An Agent is a configurable AI specialist that works inside your Workspace under rules you set. This page covers building one, running several together, answering their questions, and governing what they may do.
Agent Studio
Create an Agent and give it an identity: a name, a role, an avatar and a description. Write its system prompt, or start from one of the preset templates: Strategist, Copywriter, Designer Brief Writer, Media Buyer, Analyst, SEO Specialist, QA Reviewer and Community Manager.
Then configure how it works:
- Personality sliders for friendly, concise and autonomous.
- The tools it is allowed to call.
- Knowledge bases to consult, each with a weighting.
- Memory scope and how long it is retained.
- Spending caps per run and per day.
- Skill fragments, which are reusable prompt overlays that stack onto the base prompt.
Every published configuration is an immutable version. Compare two versions, roll back, publish, archive or duplicate. Before publishing, use the test panel to run the Agent against a scenario and watch the step trace as it happens.
Tools
Everything an Agent can do is a declared tool. Built in tools let an Agent read Workspace data, write and schedule posts, search the Media Library, query your analytics sources, read from the ad platforms and publish to a CMS.
Define your own as well. A custom HTTP tool points an Agent at an internal API, with domain allow listing and credentials pulled from a secrets vault rather than pasted into a prompt. Upload an API specification and Collabre generates the tools from it, or build one in the visual composer. Tool analytics report which tools get called, how often, and with what success rate.
The human in the loop inbox
Agents are built to ask rather than guess. When an Agent needs something, it raises a blocking question, pauses its own run and waits.
Questions come in named types: Information, Approval, Choice, Creative Review, Tool Approval and Drift Alert. Each is routed to the right person from expertise profiles, and each carries a deadline with a nudge at the halfway point. Answer with typed text, with a reusable answer template, or by recording a spoken reply that is transcribed for you to edit. Reassign, filter and act on several questions at once.
Departments
A Department is a group of Agents with a charter, a conductor that decomposes the brief, and member Agents that execute the subtasks. Presets cover Marketing, Content, Analytics, Operations, SaaS Growth, DTC E-commerce and Agency.
Write a brief, and the conductor produces a plan. Subtasks are dispatched to member Agents, dependencies are respected, independent work runs in parallel, and the Agents hand off through a shared Department memory. A blocking question pauses the run. At the end the Department reports back with its aggregate cost and duration.
Swap a member Agent, stack extra instructions onto one slot, watch the run dashboard live, and clone a Department that worked.
Activity view
The activity view shows what the AI workforce is doing now: a live event feed with filters, a step trace drawer, a kanban desks board with Department swimlanes, and an animated office scene where Agents work at desks and hand off tasks. Voice narration reads the events aloud, replay controls scrub back through a past run, and run comparison diffs two runs.
Governance
- Spending caps per Agent run, per Agent per day, and per Workspace per month.
- Kill switches that halt one Agent or every Agent at once.
- Tool approval rules that allow, deny or require confirmation, scoped to the Workspace, an Agent, a Department or a Campaign. The default is human confirmation on every call.
- A destructive action registry setting each classified action to automatic, human in the loop, or denied.
- Autopilot level per Campaign, from Off through Suggest Only and Auto with Approval to Full Auto, defaulting to Suggest Only.
- PII redaction on by default, rate limiting on runs and tool calls, and role based control over who may edit which part of an Agent.
Every Agent action is written to the audit log with its reasoning. See Security controls.