Provider Setup
Configure the LLM provider used by the inline AI menu in the editor. Open from Settings → AI Inline.
Supported Providers
| Provider | Runs | Needs |
|---|---|---|
| Ollama | Locally on your machine | Ollama installed and running |
| OpenAI | OpenAI cloud | API key |
| Anthropic | Anthropic cloud | API key |
| Google Gemini | Google AI cloud | API key |
| Custom | Any OpenAI-compatible endpoint | Base URL; API key if the endpoint requires one |
Each built-in provider has its own model presets in the dropdown.
Ollama (Local)
Best when you want privacy, no metering, and don't mind running an extra background process.
Setup
- Install Ollama on your system.
- Pull a model:
ollama pull llama3.2(or any chat model you prefer). - In memrynote, set Provider: Ollama.
- Base URL: defaults to
http://localhost:11434/v1— adjust if Ollama runs on a different port or remote machine. - Model: pick from the dropdown (presets) or type a model name you've pulled locally.
- Click Test Connection.
Notes
- memrynote talks to Ollama's native
/api/chatendpoint and automatically setsnum_ctx: 8192per request. You do not need to setOLLAMA_CONTEXT_LENGTHorOLLAMA_NUM_PARALLEL— the context window is managed for you. - No API key required for default local Ollama.
- Token costs: zero, but RAM usage scales with model size.
OpenAI
Setup
- Get an API key from platform.openai.com.
- Provider: OpenAI
- Paste the API key into the input. Toggle visibility with the eye icon.
- Model: pick from presets (e.g.
gpt-6.1-sol,gpt-6-luna). - Click Test Connection.
Notes
- Keys are stored locally in the vault (encrypted at rest).
- Keys are never sent to the memrynote sync server.
- Each AI call hits OpenAI directly from your device.
Anthropic
Setup
- Get an API key from console.anthropic.com.
- Provider: Anthropic
- Paste the API key.
- Model: pick a Claude preset (e.g.
claude-sonnet-5-5). - Click Test Connection.
Notes
Same key-storage and privacy properties as OpenAI.
Google Gemini
Setup
- Get an API key from Google AI Studio.
- Provider: Google Gemini
- Paste the API key.
- Model: pick a Gemini preset (e.g.
gemini-3.8-flash) or type any model your key can access. - Click Test Connection.
Notes
- Uses the Gemini API directly with your key; it is not tied to a Gemini CLI or Code Assist subscription.
- Same key-storage and privacy properties as OpenAI.
Custom (OpenAI-compatible)
Connect any service that speaks the OpenAI chat completions API, such as OpenRouter, Together, Groq, or a self-hosted vLLM or LM Studio server.
Setup
- Provider: Custom (OpenAI-compatible)
- Base URL: the endpoint root that serves
/chat/completions, for examplehttps://openrouter.ai/api/v1. - API Key: paste it if the endpoint requires one. Leave it empty for an endpoint without auth.
- Model: type the model ID, for example
openai/gpt-4o-minion OpenRouter. When the endpoint answers/models, its catalogue appears as suggestions while you type. - Click Test Connection.
Notes
- The model must support tool calling and JSON schema output. The inline menu edits through tool calls, and writing tools ask for structured answers. On OpenRouter, check the model's supported parameters for
toolsandstructured_outputs. - Errors name the endpoint's host, for example "Your openrouter.ai API key was rejected".
- Same key-storage and privacy properties as OpenAI. Requests go from your device to the endpoint you entered.
- Test Connection starts the local bridge but does not call the endpoint. A wrong key shows up on the first AI request.
Test Connection
The button verifies:
- The base URL is reachable
- The API key authenticates (for OpenAI / Anthropic / Google Gemini)
- The selected model exists and is accessible
A status indicator (green / yellow / red) shows the most recent test result.
Switching Providers
You can change provider at any time. Memory of past prompts is local — switching doesn't carry conversation state.
memrynote keeps one local inline-AI bridge running for the active provider settings. Re-opening the editor or another window reuses that bridge when settings are unchanged; changing provider, model, base URL, or API key restarts it with the new configuration. Open notes pick up the change, including turning inline AI on, without restarting the app.
Where Keys Live
API keys are stored locally, in the vault, encrypted with the vault key. They:
- Never leave the device unencrypted
- Never appear in sync items
- Are accessible to memrynote's main process only (renderer reads via IPC)
Rotating a key in the provider's dashboard requires updating it here too.
Agent Chat built-in model
Agent Chat's built-in model connection is configured separately, under Agent Permissions. It accepts any OpenAI-compatible base URL, model name, and API key. Two controls change how a reasoning model such as DeepSeek thinks:
- Effort in the prompt bar sets reasoning depth per turn. Default sends nothing extra; High and Max send
reasoning_effort: "high"or"max". - Thinking in the connection settings. Default sends nothing extra; Off sends
thinking: {"type": "disabled"}for faster, cheaper answers.
Both controls are hidden for the Ollama preset. Leave them at Default for a server that does not know these fields, since some servers reject a request with fields they do not recognise.
The built-in model connection has the memrynote vault tools only: no shell, no files outside the vault, and no web search. Agent Backends compares it with Claude Code, Codex and Antigravity.
Troubleshooting
| Symptom | Likely cause |
|---|---|
| "Cannot connect" with Ollama | Ollama not running; check http://localhost:11434/v1/models |
| "Unauthorized" with a cloud provider | Wrong key, expired, or rate-limited |
| "Model not found" | Model name typo or not pulled (Ollama) / not enabled in your account |
| Test passes but inline menu doesn't appear | AI Inline toggle off in settings |