AI integrations

Bedrock’s AI providers are integration plugins you enable in Settings → Integrations (⌘,). Each has a switch and, when on, its configuration. A Default send target at the top decides what new prompt cards use.

All integrations run with your own keys or local server — Bedrock adds no service of its own.

Clipboard

Always on, no configuration needed. Selecting Clipboard as a target copies the assembled payload so you can paste it anywhere.

Claude CLI

Runs your local Claude Code binary headlessly and captures the reply in the prompt card. This is the only target that can read and edit files in your project directory.

Setup: 1. Install Claude Code: npm install -g @anthropic-ai/claude-code, then claude login. 2. In Bedrock: Settings → Integrations → Claude CLI → toggle On. 3. Leave Binary path blank to use claude from your PATH, or set an absolute path (e.g. ~/.local/bin/claude). 4. Set a Permission mode: - plan — read-only; Claude explains what it would do. - acceptEdits — Claude edits files without prompting. - bypassPermissions — full access with no prompts. Use with care. 5. Optionally set a Model override (e.g. claude-opus-4-8).

Claude (Terminal)

Opens a new Terminal window in the project directory, starts Claude there, and seeds it with the prompt as the first message — giving you an interactive session. Reuses the Claude CLI settings above. macOS only.

Good for tasks where you want to guide the session after the initial prompt — e.g. debugging where you iterate with follow-up questions.

Claude API

Calls Anthropic’s Messages API directly with your API key. Returns text — it cannot read or modify files.

Setup: 1. Get an API key from console.anthropic.com. 2. Settings → Integrations → Claude API → toggle On. 3. Paste your API key and set a Model (e.g. claude-sonnet-4-6).

Gemini API

Calls Google’s Gemini API. Returns text only.

Setup: 1. Get an API key from aistudio.google.com. 2. Settings → Integrations → Gemini API → toggle On. 3. Paste your API key and set a Model (default: gemini-2.0-flash).

LM Studio (local)

Runs a model on your machine via LM Studio — fully offline, no key, no data leaves your computer. Also powers the in-editor Writing assist.

Setup: 1. Download and install LM Studio. 2. In LM Studio: open the Developer tab → Start server. 3. Load a model in the Models tab. 4. In Bedrock: Settings → Integrations → LM Studio → toggle On. 5. Set the Server URL (default http://localhost:1234/v1). 6. Click Refresh to fetch available models, then pick one.

If a run fails with “couldn’t load the model”, pick a smaller or different model. Bedrock uses Tauri’s native HTTP layer — you do not need to enable CORS in LM Studio.

Choosing a target

Turning integrations off

Toggle a provider’s switch to disable it. A disabled target disappears from the prompt board target list. If you try to run a prompt whose target is disabled, Bedrock shows a message pointing back to Settings.