Packages for Bayon AI systems

Shared packages keep Bayon AI websites, chat surfaces, and internal tools aligned around one design language and one typed agent contract instead of copying product infrastructure across apps.

Package selection

These packages are the reusable Bayon AI layer: visual primitives for product websites and typed contracts for assistant experiences.

Packages

Select one package to inspect its install command, contract, and integration boundaries.

@bayonai/website-agent

Shared public and internal Bayon AI agent contracts for website chat and operator workflows.

Use this package when a Bayon AI chat surface needs one source of truth for agent identity, model policy, tools, suggested prompts, and visibility boundaries.

  • Public Bayon Guide profile for future website chat experiences
  • Internal CEO/operator agent contract for company memory and planning workflows
  • Typed model policy, tool, prompt, visibility, and owned-document contracts

Package

@bayonai/website-agent

Current version: v0.1.0

Source: packages/website-agent

Boundaries

  • Does not send model requests.
  • Does not store chat history.
  • Does not own UI rendering.
  • Does not expose internal CEO documents to public visitors.

Install

npm install @bayonai/website-agent

Inside this monorepo, consume it as a workspace package.

npm run build -w @bayonai/website-agent

Usage

import { getWebsiteAgent } from '@bayonai/website-agent';

const agent = getWebsiteAgent();

export function getChatConfig() {
  return {
    name: agent.name,
    model: agent.modelPolicy.preferred,
    systemPurpose: agent.summary,
    suggestedPrompts: agent.suggestedPrompts,
    tools: agent.tools,
  };
}

Integration checklist

  1. Import `getWebsiteAgent` in the website, API route, or server action that prepares chat configuration.
  2. Use the exported fields to seed the chat title, starter prompts, model policy, and declared tools.
  3. Keep chat runtime behavior in the consuming app; this package owns identity and contract data only.
  4. Never surface `bayonCeoAgent` or `getCeoAgent` in public website chat. That agent is internal-only.
  5. Rebuild the package after contract changes so `dist/` and type declarations stay aligned.

Public contract

Consume the selected package contract directly instead of copying configuration into app components.

`id`, `name`, `title`, `summary`, and `mission` for stable chat identity

`modelPolicy` for preferred model, fallback model, and user-selection policy

`tools` for declared capability surfaces the chat UI can expose

`suggestedPrompts` for starter actions in the chat opening state

`contactHref` for escalation into a human conversation

Building a Bayon AI app surface?

Start with the shared packages for visual consistency, then pull agent identity from the chat package when the surface needs an assistant.

Talk to Bayon AI