More Providers
Mistral, xAI, LiteLLM, and the rest of the roster
Beyond the providers with dedicated pages, Nexgile Code supports the following entries in the provider dropdown. Each works the same way: open Settings → Providers, pick the entry, supply the credential it asks for, choose a model, and save it as a configuration profile. Keys are always stored in the editor’s Secret Storage.
The model rosters below are the ones this build actually offers. Nx IDE is the TAA-compliance edition, so models of Chinese origin are filtered out of every catalog — including the live catalogs that gateways such as LiteLLM, Requesty, Unbound, Vercel AI Gateway, and Poe fetch at runtime. A gateway’s own web console may therefore list models that the dropdown does not. See Providers Overview.
Mistral
Section titled “Mistral”European AI lab with a strong code-focused line. Create a key in the Mistral console at console.mistral.ai. The default model is codestral-latest, Mistral’s code-specialist; the roster also includes devstral-medium-latest, magistral-medium-latest (reasoning), mistral-large-latest, mistral-medium-latest, mistral-small-latest, the compact ministral-8b/3b models, and the vision-capable pixtral-large-latest.
xAI (Grok)
Section titled “xAI (Grok)”xAI’s Grok models, including a line tuned for fast coding. Create an API key in the xAI console (console.x.ai); the in-product link points at the xAI API docs. Default model is grok-4.3; also available: grok-4.20, grok-code-fast-1 (quick agentic coding), the grok-4-1-fast / grok-4-fast reasoning and non-reasoning variants, grok-4-0709, grok-3, and grok-3-mini.
LiteLLM
Section titled “LiteLLM”Self-hosted gateway that fronts 100+ provider APIs behind one OpenAI-style endpoint — popular with platform teams that want central key custody, budgets, and logging. Point the Base URL at your LiteLLM deployment, add your gateway key, hit Refresh Models to pull its model list, and pick one (default claude-3-7-sonnet-20250219). A Use prompt cache checkbox (off by default) opts into prompt caching for cache-capable models behind the gateway.
Requesty
Section titled “Requesty”Hosted routing layer with a unified catalog and per-key spend analytics. Get a key from requesty.ai (the settings panel deep-links into the app’s API-key page and shows your balance once the key is set). The model list is fetched live; the default is coding/claude-4-sonnet. A custom base URL field supports dedicated router endpoints.
Fireworks AI
Section titled “Fireworks AI”Fast serverless hosting for open-weight models. Keys come from fireworks.ai. Model ids use the accounts/fireworks/models/… form. The default is gpt-oss-120b; the roster is gpt-oss-120b, gpt-oss-20b, llama-v3p3-70b-instruct, llama4-maverick-instruct-basic, and llama4-scout-instruct-basic.
Baseten
Section titled “Baseten”Inference platform serving open-weight models on dedicated performance-focused infrastructure. Create a key at app.baseten.co/settings/api_keys. The catalog is slug-based (vendor/model), and in this edition it comes down to a single entry: openai/gpt-oss-120b, which is also the default.
Unbound
Section titled “Unbound”A security-minded gateway that puts governance and team controls in front of major providers. Get a key from gateway.getunbound.ai. The model list is fetched live from the gateway (default anthropic/claude-sonnet-4-5), and a Refresh Models button re-pulls it after your admin changes what’s allowed.
Vercel AI Gateway
Section titled “Vercel AI Gateway”Vercel’s model gateway: one key, a multi-vendor catalog, and usage visible in your Vercel dashboard. Create an API key under the AI section of the dashboard (the settings panel deep-links there). Models are discovered dynamically; the default is anthropic/claude-sonnet-4, and the catalog is browsable at vercel.com/ai-gateway/models.
Poe’s subscription catalog exposed over an API — one Poe account covers models from many labs. Generate a key at poe.com/api_key (subscription required for meaningful quota). The model list is fetched live with a Refresh Models control.
SambaNova
Section titled “SambaNova”Very fast inference for open-weight models on custom accelerator hardware. Keys come from the SambaNova cloud console at cloud.sambanova.ai. Default model is Meta-Llama-3.3-70B-Instruct; the roster also holds Meta-Llama-3.1-8B-Instruct, Llama-4-Maverick-17B-128E-Instruct, gpt-oss-120b, and the gemma-3-12b-it / gemma-4-31B-it models.
VS Code LM API
Section titled “VS Code LM API”The editor-native option: instead of calling a provider directly, Nexgile Code asks the editor’s Language Model API for models contributed by other installed extensions — most commonly GitHub Copilot’s, via the Copilot and Copilot Chat extensions. Pick a model from the Language Model dropdown (ids look like vendor/family, e.g. copilot/claude-3.5-sonnet).
Treat this entry as experimental-grade, and mind the product’s own warnings:
- Models arriving through this API may be wrapped or fine-tuned by the publishing extension, so behavior can differ from the same model used directly from its provider.
- The first use of a model typically requires accepting a consent prompt from the publishing extension (e.g. in Copilot Chat); until you do, requests can fail with errors like
400 "The requested model is not supported". - Quotas, availability, and capability limits are controlled entirely by the publishing extension and your subscription with it, not by Nx IDE.