Grabblist gives you an AI shopping assistant right inside your dashboard. But the AI brain behind it — the language model — is your choice. This guide covers every supported provider, how to get an API key, which model to pick, and how to set it up in under 2 minutes.
1. Dashboard Agent vs MCP — Two Different AIs
Grabblist has two AI-powered features. They are independent and can use different models:
| Feature | Dashboard Agent | MCP (Model Context Protocol) |
|---|---|---|
| Where | Inside grabbitapp.com dashboard | Claude Desktop or claude.ai |
| Model | Your choice (BYOK) | Claude (Sonnet 4.5 by default) |
| API Key | Required (your own) | None (uses Claude account) |
| Tools | Search, grab, collections, memory | 21 tools: browse, search, extract, grab, collections, items |
| Setup | Settings → paste API key | One-click MCP URL copy |
This guide focuses on the Dashboard Agent since it requires API key configuration. For MCP setup, see our MCP Setup Guide.
2. Supported AI Providers
The dashboard agent supports two integration modes:
- Anthropic (Native SDK) — Direct integration with Claude models. Best tool-use performance, native prompt caching.
- OpenAI-Compatible API — Works with OpenAI, DeepSeek, Groq, Together AI, OpenRouter, Ollama, LM Studio, and any API that speaks the OpenAI chat completions format.
Provider Presets
| Provider | Base URL | Recommended Models | Pricing |
|---|---|---|---|
| Anthropic | (native SDK) | Claude Opus 4.5, Sonnet 4.5, Haiku 4.5 | $3–$15/M input, $15–$75/M output |
| OpenAI | https://api.openai.com/v1 | GPT-4o, GPT-4o Mini | $2.50–$5/M input |
| DeepSeek | https://api.deepseek.com/v1 | DeepSeek V4 Pro, V4 Flash | $0.27–$1.10/M input |
| Groq | https://api.groq.com/openai/v1 | Llama 4 Maverick, DeepSeek R1 70B | Free tier available |
| Together AI | https://api.together.xyz/v1 | Llama 4 Maverick, DeepSeek R1 | $0.60–$3/M input |
| OpenRouter | https://openrouter.ai/api/v1 | GPT-4o, Claude Sonnet 4.5, many others | Varies by model |
| Custom | Your endpoint | Any OpenAI-compatible model | Depends on your setup |
3. How to Get an API Key (Step by Step)
Anthropic (Claude)
- Go to console.anthropic.com
- Sign up or log in with your Google account
- Click API Keys in the left sidebar
- Click Create Key → give it a name like "Grabblist"
- Copy the key (starts with
sk-ant-api03-...) - Paste it into Grabblist Settings with the Anthropic preset selected
OpenAI (GPT-4o, GPT-4o Mini)
- Go to platform.openai.com/api-keys
- Sign up or log in
- Click Create new secret key
- Copy the key (starts with
sk-proj-...) - In Grabblist Settings, select OpenAI preset and paste
DeepSeek
- Go to platform.deepseek.com/api_keys
- Sign up (phone verification required in some regions)
- Click Create API Key
- Copy the key (starts with
sk-...) - In Grabblist Settings, select DeepSeek preset and paste
DeepSeek is the most cost-effective option. V4 Pro costs $1.10/M input tokens — roughly 5× cheaper than GPT-4o for comparable quality.
Groq (Fastest Inference)
- Go to console.groq.com/keys
- Sign up with Google or GitHub
- Click Create API Key
- Copy the key (starts with
gsk_...) - In Grabblist Settings, select Groq preset and paste
Together AI
- Go to api.together.xyz/settings/api-keys
- Sign up
- Copy your API key
- In Grabblist Settings, select Together AI preset and paste
OpenRouter (Multi-Provider Gateway)
- Go to openrouter.ai/keys
- Sign up
- Click Create Key
- Copy the key (starts with
sk-or-v1-...) - In Grabblist Settings, select OpenRouter preset and paste
OpenRouter is a gateway to 200+ models. You get one API key and can switch between OpenAI, Anthropic, Google, Meta, and others without creating separate accounts. Pricing is pay-per-token with no subscriptions.
Custom / Self-Hosted (Ollama, LM Studio, vLLM)
- Run your local LLM server (e.g.,
ollama serve) - Note your endpoint URL (e.g.,
http://localhost:11434/v1) - In Grabblist Settings, select Custom (OpenAI-compatible)
- Paste your endpoint URL in the Base URL field
- Enter any string as the API key (e.g.,
ollama) — local servers often ignore it - Type your model name manually in the chat (e.g.,
llama3.1:8b)
4. Setting Up in the Dashboard (2 Minutes)
- Open Settings. Go to Dashboard → Settings. Scroll to "Getting Started" → "Set up your API key."
- Choose your provider. Pick from the dropdown: Anthropic, OpenAI, DeepSeek, Groq, Together AI, OpenRouter, or Custom.
