All Providers Explained

AgentX ships with built-in providers plus custom providers for anything OpenAI-compatible. This page explains every one: what it is, where to get a key, what it costs, and what it's best for.

Pricing and free tiers change often. Figures below are approximate as of October 2026 — always confirm on the provider's official pricing page before spending money.

Built-in providers

These appear directly in Settings → Providers. Tap one, paste your API key(s), pick models, done. You can paste multiple keys per provider (one per line) — AgentX rotates between them automatically to spread load and dodge rate limits.

Google (Gemini)

• What: Google's Gemini models (e.g. Gemini 2.x Flash/Pro) via Google AI Studio.
• Get a key: aistudio.google.com → Get API key. Free Google account, no credit card.
• Free tier: Generous — roughly ~1,500 requests/day on Flash-class models. Enough for heavy daily use.
• Paid: Pay-as-you-go through Google Cloud billing once free limits are exceeded.
• Best for: The best free starting point. Fast, smart, multimodal (text + images + audio).

OpenAI

• What: GPT models via api.openai.com/v1.
• Get a key: platform.openai.com → API keys. Requires a paid account with billing set up.
• Free tier: None permanent (new accounts occasionally get small trial credits).
• Paid: Per-token, pay-as-you-go. The most expensive mainstream option, but top-tier quality.
• Best for: Maximum model quality when cost doesn't matter; also the reference OpenAI-compatible API everything else imitates.

Anthropic (Claude)

• What: Claude models via Anthropic's API.
• Get a key: console.anthropic.com. Requires billing.
• Free tier: None permanent (occasional promotional credits).
• Paid: Per-token, pay-as-you-go. Generally cheaper than OpenAI for similar quality.
• Best for: Long-context reasoning, careful instruction-following, coding agents.

Heads-up

If Claude fails inside AgentX but works elsewhere, double-check the key saved in the app's provider settings — it's a common gotcha to test one key and paste a different one into the app.

DeepSeek

• What: DeepSeek's own API (api.deepseek.com) — the open-weight reasoning models as a hosted service.
• Get a key: platform.deepseek.com.
• Free tier: Occasional promotions; normally paid but extremely cheap.
• Paid: Per-token — among the cheapest reasoning-capable models available.
• Best for: Cheap deep reasoning and coding on a budget.

Qwen (Alibaba)

• What: Alibaba's Qwen models via DashScope international (dashscope-intl.aliyuncs.com/compatible-mode/v1, OpenAI-compatible).
• Get a key: Alibaba Cloud DashScope console.
• Free tier: Trial quotas for new accounts.
• Paid: Per-token, competitively priced.
• Best for: Strong multilingual (especially Chinese/English) models at low cost.

Groq

• What: Ultra-fast inference (LLaMA, Qwen, Mixtral and more) on Groq's LPUs via api.groq.com/openai/v1.
• Get a key: console.groq.com. Free account, no credit card for the free tier.
• Free tier: Generous — roughly 30 requests/min and ~14,400 requests/day on popular models.
• Paid: Pay-as-you-go for higher limits.
• Best for: The fastest free responses anywhere. Great default for quick agents and browser narration.

Ollama (local)

• What: Models running on your own computer via Ollama (http://localhost:11434/v1).
• Get a key: None needed — just run Ollama on a PC on the same network and point AgentX at it.
• Free tier: Completely free — your hardware is the limit.
• Paid: Never (just your electricity).
• Best for: Privacy (nothing leaves your network), offline use, unlimited experimentation.

OpenRouter

• What: One API for hundreds of models (openrouter.ai/api/v1) — OpenAI, Anthropic, Google, Meta, DeepSeek and more behind a single key.
• Get a key: openrouter.ai.
• Free tier: Models tagged :free are free — ~50 requests/day, rising to ~1,000/day after a one-time $10 credit purchase.
• Paid: Per-token billing across all models with one balance.
• Best for: Trying many different models with one key; free-tier experimentation.

OpenCode Go

• What: OpenCode's Zen model gateway (opencode.ai/zen/go/v1, OpenAI-compatible).
• Get a key: From your OpenCode account — see opencode.ai for current plans.
• Best for: If you already use OpenCode, reuse the same model access inside AgentX.

Local (on-device)

• What: Models running directly on the phone.
• Get a key: None.
• Cost: Free.
• Best for: Offline quick tasks. Limited by phone RAM — small models only.

TypeSafe (Jev decision engine)

• What: Not a chat provider — this is the Jev ultrafast decision engine used for browser automation. It lives on its own settings page (Settings → Jev), not in the providers list.
• Get a key: From your Jev provider; paste the base URL and model name yourself.
• Cost: Depends on the Jev key you bring.
• Best for: Fast browser clicking — ~1–2 seconds per action instead of a full model round-trip. See Jev.

Custom providers (bring anything)

Any service that speaks the OpenAI-compatible chat API can be added as a custom provider: Settings → Providers → Add custom. You choose the protocol:

Protocol	Use for
OpenAI	NaraRouter, OpenRouter-compatible gateways, local servers (LM Studio, llama.cpp server), any /v1/chat/completions endpoint
Google	Gemini-compatible endpoints
Anthropic	Claude-compatible endpoints

Popular custom setups:

• NaraRouter — free API keys for popular models. Full walkthrough: Free API Keys (NaraRouter).
• Your own gateway — e.g. a Cloudflare Worker that shims tools or routes between keys.

Which provider should you start with?

Recommended free starter stack:

1. Google (Gemini) — your main smart model, generous free tier.
2. Groq — your fast model for quick tasks and agent loops.
3. OpenRouter :free models — for trying anything else at zero cost.
4. NaraRouter — extra free keys when you need more quota.

When to pay: when you consistently hit free limits, or you want frontier quality (Claude/GPT) for important work. Start free, pay only for what the free tiers can't cover.

Next steps

• Free API Keys (NaraRouter) — step-by-step free key setup
• Add a Provider — exact fields and what they mean
• Choose Models — picking and switching models
• Free vs Paid — every cost in the whole stack, compared