ZooWork: Agents That Start With the Job Description
AI AgentsFreemiumZooWork packages a job into an Agent Skill: role, company knowledge, SOPs, delivery standards and tool permissions. Agents plan first, pause for human approval before anything moves money or data, and run on OpenAI, Claude, Gemini, DeepSeek, Kimi or GLM.
About ZooWork
The unit of work in ZooWork is not a prompt. It is an Agent Skill: a package that names the role and the outcome it owns, loads company knowledge and policies, records the SOPs your strongest people actually follow, states what counts as finished, and lists the systems that role is allowed to touch. ZooWork is aimed at forward-deployed engineers and domain experts who already know the work well enough to write it down and want something to run it overnight. That framing explains most of the design. An agent built this way starts from a filled-in job description rather than an empty box, so the first run arrives with context, standards and boundaries attached before it touches anything. Agents plan before they act. Larger tasks break into steps you can inspect, and a run produces tool calls and a deliverable rather than a paragraph of suggestions. The detail that matters for anyone deploying this inside a business is the approval gate. ZooWork holds anything that moves money or data until a person signs off on it. Its own retail walkthrough is explicit about this: the agent requests approval before writing to the warehouse system. That is what lets a long chain of reads and drafts run unattended while still stopping short of an irreversible write. Connectivity is broad and MCP-compatible, so internal tools and data sources can be added without waiting for a native connector. Google Workspace, Microsoft 365, Slack, Lark, Jira and GitHub are named on the site, alongside MCP support for anything else. Agents can be reached from Slack, Microsoft Teams, WhatsApp, the ZooWork app or an API, which matters when the people who need to approve a step are not going to open a fresh application to do it. The model layer is deliberately not locked. OpenAI, Claude, Gemini, DeepSeek, Kimi, GLM and Seedance are all available, and the stated approach is to send hard problems to the strongest model and routine work to the fastest. Consumption is metered in credits rather than sold as seats, so cost tracks actual work instead of headcount. Two adjacent products are worth knowing. ZooData turns a URL into agent-ready JSON, with the site claiming up to 75% fewer LLM tokens and billing only for the data used, across e-commerce, finance, food delivery and social platforms. The agent gallery ships prebuilt roles, including Brand Planner, Store Analyst, Deal Desk, Research Associate, Ops Manager and Contract Reviewer, each wired to specific tools such as Google Drive, Notion or Figma, so you can inspect a finished configuration before writing your own. ZooWork is sold in three shapes: Agent Builder in the app, a Managed Agent API for embedding agents inside your own product, and an enterprise tier where the customer keeps its own data and models. The published onboarding path suggests connecting systems early and having a working role within a few days, which is a reasonable expectation for a read-heavy first agent. The honest trade-off is that this suits people who already have a process worth encoding. If nobody on your team agrees what the SOP is, there is not much for an Agent Skill to package, and the effort lands on the same experts the platform is meant to free. Credits are also the billing unit, so a busy agent costs noticeably more than a chat and spend needs watching. And the $30 Pro price is a promotional rate on a plan the same page lists at $100, so budget for the possibility that it returns to list price.
Pricing & Plans
Pro is $30 per month, listed down from $100 as a limited 70% off offer, with 6,000 monthly credits plus a temporary 14,000 bonus, running on 4-core CPU, 4GB RAM and 40GB storage. Enterprise is quoted per organisation.
Pro
Limited 70% offer; the same plan is listed at $100 without the promotion
- 6,000 monthly credits plus a temporary 14,000 bonus credit
- Build agents through conversation, with 50+ industry agent templates
- 20+ models and multimodal generation
- Managed agent runtime, IM channel integrations, triggers and scheduled tasks
- Knowledge management and ZooData collection
- 4-core CPU, 4GB RAM, 40GB storage
Enterprise
Quoted per organisation, with your own agents, data and models
- Your agents, your data, your model choice
- Organisation-wide governance and deployment
- Talk to sales for scoping
Best For
Recommended use cases and scenarios where ZooWork shines.
Pros
- Approval gates halt a run before anything that moves money or data
- Agent Skills carry role, SOPs, standards and tool permissions, so agents start with context rather than a blank prompt
- MCP-compatible, so internal tools connect without waiting for a native integration
- Model choice is open across OpenAI, Claude, Gemini, DeepSeek, Kimi and GLM, metered by credit rather than by seat
- Agents are reachable from Slack, Microsoft Teams, WhatsApp, the app or an API
- Prebuilt role agents let you inspect a finished configuration before writing your own
Cons
- Credit-based metering makes a busy agent run harder to predict than a flat seat price
- The $30 Pro price is a limited-time offer, with the same plan otherwise listed at $100
- It needs a process worth encoding, so teams without an agreed SOP gain little
- Approval gates keep a person in the loop by design, which does slow long unattended runs
- ZooData token and cost-saving figures are the vendor's own claims rather than independent measurements
Reviews & Ratings
Loading reviews...
Similar Tools
More AI Agents tools you might like
NotebookLM
Google's AI research notebook that summarizes your sources, answers questions with citations, and turns documents into podcasts and study guides.
ChatGPT Agent
ChatGPT Agent lets ChatGPT plan and run multi-step tasks by itself, using apps, files, and cloud browser tools while you stay in the loop.