- Paste your API key. The placeholder text shows the expected format (e.g.,
sk-ant-api03-...for Anthropic). - Set base URL (non-Anthropic only). The preset fills this automatically. Change it only if you use a proxy or custom endpoint.
- Click Save. A green checkmark confirms your key is connected.
- Start chatting. Open the AI assistant in the dashboard. Select your model from the dropdown (Claude Sonnet 4.5, GPT-4o, etc.) and start asking.
5. Choosing the Right Model
Once your API key is saved, the assistant shows a model dropdown. The list is fetched from your provider's API (for Anthropic and preset-matched providers) or shows sensible defaults.
By Task
| Task | Recommended Model | Why |
|---|---|---|
| Quick product lookups | Claude Haiku 4.5 / GPT-4o Mini / DeepSeek V4 Flash | Fast, cheap, sufficient for simple queries |
| Multi-step shopping research | Claude Sonnet 4.5 / GPT-4o | Best tool-use ability, follows complex instructions |
| Cross-store price comparison | Claude Sonnet 4.5 / DeepSeek V4 Pro | Handles large context windows, multiple search results |
| Product description writing | Claude Opus 4.5 / GPT-4o | Best prose quality for collection notes |
| Fastest response | Groq Llama 4 Maverick | Sub-500ms token generation on custom LPU hardware |
| Budget-conscious | DeepSeek V4 Flash / GPT-4o Mini | $0.27–$0.60/M input, excellent value |
6. MCP: Which Model Does It Use?
MCP (Model Context Protocol) connects Claude to Grabblist's tools. The model is determined by your Claude client:
- claude.ai (web): Claude Sonnet 4.5 by default. Pro users can switch to Opus 4.5 or Haiku 4.5 via the model picker.
- Claude Desktop: Uses the model selected in the app. Go to Claude → Settings → Model to change it.
- Claude Code (CLI): Uses the model specified with
--modelflag or the default Sonnet 4.5.
MCP does not require an API key — it uses your Claude account's built-in quota. If you have Claude Pro ($20/month), you get 5× more messages with MCP tools. See the MCP Setup Guide for the full walkthrough.
7. Troubleshooting
| Error | Likely Cause | Fix |
|---|---|---|
"Invalid API key" | Key is wrong or expired | Regenerate key on your provider's console and re-paste |
"Insufficient credits" | Provider balance depleted | Add funds on your provider's billing page |
"Model not available" | Your API key lacks access to that model | Check provider model availability; try a different model from the dropdown |
"Rate limit exceeded" | Too many requests in a short period | Wait 30 seconds; upgrade your provider tier for higher limits |
"Connection refused" (custom endpoint) | Local server not running or wrong URL | Verify ollama serve is running; check firewall; test with curl |
| Assistant doesn't use tools | Model lacks tool-calling support | Switch to a model with native function calling (Claude, GPT-4o, Llama 3.1+) |
| Blank model list | Provider API unreachable or key invalid | Check your internet; verify the base URL is correct; try a different preset |
8. Frequently Asked Questions
Do I need to pay for the AI models?
Grabblist does not charge for AI usage. You bring your own API key from your chosen provider. Most providers offer free credits for new accounts (OpenAI: $5, Anthropic: $5, DeepSeek: $2, Groq: free tier). After that, you pay the provider directly. MCP usage through ChatGPT or Claude uses the AI account you connect it from.
Which model should I pick?
For the dashboard assistant: Claude Sonnet 4.5 or GPT-4o for complex shopping research, Claude Haiku 4.5 or GPT-4o Mini for quick lookups. DeepSeek V4 Pro if you prefer the DeepSeek ecosystem. Groq Llama 4 Maverick for the fastest responses. For MCP: Claude Sonnet 4.5 is the default and recommended — it handles multi-step shopping workflows best.
Can I use a local LLM like Ollama or LM Studio?
Yes. Select "Custom (OpenAI-compatible)" in Settings and enter your local endpoint (e.g., http://localhost:11434/v1 for Ollama). Make sure your local model supports tool calling (function calling) — most modern models do.
Is my API key stored securely?
Yes. Your API key is encrypted at rest in Supabase and never exposed to the browser. It is only used server-side to forward requests to your chosen AI provider. You can remove it at any time from Settings.
Can I switch providers later?
Absolutely. Change your provider preset in Settings, paste the new API key, and hit Save. The assistant will use your new provider immediately. Your chat history stays intact.
What is MCP and how is it different from the dashboard agent?
MCP (Model Context Protocol) connects ChatGPT, Claude Desktop, or claude.ai directly to Grabblist's 21 shopping tools — web search, product extraction, price comparison, and more. The dashboard agent is the AI assistant inside the Grabblist web app. MCP uses the model selected in the AI client you connect.
Ready to set up? Go to Settings →
Need the MCP setup instead? MCP Setup Guide →