Agents

Agents

The Zen agent model family is purpose-built for agentic workflows, function calling, and multi-step reasoning. These models are trained on real-world environment reinforcement learning to excel at tool use, code generation, and complex task execution.

Model Family

Quick Start

Below is a simple example using the Zen Agent 4B model with the Transformers library:

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "zenlm/zen-agent-4b"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype="auto",
    device_map="auto"
)

messages = [{"role": "user", "content": "What tools do you have access to?"}]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
inputs = tokenizer([text], return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True))

Zen API

For production deployments and higher throughput, use the Zen API endpoint at api.hanzo.ai. The API provides OpenAI-compatible endpoints and automatic scaling:

curl https://api.hanzo.ai/v1/chat/completions \
  -H "Authorization: Bearer $HANZO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "zen-agent-4b",
    "messages": [{"role": "user", "content": "Execute this task..."}]
  }'

Get a free API key with $5 credit at console.hanzo.ai.

Model Variants

Zen 5 Mini is the frontier-agentic tier, featuring a sparse mixture-of-experts architecture with ~10B active parameters and trained on large-scale real-world environment reinforcement learning.

Zen Agent 4B and Zen Eco variants are compact, efficient 4B models optimized for tool calling and multi-step reasoning with a 32K token context window. The GGUF variant is optimized for llama.cpp and CPU inference, while the MLX variant is tuned for Apple Silicon hardware.