Skip to main content
Back to Blog
guide

What Can AI Marketing Agents Actually Read

The Orbitable Team·AI & GTM·12 Jun 2026·5 min read

Every Orbitable agent reads from one shared world model built for your company: your ICP, your brand voice guide, your product and competitor data, and whatever documents you upload into the knowledge base. Connected integrations, HubSpot, Attio, Slack, and Apollo, add live CRM, sales engagement, and prospecting data that the same fleet of 50 specialist agents (plus the Dispatcher that orchestrates them) can draw on too.

What's actually inside the world model

The world model is not a prompt you write once and forget. It is five things, set up per customer, that stay live and get referenced every time an agent produces work.

World model componentWhat it containsHow agents use it
ICPThe buyer profile you define: roles, pain points, firmographic detailAny agent writing copy or planning outreach tailors tone, hooks, and targeting to this profile
Brand voiceYour documented tone, banned words, and style rulesGenerated content, from a landing page to a cold email, is checked against this before it ships
ProductsThe product names, features, and positioning you have enteredAgents reference the actual product and feature set instead of guessing at what you sell
CompetitorsNamed competitors and how you differ from themAgents writing comparison content or sales battlecards pull from this instead of inventing a rival
Knowledge baseDocuments you upload: case studies, pricing sheets, transcripts, decks, researchAny agent can cite a real example or a real number instead of a generic claim

Because this sits behind every agent rather than inside one tool, an agent in the content squad and an agent in the sales squad are working from the same facts about your company, not two different guesses.

What you can put in the knowledge base

The knowledge base is the uploaded part of the world model. You attach documents once, into your world, and from that point every agent operating in that world can reference them. A past case study, a pricing sheet, a competitor battlecard, a call transcript, a style guide, whatever you have that describes your business, becomes something agents can cite rather than fabricate.

How many separate knowledge bases you can run depends on how many worlds your plan gives you. Founder ($89/mo) gives you 1 world, so one knowledge base. Team ($349/mo) gives you 5 worlds. Agency ($1,999/mo) gives you 25, which is the structure an agency uses to keep each client's documents, ICP, and competitors separate from every other client's. Enterprise is custom and uncapped.

How integrations add more for agents to read

Uploaded documents are static until you replace them. Integrations add data that changes on its own. Orbitable connects to HubSpot, Attio, Slack, and Apollo. HubSpot and Attio bring your CRM records into the world an agent is working in, so outreach and reporting agents are reasoning from your actual pipeline, not a description of it. Slack connects team conversation, giving agents a channel to be triggered from or context to read. Apollo brings prospecting and contact data into the same shared context. None of this replaces the knowledge base. It sits alongside it, inside the same world model, available to the same agents.

Why the same base feeds every agent, not just one

This is the actual point of the architecture, not a side benefit. With 50 specialist agents across 10 squads, and a Dispatcher that orchestrates work between them rather than acting as a 51st specialist, the value comes from not re-briefing every agent from scratch. The GTM plan runs 6 phases across 26 agent steps. The 22 playbooks and 9 schedule templates all execute inside a world, reading the same ICP, brand voice, product data, competitor data, and knowledge base at every step. An agent three steps into the plan can use a competitor answer that a different agent wrote in step one, because it is reading the same world model, not a fresh context window. Work by one agent compounds for every other agent working in that world.

What agents cannot read

Agents are scoped to their own world. A world model is built per customer, so agents working in one client's world do not see documents, ICP, or competitor data uploaded into a different world, even on an Agency plan with 25 worlds running side by side. Agents also do not ingest anything you have not uploaded or connected. There is no passive scraping of your site or your inbox happening in the background: the knowledge base only contains what you put in it, and the integrations only surface what HubSpot, Attio, Slack, or Apollo actually hold.

The MCP server changes how the fleet is accessed, not what it can read. It exposes every agent as a tool to Claude or any other MCP client, so you can drive the fleet from outside the Orbitable app. What comes back is still an agent working from the same world model, not a raw export of the knowledge base itself.

FAQ

Do I need to upload the same document to every agent separately?

No. You upload a document once into your world's knowledge base, and every agent operating inside that world can reference it from that point on. There is no per-agent re-upload step.

Can agents in one client's world see another client's documents?

No. Each world model is scoped to one customer, so the ICP, brand voice, products, competitors, and knowledge base in one world are not visible to agents working in a different world, even under the same Agency account with 25 worlds.

Does connecting HubSpot or Apollo replace the knowledge base?

No. Integrations add live, changing data, CRM records from HubSpot and Attio, prospecting data from Apollo, team conversation from Slack, and this sits alongside the uploaded knowledge base inside the same world model. Both feed the same agents.

Can I pull the knowledge base out of Orbitable through the MCP server?

The MCP server exposes the 50 agents as tools to Claude and other MCP clients, not the raw file store. External systems trigger an agent that already has access to the world model, rather than downloading the knowledge base directly.

Does reading the knowledge base cost credits?

Reviewing work never costs credits. Credits meter the work agents produce, and when they run out, work pauses until a credit pack or the monthly reset tops it up, with no overage billing either way.

Read More