22 LLM Providers, 49 Models

Every provider uses real API endpoints - no stubs. Zero SDK dependencies (all HTTP via Node fetch).

Provider List

#	Provider	Free Tier	Default Model	Context Window
1	ZhipuAI (GLM)	✅	glm-4-flash	128K
2	Agnes AI	✅	agnes-2.0-flash	128K
3	DeepSeek	-	deepseek-chat	64K
4	Volcengine Ark	-	doubao-seed-code	128K
5	Moonshot	-	moonshot-v1	128K
6	Qwen (Tongyi)	-	qwen-turbo	128K
7	MiniMax	-	abab6.5s-chat	245K
8	Baichuan	-	baichuan2-turbo	192K
9	Yi (01.AI)	-	yi-lightning	16K
10	StepFun	-	step-1-8k	8K
11	SiliconFlow	-	Qwen/Qwen2.5-7B-Instruct	32K
12	OpenAI	-	gpt-4o	128K
13	Anthropic	-	claude-sonnet-4-20250514	200K
14	Gemini	-	gemini-2.0-flash	1M
15	OpenRouter	-	auto	varies
16	Together	-	meta-llama/Llama-3	8K
17	Groq	-	llama-3.1-8b-instant	128K
18	Fireworks	-	llama-v3p1-70b-instruct	32K
19	Mistral	-	mistral-large-latest	128K
20	Ollama (local)	✅	llama3.2	128K
21	LM Studio (local)	✅	-	varies
22	vLLM (local)	✅	-	varies

Configuration

Each provider is configured via environment variables or the config file:

```env

Example: GLM

GLM_API_KEY=your_key GLM_BASE_URL=https://open.bigmodel.cn/api/paas/v4 GLM_MODEL=glm-4-flash

Example: OpenAI

OPENAI_API_KEY=sk-... OPENAI_BASE_URL=https://api.openai.com/v1 OPENAI_MODEL=gpt-4o

Example: Ollama (local, no key)

OPENAI_API_KEY=ollama OPENAI_BASE_URL=http://localhost:11434/v1 OPENAI_MODEL=llama3.2 ```

Provider Fallback

At startup, each provider is probed. The first healthy one becomes primary. If it fails at runtime, requests fall back to the next provider in the chain.

Model Manager

Hot-swap models at runtime:

```ts import { modelManager } from "quark-agent";

// List configured models modelManager.list();

// Set active model modelManager.setActive("deepseek-chat");

// Probe a model const healthy = await modelManager.probe("glm-4-flash");

// Compare models const diff = modelManager.compare("glm-4-flash", "deepseek-chat"); ```