What is Sprout
Sprout by Willow lets anyone build their own AI agent by describing it in a sentence. You say what you want the agent to do, Sprout drafts it with the right apps, tools and skills, and you chat with it on the web, in Slack, or from an assistant you already use.
Open Sprout at sprout.withwillow.ai.
What an agent is made of
Every Sprout agent has the same few parts, and you can change any of them at any time:
- Instructions: what the agent does and how it talks, in plain language.
- Connectors: the apps and MCP servers it can use, such as Gmail, Google Calendar, Notion or GitHub, and which of their tools are switched on.
- Skills: Markdown playbooks it follows for one kind of request, like a reply format or a refund policy.
- Built-in tools: web search, reading web pages, a private sandbox for commands and files, and memory.
- Channels: where you talk to it: a web chat page and Slack.
Where an agent runs
When you build an agent you choose where it runs. You can move it later without losing its tools, skills or memory.
| Runs on | How you use it |
|---|---|
| Willow Agents | Willow runs the agent for you. Chat with it in the web chat or in Slack. |
| Instinct | Instinct opens the agent's tools through a private link and a password. |
| Grok Bot | Grok uses the agent's tools as a remote MCP server, with a password. |
Safe by default
Sprout runs on Willow, the gateway companies use to control what their AI agents can do. Every Sprout agent gets the same protection:
- Your accounts stay with Willow. When the agent uses Gmail or GitHub, Willow's gateway holds the sign-in and makes the call. The agent never sees a password or key.
- You approve what matters. Tools that change data ask you before they run. You choose which tools run freely, which ask each time, and which never run.
- Guards are always on. A prompt-injection shield, masking of secrets and personal data in tool calls, and a log of every tool call.
- A safety score shows what to tighten. The Safety tab rates what the agent can do and fixes each recommendation in one click. See Safety score.
Your agents are personal: they use your accounts and only you can chat with them.