Skip to content
NoSqlStudio
Documentation

AI Providers

Bring your own LLM. Register Anthropic, OpenAI, Gemini, Ollama, Azure OpenAI, or any OpenAI-compatible endpoint. Keys stay encrypted in the OS keychain via Electron safeStorage. Switch models in 1 click.

What it is

A registry workspace where you add LLM providers. Each entry stores a label, kind (anthropic / openai / gemini / ollama / azure / custom), endpoint URL (if applicable), model id, and an API key. Keys never leave the DBA workstation — encrypted at rest with safeStorage, decrypted on demand for each request.

When to use

Configure once per workstation. Then any AI-driven surface — Cosmos AI Advisor, query rewriter, schema migration assistant, AI-anomaly detector — uses the active provider. Switch from Claude to GPT-4 to local Ollama in 1 click, no rebuild.

How to open

  1. In the NoSqlStudio desktop app, open the Tools menu → AI Providers, or press Ctrl+Alt+Shift+I.
  2. The workspace lists configured providers. Click Add to register a new one.
  3. Pick the kind from the dropdown. Fill the required fields. The endpoint URL field auto-populates for managed providers; custom kinds expect any OpenAI-compatible /v1/chat/completions endpoint.

Registering a provider

  1. Anthropic — paste the API key starting with sk-ant-. Pick a model (claude-opus-4-7 / claude-sonnet-4-6 / claude-haiku-4-5).
  2. OpenAI — paste the API key starting with sk-. Pick a model (gpt-4, gpt-4o, gpt-4-turbo, etc.).
  3. Gemini — paste the API key. Pick a model (gemini-1.5-pro / gemini-1.5-flash).
  4. Ollama (local) — leave the API key empty. Endpoint defaults to http://localhost:11434. Pick a pulled model (llama3 / mistral / etc.).
  5. Azure OpenAI — paste the endpoint URL, deployment name, API version, and the key.
  6. Custom OpenAI-compatible — paste the base URL (e.g. https://api.together.ai/v1) and the API key. Useful for OpenRouter, Together AI, Anyscale, vLLM, and self-hosted endpoints.

Switching the active provider

Click "Set active" on any registered provider. The new selection is broadcast across all open workspace tabs that consume AI — there is no app restart and no per-feature override; one global switch.

Shadow validation guarantee

docsAiProviders.shadowBody

Limitations (v1)

  • Keys are encrypted with Electron safeStorage — encryption strength depends on the OS keyring (Keychain on macOS, DPAPI on Windows, libsecret on Linux). If your threat model needs hardware-key isolation, do not use desktop AI on shared hosts.
  • No automatic provider fallback if a primary fails mid-request. You can wire a fallback chain manually but the UI exposes only a single active provider.
  • Local Ollama performance depends on your hardware. Heavy schema analysis on llama3-70b on a laptop will be slow — pick a smaller model for interactive flows.