Provider Overview

Human uses a vtable-driven provider interface. All AI model backends implement hu_provider_t and are selected by name at runtime.

Provider vtable interface

Each provider implements:

Method	Description
chat	Non-streaming chat completion
chat_with_tools	Chat with tool calls (optional)
stream_chat	Streaming completion (optional)
supports_native_tools	Whether provider uses OpenAI-style tool schema
get_name	Provider identifier
deinit	Cleanup

Optional methods: warmup, supports_streaming, supports_vision.

Provider types

Native providers (core)

These have dedicated implementations with format-specific handling:

Name	Type	Default base URL
openai	Native	https://api.openai.com/v1
anthropic	Native	https://api.anthropic.com
gemini	Native	https://generativelanguage.googleapis.com
google	Native	(same as gemini)
ollama	Native	http://localhost:11434
openrouter	Native	https://openrouter.ai/api/v1
compatible	Native	(configurable)
claude_cli	Native	(Claude CLI)
codex_cli	Native	(Codex CLI)
openai-codex	Native	(OpenAI Codex)

Compatible providers (OpenAI chat API)

These use the compatible backend with preset base URLs:

Name	Default base URL
groq	https://api.groq.com/openai
mistral	https://api.mistral.ai/v1
deepseek	https://api.deepseek.com
xai, grok	https://api.x.ai
cerebras	https://api.cerebras.ai/v1
perplexity	https://api.perplexity.ai
cohere	https://api.cohere.com/compatibility
together, together-ai	https://api.together.xyz
fireworks, fireworks-ai	https://api.fireworks.ai/inference/v1
huggingface	https://router.huggingface.co/v1
siliconflow	https://api.siliconflow.cn/v1
venice	https://api.venice.ai
vercel, vercel-ai	https://ai-gateway.vercel.sh/v1
chutes	https://chutes.ai/api/v1
synthetic	https://api.synthetic.new/openai/v1
opencode, opencode-zen	https://opencode.ai/zen/v1
astrai	https://as-trai.com/v1
poe	https://api.poe.com/v1
moonshot, kimi	https://api.moonshot.cn/v1
glm, zhipu, zai, z.ai	https://api.z.ai/api/paas/v4
minimax	https://api.minimax.io/v1
qwen, dashscope	https://dashscope.aliyuncs.com/compatible-mode/v1
qianfan, baidu	https://aip.baidubce.com
doubao, volcengine, ark	https://ark.cn-beijing.volces.com/api/v3
byteplus	https://ark.ap-southeast.bytepluses.com/api/v3
bedrock, aws-bedrock	https://bedrock-runtime.us-east-1.amazonaws.com
cloudflare, cloudflare-ai	https://gateway.ai.cloudflare.com/v1
copilot, github-copilot	https://api.githubcopilot.com
nvidia, nvidia-nim	https://integrate.api.nvidia.com/v1
ovhcloud, ovh	https://oai.endpoints.kepler.ai.cloud.ovh.net/v1

Local providers

Name	Default base URL
lmstudio, lm-studio	http://localhost:1234/v1
vllm	http://localhost:8000/v1
llamacpp, llama.cpp	http://localhost:8080/v1
sglang	http://localhost:30000/v1
osaurus	http://localhost:1337/v1
litellm	http://localhost:4000

No API key is required for local providers.

Selecting a provider

Set default_provider in config or pass --provider on the CLI:

./human agent --provider ollama --model llama3

For compatible providers, add a providers entry with your API key:

{
  "providers": [{ "name": "groq", "api_key": "gsk_..." }],
  "default_provider": "groq"
}