The filesystem for your AI agents

Clawbox gives your AI agents a filesystem that pulls data from your tools and syncs to your OpenClaw machine or your coding agent sandboxes. Everything your agent needs, as files.

Claim your namespace

Your agent gets /home/{you} on the global filesystem

~/workspace

The Problem

Your agent has a computer. It just doesn't have the right files.

Claude Code, Codex, and OpenClaw work best when they can read and write files. But getting data from HubSpot, Linear, Notion, Gmail, and everywhere else into an agent's sandbox? That's the hard part.

Data is scattered

Your agent needs context from six different SaaS tools. Right now you're copy-pasting or writing bespoke MCP servers.

Sandboxes are ephemeral

Modal, E2B, and Daytona spin up fresh VMs. Every time, you need to re-fetch and re-mount your data.

Sync and access are broken

Agents need the right access without reading stale data or clobbering your files. Permissions and sync should be built in.


Real Workflow

How I run sales every morning with Claude Code

Yes, only an engineer would run sales on a POSIX filesystem with a coding agent. But it works unreasonably well.

01

Agents sync data automatically

Three Claude agents kick off on a schedule: one scans my inbox for Bookface posts, one checks HubSpot tasks and drafts outbound emails, one monitors the YC Companies Directory.

02

Everything lands as files

All data goes into a network-mountable filesystem accessible from Claude Code or remote sandboxes like Modal and E2B. The outputs are just markdown files.

03

Review and iterate

I review the tasks and emails Claude prepared, ask it to make changes, or kick off follow-up automations. It's all files — easy to inspect, version, and share.

clawbox demo

Another example: Mount a Linear ticket and all related Notion docs into your sandbox and let your agent rip.


Under the Hood

How Clawbox works

Scheduled data sync

Clawbox runs sandboxed scripts on a schedule that pull data from your tools — email, HubSpot, Linear, Notion, and anything else — and merge the output files into your filesystem.

Mount anywhere

Mount the filesystem into local VMs, remote sandboxes like Modal or E2B, or directly into Claude Code. Works anywhere you can mount a network filesystem.

Fine-grained access control

Control exactly how your agent interacts with data. Per mount, choose your access mode:

read-only

Agent reads files but can't modify anything

read-write

Agent reads and writes directly to your filesystem

overlay

Agent works on a copy — you decide what syncs back

clawbox.yaml
sources:
  hubspot:
    script: "./scripts/sync-hubspot.sh"
    schedule: "0 7 * * *"
    dest: "./sales/hubspot-tasks.md"
  gmail:
    script: "./scripts/sync-inbox.sh"
    schedule: "*/15 * * * *"
    dest: "./sales/inbox-leads.md"

mounts:
  claude-code:
    path: "./sales"
    access: "read-write"
  modal-sandbox:
    path: "./engineering"
    access: "overlay"

Roadmap

What's next

We're building more and would love feedback from power users of coding and computer agents.

Versioning & branching

Version every file change. Branch your filesystem like git — per directory.

Built-in web scraping

Turn webpages into parts of your filesystem. The web as a data source, automatically.

Smart data routing

Get the right data to your agent at the right time — context-aware, not just cron-based.

Security controls

Prompt injection protection, sensitive data filtering, and audit logs for what your agent sees and writes.


Get your own agent filesystem

Claim your namespace

Your agent gets /home/{you} on the global filesystem