Rixoffice does not lock you into one AI vendor. If you already pay for model access — or you want a model the built-in service does not offer — you can plug your own key into Settings and the entire AI stack, agents and skills included, runs through your provider instead. Your keys stay on your device, your usage is never metered by Rixcart, and the composer behaves exactly as it does on the default provider.
Supported providers
| Provider | What you supply | Typical reason to choose it |
|---|---|---|
| OpenRouter | OpenRouter API key | One key, access to many models |
| Anthropic | Claude API key | Long-document reasoning and writing |
| OpenAI | OpenAI API key | GPT models under your own billing |
| Google Gemini | Gemini API key | Fast multimodal work |
| DeepSeek | DeepSeek API key | Cost-effective reasoning |
| Custom | OpenAI-compatible base URL | Self-hosted or niche providers |
OpenRouter deserves a special mention for individuals: it is one account and one key with access to a broad model catalog, which makes it the easiest way to experiment. The Custom option is the escape hatch for everything else — if your server speaks the OpenAI API shape, Rixoffice can talk to it. That covers llama.cpp server, LM Studio, vLLM and most self-hosted gateways in one stroke, because the OpenAI API format has become the lingua franca of model serving.
Adding a key step by step
- Open Settings and go to the AI provider section.
- Pick your provider from the list.
- Paste your API key. It is stored locally on your device and used only to call the provider directly from your machine.
- Choose the model you want from the model list.
- Press Test. Rixoffice sends a tiny request to verify the key and shows a clear pass or fail.
- Open any document and try the composer to confirm end-to-end.
The same flow covers the Custom option, with one extra field: the base URL of an OpenAI-compatible server, for example the local address of llama.cpp server, LM Studio or vLLM. If your server needs a key, add it; if not, leave it empty and test. Connection testing is built into the settings screen precisely so you never debug a broken key from inside a document — fail fast at setup, work smoothly after.
What happens to your key
The key never leaves your device except in the direct request to the provider you chose:
- Stored locally in the app's own settings storage, not on Rixcart servers.
- Used only for calls from your machine to your provider's endpoint.
- Never metered. Rixcart counts nothing and bills nothing for BYO usage — see AI credits explained for how that differs from built-in Rix AI.
If you rotate a key at the provider, just paste the new one over the old in Settings and press Test again. Keys survive app updates and restarts, because they live in your user profile rather than inside the application.
What BYO unlocks
Bring-your-own is not only about price. It also unlocks plan features that depend on provider access:
- On the Pro API & Local plan and above, BYO keys plus Ollama give you unlimited unmetered AI, subject only to your provider quota or your own hardware.
- Teams can standardize on one provider and one billing relationship instead of per-seat credits.
- You can point different machines at different providers — your laptop at Anthropic, your desktop at a local server — and the composer behaves identically everywhere.
That last point matters more than it sounds. Because agents, skills and block-level editing are provider-independent, a mixed setup is not a compromise; it is just configuration.
Choosing a model after connecting
The model list comes from your provider, and the right pick depends on the job:
- For writing and editing in RixDocs, a strong general model beats a tiny fast one — rewrites and whole-document work reward quality.
- For chat and analysis in RixSheets, mid-tier models are usually enough and keep costs down.
- For deck generation in RixSlides, pick a capable model; the pipeline plans, scripts, audits and QC-fixes decks, and better models produce fewer fix loops.
If your provider offers several tiers, start with the strongest model you can afford for a day of real work, then step down where the tasks turn out to be easy. Usage dashboards at the provider will show you exactly where the tokens go.
Switching between built-in and BYO
Nothing is exclusive. You can use built-in Rix AI on the road and your own key at your desk, changing provider in Settings whenever it suits. Credits are only consumed while built-in Rix AI is selected; the moment you switch to your key, metering stops. And if a provider has a bad day, flipping back to the built-in proxy takes the same minute as flipping away from it.
This flexibility also changes how you evaluate providers. Instead of committing to one vendor for a year, you can run the same real workload across two providers for a week each and let your own documents decide. The composer keeps score quietly: whichever model consistently produces edits you accept on the first try is the one to keep.
What to explore next
- Ollama local models — the fully offline version of BYO.
- AI provider setup — comparing every provider option in one table.
- Pro API & Local plan — the plan built around keys and local models.