Providers Overview

Nexgile Code is bring-your-own-provider. The agent inside Nx IDE does not bundle a model or route your traffic through a first-party relay — you connect it directly to an AI provider you already have an account with, using your own API key, cloud credentials, subscription sign-in, or a model server running on your own machine. Your prompts and code go straight from your editor to the provider you chose, and you pay that provider directly at their rates (or nothing at all, for local models).

This page is the map: every selectable provider, how configuration profiles work, and how to decide which provider fits which job.

Why bring your own provider

• Choice. Frontier models, budget models, open-weight models, and fully local models are all one dropdown away — and you can switch mid-project without losing anything.
• Cost transparency. You see the provider’s own pricing, and Nexgile Code shows the live API cost of every task in the task header.
• Privacy control. Point the agent at a local Ollama or LM Studio server and no code ever leaves your machine. Point it at your own AWS or GCP account and it stays inside your cloud perimeter.
• No lock-in. Model quality moves fast. When a better model ships, you add a profile for it and keep working.

The provider roster

Nexgile Code offers 21 providers in the provider dropdown (Settings → Providers → API Provider). Grouped by how they work:

Hosted — direct provider APIs

Provider	Auth	Best known for
Anthropic	API key	Claude models — top-tier agentic coding, prompt caching
OpenAI	API key	GPT and Codex model lines, o-series reasoning models
OpenAI - ChatGPT Plus/Pro	Sign in with a ChatGPT subscription (no key)	Codex coding models covered by your existing subscription
Google Gemini	API key (Google AI Studio)	Gemini Pro/Flash models, huge context windows
Amazon Bedrock	AWS credentials / profile / Bedrock API key	Claude, Nova, Llama, and more inside your AWS account
GCP Vertex AI	Service account or application default credentials	Gemini and Claude inside your Google Cloud project
Mistral	API key	Codestral and Mistral model families
xAI (Grok)	API key	Grok models, including a fast coding line
Fireworks AI	API key	Fast hosted open-weight models
Baseten	API key	Hosted open-weight models on dedicated infrastructure
SambaNova	API key	Very fast open-weight model serving

Aggregators & gateways — one account, many models

Provider	Auth	Best known for
OpenRouter	API key	One key for a huge multi-vendor catalog
Requesty	API key	Routed access to major models with spend tracking
LiteLLM	Your gateway URL + key	Self-hosted proxy in front of any provider mix
Unbound	API key	Gateway with team-level model governance
Vercel AI Gateway	API key	Vercel’s model gateway and catalog
Poe	API key	Poe subscription catalog over an API

Local — your machine, your models

Provider	Auth	Best known for
Ollama	None (local server)	One-command local models, fully offline
LM Studio	None (local server)	Desktop app with a local OpenAI-style server
OpenAI Compatible	Base URL + key	Any OpenAI-shaped endpoint: vLLM, corporate proxies, Azure OpenAI, self-hosted gateways

Editor-native

Provider	Auth	Best known for
VS Code LM API	Handled by the editor	Models exposed by other editor extensions (for example GitHub Copilot) — experimental-grade

Note — TAA compliance

Nx IDE ships the TAA-compliance edition, which excludes providers operated by companies domiciled in China. Those providers are absent from the dropdown, blocked at connection time even if a profile names one, and their model families are filtered out of every catalog — the fixed rosters above, the live catalogs of aggregators like OpenRouter and LiteLLM, and the models discovered from local Ollama and LM Studio servers. Per-model endpoint routing on aggregators is filtered the same way, so inference never lands on Chinese-domiciled infrastructure. The edition is compiled into the build; no environment variable or setting can turn the filtering off.

API configuration profiles

Everything you set up on the Providers tab — provider, credentials, model, temperature, reasoning effort, rate limit — is saved as a named Configuration Profile. The profile selector sits at the top of Settings → Providers:

• Create several. Add a profile per provider or per purpose: claude-daily, gpt-deep-reasoning, local-ollama, cheap-bulk-work.
• Switch anytime. Change the active profile from the settings view or the chat area; the very next request uses the new provider and model.
• Pin per mode. Each mode can pin its own profile (“API Configuration — Select which API configuration to use for this mode”), so Architect can run a deep-reasoning model while Code runs a fast implementation model. See Using Modes.
• Dedicated utility profiles. Context condensing can use its own profile (“API Configuration for Context Condensing”), letting a cheap model do summarization while an expensive model does the real work. See Context Management.

Example profile set for one developer:

  claude-daily      Anthropic · claude-sonnet-4-5        default for Code mode
  opus-architect    Anthropic · claude-opus-4-8          pinned to Architect mode
  local-private     Ollama · devstral:24b                for the proprietary repo
  condense-cheap    OpenAI · gpt-5.4-mini                used only for condensing

How to choose

• Frontier quality first: Anthropic (Claude Opus/Sonnet), OpenAI (GPT-5.x / Codex), or Google Gemini Pro. Best results in agentic coding, long tasks, and large refactors.
• Enterprise procurement and compliance: Amazon Bedrock or GCP Vertex AI put the same frontier models inside your existing cloud account, IAM, and billing.
• Cost first: Gemini Flash models, GPT mini/nano tiers, or open-weight models (gpt-oss, Llama, gemma) via Fireworks AI, SambaNova, or Baseten.
• Maximum flexibility: OpenRouter (or another gateway) — one key, try everything, switch models per task.
• Privacy first: Ollama or LM Studio. Nothing leaves your machine; expect a capability gap versus frontier hosted models.
• Already paying for ChatGPT: the OpenAI - ChatGPT Plus/Pro entry uses your subscription instead of metered API billing.

Whatever you pick, the per-task cost readout, rate limits, and the Max Cost / Max Requests caps keep spending visible and bounded — see Model Behavior & Costs and Command Guardrails & Limits.

Provider pages

• Anthropic
• OpenAI — includes the ChatGPT Plus/Pro sign-in entry
• Google Gemini
• AWS Bedrock
• GCP Vertex AI
• OpenRouter
• Ollama (Local)
• LM Studio (Local)
• OpenAI-Compatible Endpoints
• More Providers — Mistral, xAI, LiteLLM, Requesty, Fireworks AI, Baseten, Unbound, Vercel AI Gateway, Poe, SambaNova, VS Code LM API

Related

• Connecting an AI Provider
• Model Behavior & Costs
• Command Guardrails & Limits
• Using Modes