Skip to main content
Every agent in Reliant needs a model, and you choose where it comes from. You can use several sources at once. When a workflow asks for a kind of model (see Model Configuration), Reliant picks the best one from the sources you’ve connected. Reliant’s servers make all model calls, never your device, whichever source you use. Your provider settings are saved to your account, so they follow you between the web app, the desktop app and every machine you use. You manage all of this in Settings → AI, which has these tabs: The same choices appear during first-run setup, on the Model step of the Quick Start.

​
Reliant AI

Reliant AI is the optional managed provider. There is no API key to create and no provider account to set up: you add credit to a wallet and Reliant pays for the model calls on your behalf. Usage is billed at the provider’s price, with no markup on the model itself. The only extra charge is a 5% service fee on wallet top-ups, which is shown as its own line before you pay. Credit from a coupon carries no fee. Your credit doesn’t expire. Manage your balance and see what you’ve spent in Settings → Billing, and your spend this period in Settings → AI → Reliant AI. Because usage is paid from your credit, your spending is bounded by what you’ve added. If you turn on automatic top-ups on the Billing page, you set a monthly limit for them. For how requests to Reliant AI are handled, see Privacy & Security.

​
Sign in with a subscription you already pay for

If you already pay for Claude, ChatGPT or GitHub Copilot, you can use that subscription instead of paying for API usage. Open Settings → AI → Your providers, choose the provider and connect it. Usage then counts against that subscription under your account with the provider, and the provider’s own terms and limits apply.

​
Signing in from the web app

Claude Code, Codex and Antigravity sign-in finish by redirecting your browser back to localhost, so something on your own computer has to catch that redirect. In the desktop app this happens automatically. In the web app, Reliant shows a short panel with two steps: install the reliant command line tool (see Installation), then run this in a terminal and leave it running:
The panel notices the helper within a couple of seconds and shows the sign-in button. The helper only passes the authorization code back to your browser; it stores no credentials, and you can stop it once you’re connected. To use a different port, run reliant auth serve --port 9999. GitHub Copilot doesn’t need the helper, because it uses a code you enter on GitHub’s site.

​
Use your own API key

Choose a provider in Settings → AI → Your providers, paste its key, and click Test Connection to check it before saving. OpenRouter can serve the Anthropic, OpenAI and Google models Reliant already knows about. Other models on OpenRouter aren’t supported unless you add them as custom models. Keys are stored encrypted with your account. Reliant doesn’t read API keys from environment variables or from a .reliant.json file, so setting ANTHROPIC_API_KEY in your shell has no effect; use the Settings page. Keep keys out of your repositories as you normally would. If you use your own key or login, you pay the provider directly and their rate limits apply to you. See Anthropic, OpenAI, Google Gemini and OpenRouter for current limits and pricing.

​
Local models

You can run models on your own hardware with Ollama, LM Studio or any server that speaks the OpenAI API (llama.cpp and vLLM work too) and use them from Reliant. Because Reliant’s servers make the model calls, a local model is reached through a daemon running on the machine that can see the model server, usually your own computer. The desktop app runs one for you; otherwise start one with reliant daemon start (see Your own computer). A cloud machine can’t see a server on your laptop, so connect your laptop as a machine, too. Then open Settings → AI → Custom & local models. Reliant looks for model servers on that machine and lists the models it finds. If yours isn’t detected, use Add endpoint and enter its address, for example: Agents need a long context window. Ollama defaults to a small one, and the page tells you when it’s too small to be useful; the fix is to raise Context length in the Ollama app’s settings. The prompt and the model’s reply still pass through Reliant’s servers and are stored in your chat; only the inference runs on your hardware. For defining your own models and tags, see Model Configuration.

​
Which models can I use?

The models available to you depend on the sources you’ve connected: a Claude login unlocks Claude models, an OpenAI key unlocks OpenAI models, and so on. The Models Reference lists every model Reliant knows about, which providers can serve it, and its tags. That page is generated from the product itself, so it is the authoritative list; model names change quickly and aren’t repeated here.

​
Troubleshooting

​
”Invalid API key”

Check that you copied the whole key with no spaces, that it belongs to the right provider (an Anthropic key for Claude models, for instance), and that it hasn’t been revoked. Test Connection shows the provider’s own error.

​
The sign-in button doesn’t appear (web app)

The local helper isn’t running. Run reliant auth serve in a terminal and keep it open; see Signing in from the web app.

​
Rate limits and usage limits

These come from your provider, not Reliant. When a subscription or key hits a usage limit, the run doesn’t fail: the chat shows which provider is limited and when it resets, and carries on by itself at the reset. Retry now tries again straight away, Stop ends the run, and if another provider you’ve connected can serve the chat, the card offers to use its model in the meantime. Check your limits in the provider’s dashboard if this happens often.

​
A model you expect isn’t offered

Models are offered only for sources you’ve connected. Check Settings → AI → Your providers, and see the Models Reference for which providers serve it.