AgentDrive is the shared folder for your agents — with a dashboard so you can actually see what they made.
Open source. Works with OpenClaw, Claude Code, Cline, and any MCP client.
The problem
Last month, an OpenClaw agent deleted 200+ emails from a Meta researcher’s inbox. The “ask before acting” rule got dropped during memory compaction. Nobody saw it happening. No dashboard. No way to stop it. — Summer Yue incident, Feb 2026
Install the skill. Your agents write to AgentDrive. You open a browser and see everything they made. Leave a comment. They read it on the next heartbeat.
How it works
Install the skill. Agents write to AgentDrive instead of ~/clawd/.
Open the dashboard. Every file, who wrote it, when.
Click a file. Type feedback. It’s attached to the file.
Next heartbeat: reads your comment, revises, saves v2.
What you get
See every file your agents wrote. Who made it. When. Click to read. Like Google Drive for agent output.
Click a file. Leave feedback. Your agent reads it on the next heartbeat. That’s how you direct agent work.
Live stream of everything happening. agent-phoenix wrote a file. You commented. agent-analytics updated metrics.
agent-phoenix writes to /reports/ but not /finances/. Your PM views but can’t delete. Scoped access for everyone.
OpenClaw skill. MCP server for Claude Code, Cline, LangChain, CrewAI. One folder, all your agents.
Agent wrote v1. You commented. Agent wrote v2. Both saved. Always go back. Always see who changed what.
Built for
3 agents on a Pi, 2 on a VPS. All in one place. See it all from anywhere.
Claude Code, Cline, Cursor. Agent work persists between sessions and across machines.
Your team runs agents. You can’t see what they produce. Now you can.
EU AI Act says humans must oversee AI. AgentDrive is the audit trail.
Get started
Open source. Install the skill. See everything in a dashboard.