# One filesystem. Every sandbox. Every agent.

Agent Drive is a modern, distributed filesystem that lets your AI agents collaborate by sharing data and session context across sandboxes in real-time.

### 1. Create a drive

Spin up a shared drive in one call.

### 2. Mount it in any sandbox

Mount the drive as a regular path.

### 3. Read & write concurrently

Multiple agents share the same files in real-time.

```
import { DriveInstance } from "@blaxel/core";

const drive = await DriveInstance.create({
  name: "shared-drive",
  region: "us-was-1",
});
```

## A shared filesystem for AI agents

Agent Drive behaves like a shared cloud filesystem mounted directly into your sandbox's file tree. Unlike volumes, drives support concurrent access from multiple sandboxes with no capacity planning.

### Share files across sandboxes in real-time

Concurrent read-write access from multiple sandboxes simultaneously, so a fleet of agents can work off the exact same files at once.

### Works like any filesystem

A POSIX-compliant FUSE client makes it behave just like another path in the local filesystem.

### Bottomless, elastic storage

Drives grow automatically with no fixed capacity limits. No pre-provisioning required.

## A real filesystem, not an object store in a costume

S3 was never designed for POSIX operations, nor for trillions of agents. We built Agent Drive differently.

### Small-file optimized

Tuned for the access patterns agents actually produce: many small reads and writes, very high concurrency.

### Custom FUSE client

An optimized POSIX-compliant FUSE client in every sandbox, built in-house for agentics.

### 5-10x faster

Beta users running both approaches report dramatic performance gains on Agent Drive for small-file I/O.

## Purpose-built for agents

Agent Drive removes the need for duplication or orchestration when agents need to share context, datasets or dependencies.

### Handoff of small files

Agents produce lots of small files: notes, JSON artifacts, tool outputs. Agent Drive is tuned for exactly this access pattern: many small files, high concurrency, no serialization in between.

### Shared memory

### Dependencies for codegen

### Parallel agents

┌─HANDOFF:SMALLFILES────────────────┐
││
│agent-A►/shared/notes/│
│├─task-001.json234B│
│├─task-002.json189B│
│├─task-003.json312B│
│└─summary.md1.2KB│
││
│agent-B◄reading...│
│├─task-001.json✓done│
│├─task-002.json✓done│
│└─task-003.json◐active│
││
│THROUGHPUT847files/s│
│LATENCYp50:2ms│
││
└───────────────────────────────────────┘

┌─SHAREDMEMORY───────────────────────┐
││
│/memory/│
│├─history/│
││├─session-2024-11.jsonl│
││└─session-2025-01.jsonl│
│├─knowledge/│
││├─decisions.md18KB│
││└─glossary.md7.3KB│
│└─embeddings.bin421MB│
││
│sandbox-new◐mounting...│
│└─resumedfrompriorctx✓│
││
└───────────────────────────────────────┘

┌─DEPSCACHE──────────────────────────┐
││
│/deps/node\_modules/│
│├─react@18.3.1✓cached│
│├─next@15.2.6✓cached│
│├─tailwindcss@3.4✓cached│
│└─…1,847pkgs✓cached│
││
│sandbox-1mount→instant│
│sandbox-2mount→instant│
│sandbox-3mount→instant│
││
│COLDINSTALL~38ssaved/spawn│
│STORAGE1×shared│
││
└───────────────────────────────────────┘

┌─PARALLELFAN-OUT/FAN-IN─────────────┐
││
│coordinator►/shared/results/│
││
│worker-01●01.json✓done│
│worker-02●02.json✓done│
│worker-03●03.json◐run│
│worker-04●04.json◐run│
│worker-05●05.json◐run│
│worker-06○pending...│
││
│coordinator◄aggregating5/6│
│PROGRESS████████████░░83%│
││
└───────────────────────────────────────┘

## Get early access to Agent Drive.

Apply for private preview to start using our distributed filesystem for your AI agents. We're onboarding new teams every week.