Governance and audit
Every conversation, tool call and project change is recorded and can be consulted from the Logs section, with export to CSV and JSONL.
Every change has an author, even when it was made through the Builder or an external assistant.
Every conversation can be reread step by step, with the tools called.
Accountability, traceability and controls are the basis of compliance processes, including the AI Act.
Versions and change control
- Every change to the project is a new version recorded under the name of whoever made it.
- You can go back to a previous version with rollback.
- Agent instructions have their own history, with restore.
- Workspace code has a git repository per node, with commit history.
The logs
| View | What it shows |
|---|---|
| Flows | One row per conversation: when, origin, person, first message, number of tools used, errors, duration |
| Tool calls | One row per call: tool, provider, caller, person, outcome, duration |
| Graph | The project with links weighted by number of calls and colored by error rate |
| Changes | Audit: who (also "via Builder" or "via MCP"), action, object, details, outcome, differences |
| Person | Everything a member did or caused |
You can filter by project, period (from one hour to 30 days, or custom), origin (console, Telegram, email, phone, scheduled task, delegation…), node and person. There are also the phone call log (Calls) and the scheduled runs log (Scheduled tasks).
Tracing a conversation
Opening a conversation gives you its complete trace, in three views. The detail also reports the input and output tokens declared by the model.
| View | What it is for |
|---|---|
| Waterfall | Every agent involved (including child conversations from delegations and subtasks) with its tool calls over time |
| Causal tree | Who caused what |
| Replay | The conversation moving through the project, step by step |
Checks before publishing
- Checks. Problems that keep the project from running, and suggestions with proposed fixes.
- Try it. A test chat with any agent of the project, with tool calls visible.
- Workspace logs. Runtime output, loading errors and git status for those who develop code.
Related pages
Security
Roles, groups, credentials, enterprise sign-in and least privilege: how to protect an agent system.
Building with external assistants
Using Claude, Codex or other assistants to build and develop TeraBrain NX projects, with the person's own rights.
Multiple agents that collaborate
Delegating a task, handing over the conversation, splitting work into subtasks.
Have an agent to take to production?Let’s talk.
We show you TeraBrain NX on your case: systems to connect, access and controls.