The guide and what loads it
| Term | Definition |
|---|---|
| The guide | The on-page assistant Puntego renders for visitors. It reads the live DOM of the surface in front of the visitor and can point, highlight, scroll, narrate, run tours, and execute confirmed actions. See Guide Tools SDK. |
| Boot script | The install snippet you add to your site — a single script tag that loads Puntego, calls /boot, and mounts the guide. It needs both data-app-id (your app ID, the gp_ token) and data-gp-api. See Install Puntego. |
| Runtime | The in-page engine the boot script loads. It inventories the live DOM, evaluates the route rules, mounts the launcher, and runs point, highlight, tour, and action requests. See Configuration. |
| Visitor JWT | The short-lived bearer token minted at /boot (roughly a 15-minute TTL) that authorizes /chat as Authorization: Bearer <visitor-jwt>. See Authentication. |
Launcher and surface modes
| Term | Definition |
|---|---|
| Bubble mode | The cursor launcher’s fallback on touch-primary or narrow (under 768px) screens, where the cursor mode switches to a bubble experience. The default floating launcher does not switch — it keeps its floating button on those screens. See Configuration. |
| Sub-app in a frame | A part of your product that loads in an iframe from another of your domains (billing, an editor). With its exact origin allowed in Dashboard → Install → Sub-apps in frames and the child snippet on its pages, the guide on your page reads it, points into it and, where you turned on Allow actions in this frame for it (off by default), acts in it after the visitor approves on your page. See Sub-apps in iframes. |
| Canvas mode | The mode the runtime enters when a canvas or WebGL surface covers more than ~60% of the viewport. The underlying controls are not exposed as DOM targets, so the guide works as a typed assistant rather than pointing at controls. See Troubleshooting. |
Targeting and configuration
| Term | Definition |
|---|---|
| Target | A control or element on the page the guide can point at, highlight, or act on. The guide resolves targets from the live DOM the visitor has loaded. See Guide Tools SDK. |
| Declared target | An owner-registered target — the highest-priority grounding tier. The guide resolves a declared target before falling back to live-DOM heuristics, and click actions must bind to one. Declared targets and actions are authored in your workspace settings (the Guide Action Contract); the server holds that manifest and it is the only thing that authorizes an action that changes the page (fill, select, click). Pointing at, highlighting, or scrolling to a visible, non-sensitive element on the current page needs no declaration; a declared target marked sensitive, or with that visual action switched off, vetoes it. The SDK’s registerTarget / registerAction add page-side grounding — they tell the guide what is on the page, and they do not grant permission. See Guide Tools SDK. |
| Manifest | The workspace configuration returned by /boot: branding, consent directive, enabled guide tools, declared targets, and allowed actions. It is held server-side and never accepted from the page. The runtime can only take page-changing actions that are enabled for the workspace and declared by this manifest; visual guidance (point, highlight, scroll) can mark any visible, non-sensitive element that the manifest does not veto. See Configuration. |
| Runtime-map | A dashboard view of your account’s live runtime state — active knowledge sources, playbooks, declared targets and actions, privacy mode, and open Inbox issues — that the owner assistant reads to reason about the account. |
| Agent-home | The owner-facing dashboard home that packs the runtime-map together with guardrails, guide tools, declared targets, model lanes, memory policy, and recommended next actions into one view. |
| Lane | An isolated runtime path with its own model, capabilities, token budget, and privacy rules. The model lanes are Visitor guide (the on-page assistant), Owner operator (dashboard drafting), Voice IO (speech transport), and Workflow eval (offline checks). Owners enable specific lanes and action instances rather than flipping one global switch. |
| Tests | Question sets an owner or admin runs against the draft (unpublished changes) and against what visitors get now (Dashboard → Guides → Tests). Publish waits while a rule fails on the draft. See Tests before you publish. |
| Inbox | The owner review queue (Dashboard → Inbox). The workspace flags items here for review — unanswered questions, failed actions, missing or stale sources, contradictions, escalations, and policy blocks — and Verified Action problems land here too. See Verified Actions. |
Verified Actions
| Term | Definition |
|---|---|
| Verified Action | A customer-approved tool that lets the guide read or change a narrow business object after policy checks. It is not a generic API agent: every action targets one fixed operation, runs server-side, and starts disabled until a workspace owner or admin enables it. See Verified Actions. |
| Curated template | One of the seven pre-built Verified Action templates, each with a fixed schema, risk class, identity requirement, and confirmation requirement. Six of the seven ship a managed adapter you connect from the dashboard. See Verified Actions. |
| Custom read-only action | A manually scoped read-only Verified Action for when no curated template fits. It calls one fixed method against one fixed base URL and path, with an input schema, output schema, per-minute rate limit, and a test input; the redaction map is optional. See Verified Actions. |
Next steps
Install Puntego
Add the boot script and its required data attributes.
Guide Tools SDK
Register targets and shadow roots so the guide can reach more of your page.
Verified Actions
Curated templates, custom read-only actions, and the policy gate.
Troubleshooting
Diagnose boot, chat, voice, and visible-page issues.