what happened and what was decided, written by the agent itself and queryable with plain SQL.

## How to start

1. Open an assistant like ChatGPT, Claude, Grok or Cursor and connect ADoor through MCP.
2. Tell it: “Set up agent traces: Agent runs (agent, task, started, outcome, summary) and Steps (run, order, action, tool, result).”
3. Give each agent add-only access to write its own runs and steps.

[Get your account](https://app.adoor.halmoslabs.com/signup)

## Questions people ask

### Can an agent edit its own trace afterwards?

Not if you give it add-only access. The history records any change that is made.

### How much detail should go in?

A summary per step works well. Keep raw model logs in your observability tool.

### What does it cost?

Nothing on the free plan, which holds 3 tables and 100 rows. Pro is $10 a month per workspace, with 1 GB of storage and no per-GB charges.

### Which assistants can I use?

Any of ChatGPT, Claude, Grok or Cursor, connected through MCP. Each one asks for your approval, and you decide which lists it can use.

## Related

- [Keep a daily work log](/use/daily-work-log/)
- [Let agents build on each other’s findings](/use/shared-agent-notes/)
- [Run a task queue for your agents](/use/agent-task-queue/)
- [ADoor vs Mem0](/compare/adoor-vs-mem0/)
- [ADoor vs Letta](/compare/adoor-vs-letta/)
- [All use cases](/use/)

Updated October 20, 2026.