Models & limits
Ingot works with three kinds of models, side by side in one picker: Ingot's hosted models, your own provider keys, and local models.
Hosted models
Signed-in users call models through the Ingot model gateway. The aliases route to a chain of models, so if one provider has an outage your agent falls back to the next.
| Model | Plans | Good for |
|---|---|---|
Ingot Fast (ingot-fast) | all plans | quick edits, chat, autocomplete |
Ingot Smart (ingot-smart) | paid plans | everyday agent work |
Ingot Reasoning (ingot-reasoning) | paid plans | hard problems, planning |
You can also pick a specific model, for example GPT-4.1 and GPT-4.1 mini, Claude Sonnet 4.5 and Haiku 4.5, Gemini 2.5 Flash and Pro, DeepSeek Chat and Reasoner. Which ones you see depends on your plan. Auto lets Ingot choose.
Limits
| Free | Pro | Team | |
|---|---|---|---|
| Tokens per month | 1M | 20M | 40M |
These allowances are provisional and may change before 1.0; your exact numbers are always in the dashboard and the IDE status bar. A single request can carry up to 200k tokens, and rate limits are counted per minute.
| Error | Meaning |
|---|---|
402 token_allowance_exhausted | monthly tokens used up |
402 spend_cap_reached | spend cap reached |
429 rate_limited | too many requests or tokens this minute; wait and retry |
403 model_not_in_plan | that model needs a paid plan |
Bring your own key (BYOK)
In the IDE, turn on ingot.models.byok and add your keys; the IDE stores them in the OS keychain. Keys can also be saved to your account in the dashboard under API keys. Supported families: Claude, GPT, Gemini, DeepSeek, and custom for any endpoint that speaks the standard chat-completions API (https, or http on localhost). Keys are encrypted at rest (AES-256-GCM) and never shown again; the dashboard only shows the last four characters. Using your own keys is free on every plan.
Local models
Ingot talks to Ollama at http://127.0.0.1:11434 by default. Change it with the ingot.models.ollamaUrl setting (Settings › Tools › AI Models). With local models nothing leaves your machine.
Autocomplete
Inline completions are controlled by ingot.autocomplete.enabled, ingot.autocomplete.model and ingot.autocomplete.debounceMs (Settings).