AI
AI Providers
Connect RaiseDB to Claude, OpenAI / OpenAI-compatible, Gemini, or Ollama.
RaiseDB supports four API protocols (one of which also serves OpenAI / DeepSeek / Qwen / etc.). Pick at Settings → AI, or from the panel on first use.
Supported providers
| Provider | API style | Default base URL | Native tools |
|---|---|---|---|
| Anthropic Claude | Anthropic Messages | https://api.anthropic.com/v1 | yes |
| OpenAI / OpenAI-compatible | OpenAI /chat/completions | https://api.openai.com/v1 | yes |
| Google Gemini | generateContent | https://generativelanguage.googleapis.com/v1beta | yes |
| Ollama | Native /api/generate (NDJSON stream) | http://localhost:11434 | no (falls back to ReAct) |
The OpenAI API style covers OpenAI, DeepSeek, Qwen, and any other OpenAI-compatible endpoint (vLLM, LM Studio, …). The same
OpenAIProviderpowers them all — set the base URL to point elsewhere.
Configure
Settings → AI → Add Provider. Per-provider fields:
- API key (or leave blank for no-auth local endpoints)
- Base URL (override for self-hosted or proxies)
- Temperature (default 0)
- Max tokens (some call sites currently hardcode 4096)
- Model, context window size, max output
- Tool-use / thinking flags (Claude, Gemini)
Test connection
Hit Test before saving to confirm credentials.
Choosing a model
- SQL generation: any mid-tier model works well. Claude / GPT-4o are best.
- Agents / long reasoning: pick a model with strong tool use. Claude Sonnet/Opus and GPT-4o are recommended.
- Fully offline: Ollama with a local model. RaiseDB will run the agent via ReAct text-tool-calling since Ollama doesn't expose native tools.
API key storage
Keys are encrypted alongside connection credentials. See the security model.
Quota
| Tier | AI requests / day |
|---|---|
| Free | 5 |
| Pro | Unlimited |
See Pricing.