Open source · MCP native

Your agents work in isolation.
Give them a shared workspace.

37 agents per org. Each writes to local files that no one else can see. No persistence across sessions. No way for humans to review, comment, or intervene. AgentDrive is the persistent, collaborative storage layer where agents and humans work on the same files — in real time.

Works with OpenClaw, Claude Code, Cline, LangChain, CrewAI, and any MCP client.

Agents
write analysis-q1.md
write metrics.json
read feedback on brief.md
update brief-v2.md
Humans
analysis-q1.md
“add region data”
metrics.json
brief-v2.md

Imagine hiring 37 employees and telling them: you have no desk.

No shared drive. No way to leave work for others to review. Every night, we shred everything you produced.

That’s how we treat AI agents today. They write reports, analyses, datasets, code — then it sits in ~/clawd/ where nobody reads it. Sessions end, context compacts, files vanish. Other agents can’t access each other’s work. Humans have zero visibility.

88% of orgs report agent security incidents — with no audit trail
agent-fleet-status
$ ls ~/clawd/
AGENTS.md
report-q1.md
analysis.md
3 files, local only

# who wrote these?
unknown — no attribution

# can other agents read them?
no — single workspace, no sharing

# can I comment on analysis.md?
no — no feedback mechanism

# what if this machine crashes?
gone — no persistence, no backup

An OpenClaw agent autonomously deleted 200+ emails from a Meta researcher’s inbox. The safety rule got dropped during memory compaction. No dashboard. No oversight. No way to intervene. — Summer Yue incident, Feb 2026

The solution

A persistent shared workspace where agents and humans collaborate on real files.

Agents write to AgentDrive instead of local disk. Every file is attributed, versioned, and visible to your team. Humans review, comment, and direct agent work — without breaking agent execution.

Agents

Write files, read feedback,
update and version

files →
← feedback

Humans

View dashboard, review,
comment, approve

The dashboard

See what every agent is building. Right now.

Every file, who wrote it, when. Live activity feed. Click to review. Leave feedback agents read on their next run.

app.agentdrive.sh
acme-corp
Workspace
Activity
Agents
Permissions
Workspace 12 files · 4 agents
md
analysis-q1.md agent-phoenix 2m ago
md
campaign-brief-v2.md rakesh 14m ago
json
revenue-metrics.json agent-analytics 38m ago
csv
user-segments.csv agent-data 1h ago
Live activity
agent-phoenix updated analysis-q1.md (v3)
rakesh commented: “Add regional breakdown”
agent-analytics created revenue-metrics.json
agent-data read feedback on user-segments.csv
Capabilities

The infrastructure your agent fleet is missing

Persistent storage, real-time collaboration, identity-aware permissions — built for multi-agent workflows.

Persistent shared storage

Files survive session ends, memory compaction, and machine crashes. Every agent writes to the same workspace. Nothing gets shredded.

Human-in-the-loop feedback

Click a file, leave a comment. Your agent reads it on the next heartbeat and revises. This is how humans direct agent work without breaking execution.

Multi-agent orchestration

agent-phoenix writes to /reports/. agent-analytics reads it and enriches with metrics. Inter-agent file sharing with full attribution and provenance.

Identity-aware permissions

Each agent gets a unique identity. Scoped write access per directory. Your PM views but can’t delete. RBAC for agents and humans.

Universal MCP protocol

OpenClaw skill. MCP server for Claude Code, Cline, LangChain, CrewAI. One workspace, every agent framework. Install in 2 minutes.

Full version history

v1, v2, v3 — every revision saved with who changed what and when. Diff any version. Roll back instantly. Complete audit trail for compliance.

How it works

From zero visibility to full collaboration in 5 minutes

1
Agent

Install the skill

One command. Agents write to AgentDrive instead of local disk.

2
Agent

Writes to shared workspace

Files sync to persistent storage. Attributed, versioned, accessible to other agents.

3
Human

Reviews in dashboard

Open the browser. See every file, who wrote it, when. Leave comments.

4
Agent

Reads feedback, iterates

Next run: reads your comment, revises, saves v2. The collaboration loop completes.

# Install for OpenClaw
openclaw skill install agentdrive

# Or for Claude Code / Cline / any MCP client
npm install @agentdrive/mcp-server

✓ agent-phoenix saved /workspace/analysis-q1.md
✓ agent-analytics saved /workspace/metrics.json
✎ rakesh: "Add revenue breakdown by region"
✓ agent-phoenix updated /workspace/analysis-q1-v2.md
Built for

Anyone running multi-agent workflows

OpenClaw operators

3 agents on a Pi, 2 on a VPS, 5 in the cloud. All writing to one persistent workspace. Monitor from anywhere.

AI engineering teams

LangChain, CrewAI, custom pipelines. Agents share intermediate files. Humans review before the next stage runs.

Team leads & managers

Your team deploys agents. You can’t see what they produce. Now you have a dashboard. Comment and direct.

Compliance & governance

EU AI Act requires human oversight of autonomous AI. AgentDrive is the audit trail, attribution log, and control layer.

Your agents are working right now. Can you see what they’re building?

Open source. Persistent storage. Real-time human-agent collaboration. Install in 5 minutes.

Free for individuals. Open-source MCP server.