Raccoon is an AI agent platform that turns plain-language descriptions into finished, deployable work by planning, building, and shipping tasks across connected tools. It operates on a persistent cloud computer, allowing users to steer, rewind, or fork its work in real time. The platform integrates with 50+ connectors and supports custom MCP servers, with per-tool permissions and editable memory for transparent control.
Funding
Funding not disclosed
Founders
Product
Problem
Building digital work—whether a web app, a slide deck, or a data analysis—typically requires navigating multiple disconnected tools, writing complex prompts, and manually managing files, versions, and permissions. Users often lose context between sessions, struggle to steer AI outputs mid-task, and face opaque processes that make it hard to trust or verify the final result.
Solution
Raccoon provides an AI agent that operates on a persistent cloud computer, turning plain-language descriptions into finished, shipped work. Users can describe a task by typing, speaking, sketching, or dropping in files, and Raccoon plans the work openly as a live checklist, executes it in full view, and allows real-time steering, rewinding, or forking to explore alternative paths. It connects to real accounts like Gmail, Drive, GitHub, and Notion, plus any MCP server, with per-tool permissions and allow-lists for control. Raccoon also remembers user preferences and project context across sessions, and supports reusable, guardrailed skills that package workflows with enforced safety rules.
Target Audience
Primary users are individual professionals, small teams, and developers who need an autonomous AI agent to build and ship digital work—such as web apps, slides, and data analyses—while maintaining control over connected tools and workflows.
Features
- Persistent cloud computer with live file system, app previews, and backend console beside every chat
- Multi-modal input: type, voice, sketch, or file uploads, with no prompt-craft or pre-configuration required
- 50+ built-in connectors (Gmail, Drive, GitHub, Notion, Supabase) plus any MCP server, with per-tool allow-lists and read-only or mutating action controls
- Editable, transparent memory that stores preferences and project context, with every read/write shown and manageable by the user
- Skills system with hard guardrails (per-skill tool allow-lists/deny-lists), importable from zip or public repos, and auto-loading when relevant
- Rewind, fork, and subagent capabilities for parallel task execution and flexible exploration
- One-click publishing to live links or custom domains, with full state persistence across sessions