Skip to main content

​
What agents can do

After connecting, an agent sees two groups of tools. Which ones it may call depends on the permissions you granted and on your role in each project.

​
Watch and read

​
Deploy

Available with Read + deploy, on projects where an admin enabled agent write access. Any of these aimed at a production deployment returns confirm_required first. The agent has to call again with confirm: true, which in practice means it asks you. Nothing destructive is exposed. Agents cannot delete projects or deployments, restore backups, or change domains.

​
Odoo data

The same odoo_*, doploy_sql_* and doploy_schema_* tools available in doploy AI chat are available to connected agents. A call can name a connection_id from doploy_list_connections. If you can reach only one database, or your login has a default one, the agent can leave it out and doploy picks that database. Write tools always need a named database when you can reach more than one. Every answer says which database it came from. What the agent may do on a database follows your own Odoo identity and tier there: read, write, or execute, and whether direct SQL is allowed. Access is checked on every call, so when a project admin adds or removes a database for you on the AI users page, the agent sees the change on its next request without signing in again.

​
The loop

The tools are designed around one workflow:
  1. Agent pushes code to a branch, or creates one with platform_create_dev_branch.
  2. doploy builds it and runs the tests.
  3. Agent reads platform_get_test_results and platform_get_build_logs, fixes the code, pushes again.
  4. Build goes green. The agent opens the pull request.
  5. After merge, the agent runs platform_update_modules on staging and checks the Odoo log.
Every write is recorded in the project’s audit log with the app that made it.