Knowledge base

What to put in it

Add the reference material an assistant would need to represent you well: how your business works, who your customers are, how you talk, and the facts agents should treat as ground truth.

• Company basics — what you do, your products, pricing, and policies.
• Voice and style — how you want emails and content to sound.
• Playbooks — how you handle common situations (a refund request, a sales follow-up).
• Reference documents — contracts, spec sheets, FAQs, onboarding material.
• Standing facts — key people, accounts, and details agents should always know.

How agents use knowledge

When you give Alex a task, the agents retrieve the relevant pieces of your knowledge base and use them as context for the work — grounding a drafted email in your voice, or a finance summary in your policies. Because retrieval is scoped to your workspace, this context never leaks across the tenant boundary (see the workspace guide). Knowledge complements integrations: integrations are live data, knowledge is the stable context that makes agents act like they know your business.

Supported formats

Upload common document and text formats — for example PDFs, Word documents, plain text and Markdown, and pasted notes — plus short typed-in facts. Well-structured text works best: clear headings and plain language are easier for agents to use accurately than dense, heavily formatted files.

Organizing topics

• Group related material by topic or department so retrieval pulls the right context.
• Prefer several focused documents over one giant catch-all file.
• Keep it current — remove or replace outdated facts so agents don't act on stale information.
• Name things clearly; a descriptive title helps agents pick the right source.

The knowledge graph

Beyond storing documents, AgentCorp connects the entities in your knowledge — people, accounts, products, and the facts that relate them — into a graph. That lets agents follow relationships (this contact belongs to that account, which is on this plan) rather than treating every document in isolation, which makes their answers more consistent and their work more context-aware over time.