Glossary

Definitions of the terms used throughout Teaflask. Each entry links to the page that explains the concept in more detail.

Some distinctions matter when configuring your integration: adding an action to the catalog is different from granting it to an agent, and releasing a doc is different from publishing an agent or help center.

A

Action

An API operation registered in your organization's catalog, defined by its HTTP method, path template, and parameter schemas. An agent must be granted the action before it can use it, and execution remains subject to the action's policy and your API's permissions. See Actions.

Adapter

The frontend code that connects the assistant's action requests to your API. Requests run in the user's browser through the authentication or API client you configure, rather than being made by Teaflask's servers. See Adapters.

Agent

A configured assistant that understands your product, answers questions, and can act on a user's behalf. Its configuration includes a prompt, model, and grants that determine what it can access and do. You can test a draft in the Playground; publishing makes a configuration available to the embedded assistant. See Agents.

Agent turn

One response cycle triggered by a user's message. A turn can include reading docs, calling tools, waiting for approval, and performing actions before the agent finishes its response. A conversation can contain many turns. See Lifecycle.

B

Balance

Your organization's prepaid funds, held in US dollars, used to pay for model usage not covered by your own provider keys. Top-ups and credits increase the balance; usage deductions reduce it. See Billing.

Brew

The assistant inside the Teaflask dashboard that can help you draft documentation from your connected codebase. Its drafts appear under Docs for review and editing before release. It is separate from the agents you configure for your product's users. See Drafting.

C

Capability

A category of functionality an agent can be granted, such as searching docs or navigating your app. Enabling a capability makes its associated tools available to the agent. Capabilities are distinct from the individual API actions in your catalog. See Agents.

Claim

A specific statement of fact in a doc, such as “the cancel button says Cancel plan.” For code-backed verification, claims are turned into executable checks in the doc revision's verification suite. See Verification.

Companion

The floating launcher in the corner of your app that opens the assistant's chat drawer. See Surfaces.

Consequence tier

An action's classification by effect: read-only, idempotent, or destructive. An idempotent operation has the same effect when repeated as when performed once. Some actions have no assigned tier. Tiers help you prioritize review; they do not determine whether an action is allowed or requires approval. See Actions.

Conversation

An assistant thread containing a user's messages, the agent's responses, tool calls, and action approvals and results. Your team can review conversations in the dashboard. See Conversations.

D

Dispatch

The record of a task delegated to a background subagent, including the work requested and the report returned. See Dispatches.

Doc

An article in your knowledge base with a stable address and a history of revisions. Its released revision is the version available to agents and other configured readers; draft edits do not replace it until you release an update. See Knowledge.

Doc provenance verification suite

The executable checks associated with a doc revision, derived from its factual claims and run against your code. Re-running the suite helps identify when code changes contradict the documentation. See Verification.

Doc request

A request to draft a doc, recorded with its prompt, the member who asked for it, and the resulting draft for review. See Drafting.

E

Embedded assistant

The assistant interface installed inside your product. It can appear as a widget or a dedicated assistant page; the agent configuration determines how it behaves, while the frontend integration determines where and how it appears. See Assistant UI.

End user

A person using the assistant inside your product. They do not need a Teaflask account. Visitors can remain anonymous or be identified by your backend using an end-user token. See Identity.

End-user token

A short-lived JSON Web Token (JWT) signed by your backend to identify the current user to Teaflask. It establishes identity, not permission to perform actions, and does not replace your API's authentication or authorization. See Identity.

F

Folder

A place to organize docs in a hierarchy. Folders can contain docs and subfolders, and can be used alongside tags to limit which docs an agent may access. See Organizing.

G

Grant

An agent configuration setting that gives it access to a capability, action, or skill. Adding an action to your organization's catalog does not grant it to an agent; you separately select which agents may use it. See Agents.

K

Knowledge scope

The set of released docs an agent may access, controlled by its tag and folder filters. Search docs and Browse docs as files use the same scope. When both filters are enabled, a doc must match both to be included. See Agent knowledge.

M

Member

Someone who belongs to your Teaflask organization, rather than an end user of your embedded assistant. Members have either the owner or member role, which determines their permissions. See Members.

