Does Odoo’s native AI have an audit trail and permissions?

It keeps a record, but not one built to be evidence. First, be fair: Odoo 19 ships a real AI agent, a genuine tool-calling loop, and it is capable. When that agent acts through a server action, the trace it leaves is a chatter message on the record. We read the Odoo 19 Enterprise source, and a chatter message records what the agent attempted rather than the before-and-after state of the row, and it can be edited or deleted like any other message. So you can see that something was tried; you cannot reliably reconstruct what actually changed, or prove it was not altered afterward. An audit trail meant to stand as evidence keeps the values on both sides of every write, reads them back from the database, and is tamper-evident.

An audit trail logs what your AI said it did. It does not tell you what your database actually contains. A log records each action an agent took and what the tool reported back, which is worth having and something a number of tools now do well. But the tool’s report is a claim: if a write lands on the wrong record, or a success sentence is composed from what the action intended, a faithful log records that false success exactly as written. Verification is the separate step of re-reading the record from the database on an independent connection after the write commits, and comparing it field by field against what the caller intended. Logging tells you what the agent claimed; verification tells you what is actually in the ledger. Only the second catches the write that reported success but did not land.

Yes, at the level of the acting user’s Odoo permissions, which is real control and worth using. What Odoo’s own documentation is candid about is where enforcement lives: a server action, in Odoo’s words, "does not enforce business rules, modify records directly, or guarantee the correctness of the operation. Its role is limited to decision-making." The tools that actually write "must enforce business rules explicitly in Python code." So the guardrail is whatever the tool author wrote, not a standing per-agent policy. And we found in the source that when a permission error is raised, it is caught, converted to text, and handed back to the model as a tool result, so a denial arrives as advice the model may try to route around, rather than a wall.

Two reasons we read in the Odoo 19 source. First, there is no savepoint in the AI module’s tool execution, so if step three of a five-step action is denied, steps one and two remain written. Second, when a server action returns nothing, Odoo composes a success sentence from what the action intended and hands that to the model, so the agent can report success without the database being re-read. The fix is structural, not a better prompt: run the write in a transaction that rolls back as a whole, then read the record back from the database before anyone calls it done.

Four things. Keep the before-and-after values, not just the event. Read the record back from the database on an independent connection after commit and compare it, field by field, against what was intended. Commit a refusal on its own transaction so the record of a denial survives the rollback of the action it denied. And be tamper-evident: make alteration visible with a hash chain, while being honest that a database superuser who disables the trigger can still alter rows. That last honesty is the tell. Tamper-evident is defensible; tamper-proof is a claim no ERP add-on can keep.

For reading, displaying reports, and drafting, it is genuinely useful today, and Odoo’s standard Ask AI agent, in Odoo’s own words, "cannot make changes to the database... it can open views and display reports, but it cannot create leads or alter data," so the low-risk path is well fenced. The care is needed the moment an agent can write. There the questions are whether a failed multi-step write rolls back cleanly, whether success is confirmed by re-reading the database, and whether the record of what happened can be trusted afterward. We work those through in detail on can AI write to Odoo safely (nanti.ai/can-ai-write-to-odoo-safely).