ZooWork

• A consultant turning a client's proposal or reporting playbook into an agent the client's team can mention in Slack or Teams
• A retail operations lead running a daily stockout-risk check that drafts replenishment orders and waits for approval before writing to the WMS
• A forward-deployed engineer embedding a ZooWork agent in a customer's internal app through the TypeScript or Python SDK, with one Agent per customer
• Scheduling a weekday research or market-signal brief that publishes its output as a document artifact
• Replacing an application-owned model and tool loop with managed Sessions, saved events, and an idle-pausing sandbox

ZooWork stands out for being built around delivering agents to other people rather than automating your own inbox. The Builder lowers the bar for a domain expert to package a playbook, and the Managed Agent API gives an engineer the same runtime to embed per client. The documentation is the most useful part of the evaluation because it is explicit about the gaps. Client isolation is per Agent, not per session. Public keys cannot bring authenticated MCP servers or custom sandboxes. Spending caps are your job. The API is still a Developer Preview. For a consultant or FDE who can live inside a hosted platform, that is a workable start. Teams that need self-hosting, credentialed connectors through the API, or a free tier to experiment should look elsewhere for now.

About ZooWork

ZooWork, from Serendipity One (the company behind the Gensmo shopping agent) and formerly called ZooClaw, has two ways in. Agent Builder is the no-code workspace: you describe how the work should be done, ZooWork creates a runnable agent, and further conversation in Build updates the agent's own instructions and Skills for future tasks. You can start from 50+ industry and role templates, pick from 20+ models, add knowledge, set triggers and schedules, and put the agent in chat apps so teammates can mention it in Slack, Microsoft Teams, or WhatsApp. The homepage shows agents planning a task, calling business systems, and pausing at a human approval gate before writes such as an ERP update. The Managed Agent API is the developer side and is in Developer Preview. Your backend creates an Agent (instructions, model, tools, Skills), opens Sessions, and streams events. ZooWork runs the model and tool loop, keeps conversation history, and executes code and files in a managed E2B cloud sandbox that pauses when idle. Tools can be built-in, remote MCP servers, or application-executed custom tools, and each tool can be set to always run or to wait for approval. Pricing is split. Agent Builder Pro is listed at $30 a month as a limited 70%-off offer (normally $100) with 6,000 monthly credits plus 14,000 temporary bonus credits. The API uses a separate prepaid balance starting at $5, with model tokens at each provider's list price, tool fees at the provider's price, and $0.20 per running hour for the default 4 vCPU sandbox. The docs are unusually candid about current limits. Sessions on one Agent share its workspace, so isolating clients means one Agent per client. Public API keys cannot create custom environments, manage chat channels, or connect authenticated or private MCP servers. There is no server-enforced spending cap, and no self-hosted deployment is documented. ZooWork is a good fit for experts packaging a repeatable workflow for several teams or clients. It is a weaker fit for teams that need self-hosting, open-source control, or credentialed connectors through the API today.