Memory

Facts an agent retains between conversations. Memory is a single on or off setting: when it is on, the agent keeps notes about each signed-in end user and reads them back in that person's later conversations, and it can save a fact when the user asks it to. Memory is private to the individual end user — anonymous visitors and playground conversations are never remembered. See Memory.

O

Organization

The account boundary that owns your docs, agents, actions, keys, members, and balance. These resources belong to the organization and are isolated from other organizations. See Organization.

Origin

The protocol, hostname, and port, if present, of a web address, such as https://app.example.com or http://localhost:3000. It does not include a page path. A publishable key's allowed origins determine which sites may use it to connect to the assistant. See Publishable keys.

P

Permission

A right held by a member of your organization, such as reading docs or publishing an agent. Member permissions govern access to the Teaflask dashboard and its resources; agent grants govern what an agent may do. See Permissions.

Policy

The approval rule set on an action in your catalog: Allow runs without user approval, Ask requires approval before execution, and Deny prevents execution. An agent's grants cannot override the catalog policy. These rules do not replace your API's authorization checks. See Actions.

Proposal

A suggested change waiting for review rather than being applied automatically. Scans propose actions and pages for your catalogs; verification can also produce proposals to update docs when code changes contradict them. See Actions and Verification.

Publish

Make a version available for use. Publishing an agent activates its configuration for the embedded assistant; publishing a skill makes its instructions available to agents that have access to it. Help centers publish their site configuration and have a separate live switch. Docs use release instead. See Revisions for agent publishing.

Publishable key

A public browser credential that connects an embedded assistant to a particular agent. It has an allowed-origin list and is safe to include in frontend code. It identifies the agent, not the end user, and is not a secret API key. See Publishable keys.

R

Release

Make a chosen doc revision the current version available to agents, MCP clients, and help centers, subject to their access and visibility settings. Releasing a doc does not automatically make it public or grant every agent access to it. See Releasing.

Revision

A version of a doc's content, an agent's configuration, or a skill's instructions. Revisions preserve history so you can compare changes and restore an earlier version. A rollback creates a new revision rather than erasing the intervening history. See Revisions.

S

Scan

An inspection of your connected repositories that discovers API actions and product pages and submits them as proposals. You review the proposals before adding them to your catalogs; discovery alone does not grant an agent permission to use them. See Actions and Sitemap.

Sitemap

Your organization's catalog of product pages, discovered from your code and approved by you. It tells agents which pages exist. Permission to navigate to those pages is configured separately under App access. See Sitemap.

Skill

A reusable set of instructions that you write once and grant to one or more agents. An agent loads the skill when a request calls for it. See Skills.

Subagent

A background agent assigned a task by a parent agent. It receives a self-contained brief rather than the parent's conversation history, and its tools are limited to those permitted by the parent's grants. See Delegation.

Surface

A place where the embedded assistant appears, such as the widget or a dedicated full-height assistant page. Surfaces using the same session identity share a conversation. See Surfaces.

T

Tag

A label applied to a doc or folder. Tags on a folder are inherited by the docs and folders beneath it, and can be used to scope an agent's knowledge. See Organizing.

Tool

A function the model can call during an agent turn, such as searching docs or executing a registered action. Capabilities make their associated tools available, and each granted API action provides its own tool. See Agents.

Transaction

An entry recording a change to your organization's balance, such as a top-up, usage deduction, credit, or refund. See Transactions.

V

Verification run

One execution of a doc revision's verification suite against your code. Its results show which claims were checked and whether those checks passed. See Verification.

W

Workspace

An isolated computer an agent can be granted for working with a shell and file editor. It is scratch space for agent work, not your Teaflask organization or a place to rely on for permanent storage. See Workspace.

More resources

• Fundamentals: how these concepts fit together.
• Getting started: setting up your knowledge base, actions, and embedded assistant.
• Actions: the relationship between catalogs, grants, policies, and consequence tiers.
• Releasing: making doc revisions available without exposing unfinished edits.
• Permissions: the permissions held by members of your organization.