AI Providers & Key Encryption

PromptBranch connects directly to major cloud AI providers and any OpenAI-compatible local endpoint. API keys are encrypted with your operating system's native keychain before they are stored.

---

Supported Providers

The Connect a provider… dialog lists every provider in the community-maintained models.dev catalog:

Provider Type	Examples	Execution Driver
OpenAI	GPT family, o-series reasoning models	openai
Anthropic	Claude family	anthropic
Google	Gemini family	google
OpenAI-compatible	Groq, Mistral, Together AI, xAI, DeepSeek, Perplexity, …	openai-compatible
Custom endpoint	Ollama, LM Studio, vLLM, LocalAI — anything speaking the OpenAI HTTP API	openai-compatible

Popular providers (OpenAI, Anthropic, Google) are pinned at the top of the list; the rest follow alphabetically with model, context-window, and pricing data from the locally cached catalog when available.

---

Connecting a Provider

1. In PromptBranch, open Settings (⌘, / Ctrl+,) and navigate to AI Providers.
2. Click Connect a provider….
3. Select a provider from the catalog list — or Custom OpenAI-compatible provider for a local or self-hosted endpoint.
4. Enter your API Key.
5. Click Connect. PromptBranch verifies authentication with a minimal test request ("Reply with: ok") before saving the connection.

One-Click Environment Key Detection

If standard provider API keys are defined in your shell environment (OPENAI_API_KEY, ANTHROPIC_API_KEY, or GOOGLE_GENERATIVE_AI_API_KEY), PromptBranch automatically detects them and displays a Use environment key button for instant setup.

---

Connecting Local Endpoints (Ollama, LM Studio)

PromptBranch fully supports local, private, offline inference via any OpenAI-compatible HTTP endpoint.

Ollama Setup

1. Ensure Ollama is running locally (ollama serve).
2. In Settings → AI Providers, click Connect a provider… and choose Custom OpenAI-compatible provider.
3. Set Base URL: http://localhost:11434/v1.
4. Pick or enter a Test model you have pulled locally, then click Connect.
5. After connecting, open Manage models and add your installed model IDs (e.g. llama3.1:8b, deepseek-r1:8b, qwen2.5-coder:7b) so they appear in the run picker.

LM Studio Setup

1. In LM Studio, start the Local Server on port 1234.
2. Connect a custom OpenAI-compatible provider with Base URL http://localhost:1234/v1.
3. Add your loaded model IDs under Manage models.

---

Security & OS Keychain Encryption (safeStorage)

PromptBranch treats API keys with uncompromising security:

User->>Renderer: Enters API Key Renderer->>Main: Send once via the validated preload IPC bridge Main->>Keychain: Encrypt plaintext key Keychain-->>Main: Return opaque ciphertext blob Main->>DB: Store only the encrypted key blob Note over DB,Renderer: Plaintext key is NEVER stored in SQLite<br/>and NEVER sent back to Renderer

• safeStorage: API keys are encrypted at rest using your operating system's native cryptographic keychain (macOS Keychain, Windows DPAPI, or Linux Secret Service).
• Renderer Isolation: The renderer necessarily holds a key while you type it, then sends it once to the main process. Stored keys are never returned to or decrypted in the renderer.
• Export & Sync Safety: Library JSON exports can include provider metadata and model declarations, but the api_key_enc field is cleared. Multi-device sync also redacts that field, so keys must be configured separately on each device.

---

The Model Catalog (models.dev)

PromptBranch integrates with the community-maintained models.dev catalog:

• Catalog Model Data: Context window sizes, output limits, and input/output pricing per million tokens when models.dev supplies them.
• Offline Cache: Catalog data is cached locally. If your machine is offline, PromptBranch keeps operating with the cached catalog; a failed refresh never clobbers a good cache.
• Managing Model Visibility: In Settings → AI Providers, click Manage models to hide models you don't use from the run picker, add custom model IDs for local endpoints, or bring hidden models back.