Agents

Agents are AI specialists you hire into your workspace. Each one has a focus, a set of custom instructions, and a corner of your engine it keeps current. Think of them as team members, not a chatbot: a competitor analyst, a user researcher, a content strategist — each doing its job on a schedule and answering when you call it.

Every workspace starts with five, and you can add your own. They're shared across the whole team.

The five built-in agents

Agent	Handle	What it owns
Ecosystem Agent	@ecosystem	The market and ecosystem you operate in — platforms, channels, communities, where attention is moving. Keeps ecosystem briefs in your reference library.
User Agent	@users	Your audience — who they are, what they want, how they talk, what makes them switch. Adds and refines your audience personas.
Competitor Agent	@competitors	Your competitive set — new entrants, positioning moves, strengths, weaknesses, and where you win. Maintains your competitive landscape.
Content Strategist	@strategist	What to publish and why — topics, angles, formats, and cadence, grounded in what's converting.
Analyst Agent	@analyst	The numbers — reach, conversion, growth, anomalies, and simple projections, always from your real analytics.

You can edit any built-in's focus, layer on your own instructions, pause it, or remove it. A removed built-in stays removed until you restore it.

Four ways to invoke an agent

1. @-tag in the assistant. Type @ in the chat sidebar, pick an agent, and it answers as that specialist — grounded in your engine and its domain. Example: "@competitors, what changed this week?"
2. On a schedule. Every agent runs on its own cadence (daily by default). No prompt needed — it sweeps its area and updates your engine.
3. From MCP. Any connected AI client can call list_agents to see your roster and run_agent to run one. See Build with AI.
4. On the canvas. When you create a piece from a prompt, choose an agent to write it through — the Content Strategist's lens, say — and its focus frames the generation.

Keeping your engine current

On each scheduled sweep, an agent checks its area for what's new and updates your engine:

• New findings are added automatically — a competitor that just launched, a sharper persona, a fresh ecosystem brief, a growth insight.
• Changes that would overwrite something you curated are held back and collected into one daily Suggested updates digest on the Agents page. Review them per item — apply or dismiss with a click.

You decide how much autonomy each agent has:

• Auto-add new, review overwrites (default) — the balance above.
• Auto-apply everything — including overwrites, no review. Everything is logged and revertable.
• Review everything — even new items wait for your approval.

Model and billing

Agents run on your workspace's model setting, just like generation:

• Auto by default: the best model while your included usage lasts, then the most economical model once it's spent. You can pin a specific model per agent, within your plan's allowed models.
• Runs count toward your plan — they're metered like any other AI usage.
• When you hit your limit, scheduled agents automatically drop from daily to weekly so they keep working without running up overage.

Managing agents

From the Agents page you can:

• Add an agent — give it a name, an @handle, a focus, a domain, a schedule, and an autonomy setting. It's shared with the whole workspace.
• Edit any agent's focus and instructions, including the built-ins.
• Pause / resume an agent (paused agents don't run or answer).
• Remove an agent.
• Run now and watch its run timeline — status, model, and a summary of what it found.

Creating, editing, removing, scheduling, and setting auto-apply are manager-level actions; any member can run an agent and see the roster.