Quick start

This page is a sequential checklist to plan your first end-to-end bot experience. Each step lists a success criterion and common pitfalls. For vocabulary, read Concepts and architecture.

Screenshot checklist

You can add a screenshot per step (e.g. ![Step 2 — Project list](./images/quick-start-02.png)).

Prerequisites

Requirement	Why it matters
A valid Cere Insight account (sign-up or invite)	Protected routes require a session
The web app URL in a browser	Your deployment’s https://…
A reachable API	NEXT_PUBLIC_API_URL must point to your backend
(Live) Messaging environment and API credentials	Inbox wiring is configured per org/environment

Roles: Creating a new project usually requires ORG_ADMIN; members may edit but not create — match your org policy to the API guards.

Step 1 — Account and org context

1. Sign in from Login or Register (/login, /register).
2. If you belong to multiple orgs, pick the right tenant in the organization switcher; this drives the X-Organization-Id header.
3. After sign-in you should see the shell: Home, Projects, etc. in the sidebar.

Success: Dashboard loads; network tab shows Authorization and X-Organization-Id on API calls.

Stuck? Redirect loop to /login often means no access_token cookie — check privacy/incognito settings.

Step 2 — First project (Bot App)

1. Open Projects (/projects).
2. Create a new project: name, short description, main system prompt.
3. LLM settings (model, temperature, token cap) must satisfy the plan’s allowedModels; otherwise creation or runtime calls fail.

Success: Project card appears; detail (/projects/[id]) opens board/settings.

Stuck? 403 — role too low or super admin without org context. See Authentication.

Step 3 — Define agents

1. Create at least one agent: display name, system prompt.
2. For multiple specialists, add a router agent — the routing description must be explicit.
3. If priority fields exist, document team rules for conflicting routes.

Success: Agents appear in the project; they can be placed on the board per product version.

See Agents.

Step 4 — Knowledge base (recommended)

1. Create a base under Knowledge bases (/knowledge-bases).
2. Upload documents; wait until embeddings are COMPLETED.
3. Attach a tool or knowledge base node on the project board.

Success: Test-mode answers lean on sources or hallucinations drop.

See Knowledge bases.

Step 5 — Tools and integrations

1. Declare an API integration (base URL, auth) if you call REST services.
2. Create a tool and link it to the right agent.
3. Never publish secrets in client-side snippets — prefer secure server patterns summarized in API integrations.

Success: In test mode the model calls the tool with meaningful HTTP errors when failing.

Step 6 — Board

1. Open the project board.
2. Place nodes, e.g. Router → Specialist → KB tool.
3. Draw edges; the right detail panel edits the selected node.
4. Confirm save/autosave behavior for your version.

Screenshot

Show ≥3 nodes, one selected edge, and the detail panel — readers instantly see “where to click”.

Stuck? Empty board may mean the graph failed to load — check 401/403 in the console.

Step 7 — Test mode

1. Open test mode in the project context.
2. Send realistic user utterances; observe which agent/tool fired (highlights if available).
3. Fix routing text or edges when branching is wrong.

See Project management — Test mode.

Step 8 — Live chat (ops)

1. Create a live messaging inbox (web, email, WhatsApp…).
2. Bind the project to the inbox and configure bot / live agent assignment.
3. Embed the widget or validate with a test user.

See Live chat and messaging.

Step 9 — AI Builder (optional)

Use the AI Builder panel for natural-language sessions; periodically verify the board manually — suggestions are not always production-grade.

See AI Builder.

Next steps

• Project management
• Orchestration
• API: Projects