AI Providers
MemryLab has 9 provider presets, 8 of them with free tiers (Ollama runs locally and needs no key), plus a custom OpenAI-compatible endpoint. You can switch providers at any time without losing data. LLM and embedding providers can be configured independently — for example, use Gemini for analysis and Ollama for private local embeddings.
Recommended Local Models (Ollama)
For local inference, pick models based on your GPU VRAM. Embedding model is always nomic-embed-text.
| VRAM | LLM Model | Speed | Install |
|---|---|---|---|
| 4 GB | llama3.2:3b | ~40 tok/s | ollama pull llama3.2:3b |
| 8 GB | llama3.1:8b | ~35 tok/s | ollama pull llama3.1:8b |
| 12 GB ★ | qwen2.5:14b-instruct-q5_K_M | ~25 tok/s | ollama pull qwen2.5:14b-instruct-q5_K_M |
| 16 GB+ | qwen2.5:32b-instruct-q4_K_M | ~15 tok/s | ollama pull qwen2.5:32b-instruct-q4_K_M |
Larger models produce better belief extraction and contradiction detection. For no-GPU setups, use Gemini Flash (free) or Groq (free) as cloud providers.
💡 Pro Tip: Mix Providers
In Settings → Embedding Provider, you can use a different provider for embeddings vs LLM. The best privacy setup: use Gemini Flash (free, fast) for analysis + Ollama (local) for embeddings. This way your search index stays fully private while analysis uses a powerful cloud model.
Quick Comparison
| Provider | Free Tier | Details |
|---|---|---|
| Ollama (Local) | Yes | Runs on your own hardware, no limits and no internet needed |
| OpenRouter | Yes | Free models from several labs, about 20 requests/min and 200/day per model |
| Groq | Yes | Free tier: 30 requests/min, about 1,000 requests/day on the 70B model |
| Google Gemini | Yes | Free tier: 10 requests/min on Flash, 5 on Pro, no credit card. Includes embeddings |
| Cerebras | Yes | Free tier: 1M tokens/day, very fast inference |
| Mistral | Yes | Free experiment plan: 1 request/sec, up to 1B tokens/month. Includes embeddings |
| SambaNova | Yes | Free tier: 10-30 requests/min depending on the model |
| Cohere | Yes | Free tier: 1,000 API calls/month. Includes embeddings |
| Claude (Anthropic) | Paid | Pay-per-use API key, no free tier |
| Custom (OpenAI-compatible) | Varies | Any OpenAI-compatible endpoint (LM Studio, vLLM, OpenAI and others) |
Ollama (Local)
1. Install Ollama from ollama.com/download
2. Pull a model: ollama pull llama3.1:8b
3. Pull the embedding model: ollama pull nomic-embed-text
4. In MemryLab: Settings → AI Provider → select "Ollama (Local)"
5. No API key is neededOpenRouter
1. Go to openrouter.ai and create an account
2. Generate an API key
3. In MemryLab: Settings → AI Provider → select "OpenRouter"
4. Paste your API key — a free model is pre-selectedGroq
1. Go to console.groq.com and create an account
2. Generate an API key from the dashboard
3. In MemryLab: Settings → AI Provider → select "Groq"
4. Paste your API key and test connectionGoogle Gemini
1. Go to aistudio.google.com and sign in
2. Create an API key
3. In MemryLab: Settings → AI Provider → select "Google Gemini"
4. Paste your API key and test connectionCerebras
1. Go to cloud.cerebras.ai and sign up
2. Create an API key from the dashboard
3. In MemryLab: Settings → AI Provider → select "Cerebras"
4. Paste your API key and test connectionMistral
1. Go to console.mistral.ai and create an account
2. Generate an API key
3. In MemryLab: Settings → AI Provider → select "Mistral"
4. Paste your API key and test connectionSambaNova
1. Go to cloud.sambanova.ai and sign up
2. Create an API key
3. In MemryLab: Settings → AI Provider → select "SambaNova"
4. Paste your API key and test connectionCohere
1. Go to dashboard.cohere.com and sign up
2. Copy your trial API key
3. In MemryLab: Settings → AI Provider → select "Cohere"
4. Paste your API key and test connectionClaude (Anthropic)
1. Go to console.anthropic.com and sign in
2. Create an API key
3. In MemryLab: Settings → AI Provider → select "Claude (Anthropic)"
4. Paste your API key and test connectionCustom (OpenAI-compatible)
1. In MemryLab: Settings → AI Provider → Custom OpenAI-compatible endpoint
2. Enter the base URL (e.g., http://localhost:1234/v1)
3. Enter an API key if required (or leave blank for local)
4. Select a model name and test connectionAPI Key Security
MemryLab stores API keys in your operating system's secure keychain — never in plain text files, environment variables, or the SQLite database.
- Windows: Windows Credential Manager
- macOS: Keychain Access
- Linux: libsecret (GNOME Keyring / KWallet)