Build Agents That Get Work Done.

Add agents to your app. ZooWork runs them and saves session history.

What ZooWork handles

Agent setup

Set instructions, a model, tools, and skills once. Start sessions as needed.

Set up an agent

Tools and sandbox

Search the web, run code, and create files. Connect services with MCP or custom tools.

Agent tools

Sessions and events

Get replies and tool updates as they happen. Resume with saved history.

Stream events

Make your first API call

Install the SDK. Set ZOOWORK_API_KEY to your project key on your server, then list available models.

verify-key.mts
import { createZooworkClient } from '@zoowork-ai/sdk'

const client = createZooworkClient()
console.log(await client.listModels())

ZooWork Agent Runtime

Focus on the skills your agents need. ZooWork handles how they run and scale.

Execution harness

Plan and run work in steps, with guardrails and human review steps you define.

Parallel tasks converge at a review step before completion.

Continuous improvement

Use feedback and results from each run to improve agent skills.

A feedback loop connects execution, reflection, and learning.

Cloud-native infrastructure

Scale compute, storage, networking, and task orchestration as workloads grow.

An orchestrator distributes work across compute, storage, and network resources.

Pay as you go

Add funds in USD. Projects in your organization share one balance.

Models

Each model has separate rates per million input, cached input, and output tokens.

Model pricing

Tools and media

Generate images, video, and voice, or use search and data tools. Rates vary by task.

Tool pricing

Sandbox compute

The default sandbox (4 vCPU, 4 GiB) is billed by runtime.

Compute pricing

Platform and Agent Builder have separate balances. View usage

ZooWork Token Cashback

FDEs and developers can apply for cashback on eligible ZooWork API model costs.

30% simulated cashback on DeepSeek Flash Latest.

Final terms set who qualifies and how much they get.