OpenAI Integration

Honeycomb uses the OpenAI API for AI-powered content features in the site builder and AI docs apps. All calls are server-side only — the API key never reaches the browser.

SDK Setup

The OpenAI client is a lazy singleton in src/lib/modules/sitebuilder/services/ai.service.ts:

import OpenAI from "openai";
import { OPENAI_API_KEY } from "$env/static/private";

let _openai: OpenAI | null = null;
function openai(): OpenAI {
  if (!_openai) {
    _openai = new OpenAI({ apiKey: OPENAI_API_KEY });
  }
  return _openai;
}

A separate Supabase service-role client writes usage records to the ext_builder.ai_generations table.

Capabilities

Text Generation

Generates website copy using gpt-4o-mini with a system prompt tailored for conversion-focused copywriting.

Text Modification

Simplify, shorten, lengthen, fix spelling, change tone, or translate existing text.

Image Generation

Creates images via DALL-E 3 with configurable size and style (vivid or natural).

Usage Tracking

Every generation is logged with token count, model, prompt, and response for billing and audit.

Text Generation

export async function generateText(
  userId: string,
  prompt: string,
  context?: string,
): Promise<{ text: string; tokensUsed: number }>

Model: gpt-4o-mini
Temperature: 1
Max tokens: 1024
System prompt: Instructs the model to act as a professional copywriter producing clean, conversion-focused copy without markdown formatting.
An optional context parameter injects background about the website before the user’s prompt.

Text Modification

export async function modifyText(
  userId: string,
  text: string,
  action: AiModifyAction,
  params?: { tone?: AiTone; language?: string },
): Promise<{ text: string; tokensUsed: number }>

Supported actions:

Action	Description
`simplify`	Make text easier to understand
`shorten`	Condense while keeping the key message
`lengthen`	Expand with more detail
`fix_spelling`	Correct spelling, grammar, and punctuation
`change_tone`	Rewrite in a specified tone (requires `params.tone`)
`translate`	Translate to a target language (requires `params.language`)

Model: gpt-4o-mini
Temperature: 0.7 (lower than generation for more faithful rewrites)

Image Generation

export async function generateImage(
  userId: string,
  prompt: string,
  size?: "1024x1024" | "1792x1024" | "1024x1792",
  style?: "vivid" | "natural",
): Promise<{ url: string }>

Model: DALL-E 3
Default size: 1024x1024
Default style: vivid

The returned URL is a temporary Stripe-hosted image. The caller is responsible for downloading and storing it in Supabase Storage if persistence is needed.

Usage Tracking

Every call writes a row to ext_builder.ai_generations:

Column	Description
`user_id`	The authenticated user
`type`	`generate_text`, `modify_text`, `generate_image`, or `modify_image`
`prompt`	The input prompt or action description
`response`	Generated text or image URL
`tokens_used`	Total tokens consumed (1 for image generations)
`model`	Model identifier (`gpt-4o-mini`, `dall-e-3`)

Querying Usage

// Get paginated generation history
await listAiGenerations(userId, { limit: 25, offset: 0 });

// Get current month stats
await getUsageStats(userId);
// Returns: { text_tokens_used, image_count, period_start }

Claude and Other Models

The AI Docs app (src/lib/apps/aidocs) supports multiple model providers. The model constants are defined in src/lib/apps/aidocs/constants.ts:

export const AI_MODELS = [
  "gpt-4o",
  "gpt-4o-mini",
  "gpt-4-turbo",
  "gpt-3.5-turbo",
  "claude-sonnet-4-20250514",
  "claude-haiku-4-5-20251001",
  "gemini-2.5-pro",
  "gemini-2.5-flash",
] as const;

export const DEFAULT_MODEL = "gpt-4o" as const;

Key Files

File	Purpose
`src/lib/modules/sitebuilder/services/ai.service.ts`	OpenAI client, text/image generation, usage tracking
`src/lib/apps/aidocs/constants.ts`	Multi-provider model list and defaults