The agent registry
What an AI agent record holds, the five agents that ship, and how to add or pause one.
An agent is a record you manage under Social Media → AI Workforce → AI Agents. It is not hard-coded behaviour: you can read what it is for, change its limits and pause it.
The form#
| Field | Meaning |
|---|---|
| Agent Name | Required, unique per company. |
| Role | How you would describe it if it were a human colleague. |
| Category | Sales & Prospecting, Marketing & Campaigns, Content & Social, Support & Retention, Brand & Compliance, or Intelligence & Analytics. |
| State | Not Deployed, Active or Paused. |
| Autonomy | How much it may do without a human. See Autonomy levels. Managers only can change it. |
| What This Means | A plain-language explanation of the chosen level, filled in automatically. |
| Max Posts / Day | The daily cap on autonomous publishing. 0 means no limit. |
| Approval Required for Media | On: posts carrying an image or other attachment always need a human, whatever the autonomy level. |
| Odoo Data Access | Comma-separated technical model names the agent is meant to read, for example crm.lead, sale.order. |
| Channels | Comma-separated channels it works on, for example Telegram, Facebook. |
| What This Agent Does and Responsibilities | Free text; one responsibility per line. |
| Company | Multi-company setups. |
Warning
Odoo Data Access and Channels are declarations: they document what the agent is meant to reach, for people reading the registry. The module does not use them to restrict database access or to route posts. Do not rely on them as a security control. The controls that are enforced are the ones under Autonomy levels.
Buttons and smart button#
- Activate and Pause (Managers only) change the state. Only Active agents may publish autonomously; a paused agent is refused.
- The Actions smart button opens this agent's activity log.
- The Activity tab lists the same entries inline.
The list view greys out paused agents. Filters: Active, Paused, Needs Approval to Act, Acts Independently. Group by Category, Autonomy or Status.
Counters#
The form and list show Posts Handled, Awaiting Approval and Logged Actions for each agent. They are computed with grouped queries, so the list stays fast with many agents.
The five agents that ship#
They are created on install and are all Active. They are yours to edit: the data is loaded once and updates do not overwrite your changes.
| Agent | Role | Autonomy | Daily cap | Media needs a human |
|---|---|---|---|---|
| AI Content Creator | Multilingual copywriting and asset production: English, Amharic and Afaan Oromo. Never publishes on its own. | Approval Required | none | yes |
| AI Social Media Manager | Schedules approved content and publishes routine text posts within its cap. | Semi-Autonomous | 8 | yes |
| AI Brand Guardian | Checks copy against brand vocabulary and Ethiopian advertising rules. Only ever blocks and advises. | Autonomous | none | no |
| AI Engagement Agent | Triages comments, drafts replies for a human to send. | Assisted | none | yes |
| AI Marketing Analyst | Reads published post metrics and reports what works. Read-only. | Autonomous | none | no |
Note
Brand Guardian and Marketing Analyst are Autonomous only because they never publish: one blocks and advises, the other reads. The autonomy level describes what an agent is allowed to do, and these two are not given the ability to post.
Assigning an agent to a post#
On a post's form, AI Workforce & Governance → Handled By sets the responsible agent. That choice is what the publishing gate consults. A post with no agent is treated as human work and needs approval unless a Manager publishes it.
Adding your own agent#
- AI Agents → New.
- Give it a name, role, category, description and responsibilities.
- Choose the lowest autonomy that does the job. Start at Approval Required and raise it once you trust its work.
- Set Max Posts / Day and leave Approval Required for Media on unless you have a reason.
- Press Activate.
Changing an agent's autonomy is tracked in its chatter, and each log entry records the autonomy the agent had at that moment, so a later change never rewrites history.