Hedhog's built-in AI chat (AiService, POST /ai/chat) talks directly to OpenAI, Google Gemini, and DeepSeek — these are the providers the core module supports today.

Status: ✅ Implemented — OpenAI, Gemini, and DeepSeek all work through /ai/chat.

1. Create an Integration Profile (type AI, provider openai, gemini, or deepseek) with an api_key at Settings → Integration Profiles
2. Go to Settings → Configurations → General, find AI Configuration, and point ai-openai-profile-id, ai-gemini-profile-id, and/or ai-deepseek-profile-id at that profile
3. Call POST /ai/chat with { message, provider } (provider defaults to openai) and confirm you get a real completion back

1. Create an Integration Profile of type AI, provider openai, gemini, or deepseek, with an api_key in its config
2. Go to Settings → Configurations → General, find the AI Configuration section, and set the matching field — ai-openai-profile-id (OpenAI Profile), ai-gemini-profile-id (Gemini Profile), or ai-deepseek-profile-id (DeepSeek Profile) — to that profile
3. AiService reads whichever profile is configured for the provider the caller requests (data.provider defaults to openai) and calls the corresponding API directly

If none of the profile settings (ai-openai-profile-id, ai-gemini-profile-id, ai-deepseek-profile-id) is set, the server logs a warning on boot ("No AI profiles configured") but still starts — calls to /ai/chat simply fail until you configure at least one.

data.model is optional on POST /ai/chat — if you leave it out, AiService falls back to:

---

Console: platform.openai.com

1. Go to API keys and create a new secret key
2. Optionally create a project and restrict the key to it, and set usage limits under Settings → Limits so a bug somewhere can't run away with your budget

---

Console: aistudio.google.com or console.cloud.google.com

1. In AI Studio, click Get API key for the fastest path, or in GCP enable the Generative Language API and create credentials if you need the key tied to a billing project

---

Console: platform.deepseek.com

1. Under API keys, create a new key (sk-…) and add credit under Top up — there is no free tier
2. The API is OpenAI-compatible, so Hedhog reuses the same request format and only swaps the base URL (https://api.deepseek.com)

DeepSeek models have no vision: an image attachment goes into the prompt as a text placeholder, not as an image.

---

Providers charge per token (input + output) directly to your account — Hedhog adds no markup. Set a model per use case rather than defaulting everything to the most expensive option:

• High quality / complex tasks: gpt-4o, gemini-1.5-pro
• Balanced / default: gpt-4o-mini, gemini-1.5-flash
• High volume / simple tasks: gpt-3.5-turbo, gemini-2.0-flash, deepseek-chat

---

• Call /ai/chat with provider: "openai" and confirm you get a real completion back, not a 400
• Call it again with provider: "gemini" and provider: "deepseek" and confirm the same

---