What is the Handbook?
The Handbook is the internal counterpart to the Knowledge Base. The Knowledge Base is what you’d happily put on your public website — answers customers can read directly. The Handbook is everything you’d never put there: escalation scripts, refund policies, “if X happens, do Y”, the wording you use when a VIP customer is unhappy, the specific checklist when a payment dispute arrives.
Atender’s AI agents — Sidekick, Web Chat, voice — read the Handbook as context. It shapes how they behave: which escalation paths to suggest, what tone to take, which policies apply. They use the Handbook to inform their answers, but they never quote it directly to customers. The Handbook stays private to your team.
What lives in the Handbook
Things that are right for the Handbook:
- Escalation procedures. “If the conversation involves a chargeback, transfer to the Billing team and tag
legal-review.” - Refund policies. The actual rules — who’s eligible, how much, what approvals are needed.
- Tone and language guidance. “When a customer mentions a death in the family, lead with empathy, never offer a discount.”
- Internal contact protocols. “Issues affecting more than 10 customers go to #incidents. Issues affecting an enterprise account go to the account owner.”
- Step-by-step troubleshooting that includes internal tools. “Open the admin panel, look up the customer’s user ID, paste it into the dashboard…”
- Specific scripts. “Use this exact wording when offering a goodwill credit.”
Things that belong in the Knowledge Base instead, not the Handbook:
- Customer-readable how-tos. “How to update your payment method” goes in the KB.
- Product feature explanations. “What is your refund policy?” — the customer-facing version goes in the KB; the internal enforcement of that policy goes in the Handbook.
- FAQs that shouldn’t be a secret. If you’d publish it on your help center, it’s a KB article.
How the Handbook is structured
The Handbook is intentionally simpler than the Knowledge Base:
- Handbook — A tenant can have more than one Handbook. Each Handbook is its own partition, with its own categories and procedures. If more than one Handbook exists, use the Handbook picker to choose which one you’re working in.
- Procedure — A single Handbook entry — title, body, keywords, category. Authored in a rich-text editor.
- Category — Top-level grouping for procedures within one Handbook. Two-level depth only: categories → procedures. No subcategories, sections, or roles.
- Keywords — Up to 10 per procedure for AI retrieval.
- Visibility — A simple on/off toggle. Off means the procedure is hidden from agents and AI; useful for drafts.
- Access rules — Per-procedure or per-category settings: confidential flag plus an optional scope.
There’s no draft / published / needs-review workflow. There are no tags. There are no cross-cutting sections. The Handbook trades flexibility for speed — your team is the audience, and the audience is small enough that simpler structure works.
Access rules
Every procedure (and every category) can have one or more access rules attached. An access rule has two parts:
- Confidential flag — when on, the procedure or category is locked. Visible only to admins; hidden from agents who don’t have explicit permission.
- Scope — optional. Limits the procedure to:
- Main agent — the global AI behavior (Sidekick, default specialist).
- A specific agent config — a particular Specialist Agent or Agent Stack you’ve built. The procedure only informs that agent’s behavior.
A procedure with no scope is available to every AI agent using the same Handbook partition, before any confidential rules are applied. If your tenant has more than one Handbook, the Handbook picker determines which partition you’re editing and which set of procedures the agent can retrieve from. A procedure scoped to “agent config X” is invisible to other agents and never enters their context window.
This is how you keep the Refunds Specialist focused on refund policy without polluting the Voice Reception agent’s context with refund minutiae.
How AI agents use the Handbook
When a Specialist Agent (or Sidekick on a conversation) is composing a response, it retrieves Handbook procedures relevant to the question — same hybrid retrieval pipeline as the Knowledge Base. The retrieved procedures are added to the agent’s context as guidance, not as quotable content.
“Available to every AI agent” means every eligible agent using that same Handbook partition, before confidential flags or scope rules narrow access. If your tenant has multiple Handbooks, the selected Handbook partition controls which categories and procedures are considered for retrieval.
This is why Handbook procedures should be written in instruction language (“If X, do Y”) rather than article language (“This document explains X”). The AI follows them; it doesn’t transcribe them.
When to write a Handbook procedure vs. a KB article
Use the Handbook when:
- The content is internal — you wouldn’t show it to a customer.
- It guides agent behavior or AI behavior.
- It contains specific tools, internal jargon, or sensitive details.
Use the Knowledge Base when:
- The content is customer-facing.
- A customer searching the help center should find it.
- You’d happily link to it from your product or marketing site.
If you’re not sure, ask: “Would I be comfortable if a customer read this verbatim?” If yes, KB. If no, Handbook.
Where to go next
- Write a procedure: Write a Handbook procedure
- See the difference clearly: Knowledge Base vs. Handbook
- Configure who sees what: access rules are configured inline in the Handbook procedure editor.