AI models
Nomima doesn’t lock you into anyone’s AI. Run open models locally with no key at all, or bring your own key for Claude, GPT, Gemini and a dozen more clouds. You pick the provider; Nomima just talks to it.
Your models, your choice. AI in Nomima (the Summon assistant) is free — run a model locally for fully offline use, or supply an API key from any supported cloud. Everything else in Nomima — including the Knowledge Graph, Smart Views and search — works without any AI at all.
Supported providers
Providers come in three groups — the same grouping you’ll see in the picker. Local providers run on your own machine and need no key; the Cloud and OpenAI-compatible groups are bring-your-own-key.
Local — offline, no key
| Provider | What it is | Needs internet? |
|---|---|---|
| Ollama | Run open models locally. Stays offline — no key, no cloud. | No — runs on your machine |
| LM Studio | Local models via LM Studio’s server. Offline, no key. | No — runs on your machine |
Fully offline AI. Point Summon at Ollama or LM Studio and no note ever leaves your Mac — the model runs locally. Summon needs a tool-capable model; the “Check” button confirms a local one qualifies.
Live status & capability check. For a local provider Nomima shows a green/red server status pill — in Settings and in the Summon header — and lists the models you actually have installed. Picking one runs a one-time capability probe and shows an honest badge: Summon-ready, Summon · basic (a small model), AI blocks only (too weak to drive Summon), or Not checked. If the server is down or the model can’t drive tools, Summon’s composer is disabled with a clear nudge — and even a weak model can still answer from your notes through a grounded, no-tools fallback.
Cloud — bring your own key
| Provider | Models | Needs internet? |
|---|---|---|
| Anthropic | Claude — Opus, Sonnet, Haiku | Yes — calls Anthropic with your API key |
| OpenAI | GPT-5.1, GPT-4o and the o-series | Yes — calls OpenAI with your API key |
| Google Gemini | Gemini 3 Pro and the Flash tiers | Yes — calls Google with your API key |
| AWS Bedrock | Dozens of models on AWS — Claude, GPT-OSS, DeepSeek, Qwen, Mistral and more — from a single Bedrock API key | Yes — calls AWS with your Bedrock API key |
AWS Bedrock — just paste the key
Bedrock needs a Bedrock API key, not an AWS access key and secret. Create one in the AWS console under Amazon Bedrock → API keys (a long-term key is easiest), then paste it into the single field in Settings → Summon → AI model → AWS Bedrock.
That’s the whole setup — there’s no region to choose. Nomima works out which AWS region your key belongs to on its own and remembers it, then fills the model list live from your own account, so you only ever see models you actually have access to. Pick one and Summon uses it like any other provider.
Two things worth knowing:
- Enable the models you want in AWS first. The list only shows models your account has access to, so if something you expected is missing, grant it model access in the Bedrock console and reopen the page.
- If you replace the key, paste the new one. An old or revoked key is the usual reason Bedrock suddenly stops working — pasting the current key and saving fixes it.
If you need Bedrock pinned to a specific region (for data-residency reasons, say), use the change link next to the detected region and enter it yourself.
OpenAI-compatible clouds — bring your own key
These share one OpenAI-compatible adapter, so any of them is just a key away.
| Provider | What it is | Needs internet? |
|---|---|---|
| Groq | Very fast inference for open models. | Yes — calls Groq with your API key |
| OpenRouter | One key, hundreds of models across providers. | Yes — calls OpenRouter with your API key |
| Mistral | Mistral’s open + premier models. | Yes — calls Mistral with your API key |
| DeepSeek | DeepSeek’s chat + reasoning models. | Yes — calls DeepSeek with your API key |
| xAI (Grok) | Grok models from xAI. | Yes — calls xAI with your API key |
| Together AI | Open models hosted on Together AI. | Yes — calls Together with your API key |
Setting up a provider
- Open Settings → Summon → AI model. It’s a two-pane view: the provider list (grouped Local / Cloud / OpenAI-compatible) on the left, the selected provider’s detail on the right.
- Pick a provider. For a local one (Ollama, LM Studio) there’s no key — Nomima probes for the models you have installed. For a cloud one, paste your API key.
- Choose a model. For local providers this is whatever you’ve pulled; for clouds it’s the models that provider offers.
- Confirm it works — a small test call checks the credentials, and Check verifies a local model is tool-capable enough for Summon.
- Set your default provider; that’s what Summon uses unless told otherwise.
How your keys are protected
- Keys are stored in the macOS Keychain, not in a config file. In the UI you’ll only ever see the last few characters.
- Local providers need no key at all — nothing to store, nothing to leak.
- Keys are used for exactly one thing: calling the provider you configured, when you invoke AI. Nomima has no server of its own and never proxies your traffic.
- Delete a key any time from the same panel.
What actually uses AI
| Feature | Uses your AI model? |
|---|---|
| Summon assistant | Yes — the only feature that calls your AI model |
| Knowledge Graph, briefs, themes | No — built with on-device analysis |
| Search, Smart Views, Home agenda | No |
| Date detection, smart tags | No |
If you never configure a provider, Nomima simply never makes an AI call. The app is fully functional as a private, offline notebook.
If something fails
Nomima tells you exactly what went wrong rather than failing silently: an invalid key (fix the key), quota exhausted (top up with your provider), rate limited (wait a moment and retry), or a network error (check your connection).