Persistent Agent Volumes
On-demand Shared Persistent Volumes for Agents
Give coding agents, research workflows, and automation jobs a real workspace with POSIX semantics, durable state, and storage that follows the work wherever it runs. Ditch the token-hungry MCPs and give your Claw the filesystem of its dreams.
Mount it with `clawfs mount` over FUSE or NFS, launch an explicit NFS server with `clawfs serve`, or inject it through the preload shared library. One install gives you `clawfs up`, `clawfs mount`, and `clawfs serve`.
curl -fsSL https://clawfs.dev/install.sh | bashStart in minutes with the CLI, then scale up with team access, cloud storage, and enterprise controls. Need SSO, audit streams, managed acceleration, or advanced deployment support? See Enterprise.
Operator Flow
Need the practical FAQ or the filesystem deets? Read the FAQ.
Single CLI Command
Same Volume, Later
Why ClawFS
Agents keep filesystem state between invocations, so long-running work survives crashes, restarts, and rescheduling.
Mount existing state instead of rebuilding toolchains, recloning repos, or rehydrating caches every time a worker starts.
Run on-prem or across cloud environments while keeping a single durable volume identity for the agent.
Keep large working sets in the filesystem instead of squeezing them into prompts, summaries, or brittle custom memory layers.
Multiple agents can attach to the same namespace for review, continuation, or specialization without export/import glue.
Adopt ClawFS without rebuilding your stack: install the CLI, mount a workspace, and return to it whenever the job resumes.
Good Fit
Keep repos, build outputs, package caches, and branch state across task sessions.
Preserve scraped corpora, experiment artifacts, notebooks, and checkpoints across iterative runs.
Let planner, executor, and reviewer agents share one durable workspace instead of passing tarballs around.
What Changes
Agents mount warm volumes with existing tools and data instead of spending minutes reconstructing context.
If a worker disappears, the volume does not. Resume from the same filesystem state on a fresh runtime.
Escalate work to a stronger model or human operator by reattaching the exact workspace, not a summary of it.
Cut repeated downloads, cache rebuilds, and duplicated scratch disks across fleets of short-lived workers.
Keep the data model stable while moving agents between laptop, VM, Kubernetes job, or hosted runner.
Use a real mount point and real files, which fits existing tools better than bespoke memory APIs.
Comparison Matrix
| Capability | ClawFS | Ephemeral VM Disk | Object Store Buckets | Managed NFS |
|---|---|---|---|---|
| Single-command volume up | Yes | No | No | Partial |
| Persistent state across agent runs | Yes | No | Yes | Yes |
| Native filesystem mount semantics | Yes | Yes | No | Yes |
| Easy handoff between agents | Yes | No | Partial | Partial |
| Portable across clouds and local runners | Yes | Limited | Yes | Limited |
| Checkpointing and restore path | Yes | Limited | Partial | Provider tooling |
| Best fit | Persistent agent workspaces | Disposable single-run jobs | Artifact blobs and archives | Traditional shared app storage |
ClawFS is aimed at agent runtime continuity: durable mounted state without adding a separate data handoff system.
ClawFS gives operators one durable substrate for manageable agents. Keep storage in your own environment, attach it when needed, and stop paying the cold-start penalty of rebuilding state from zero.
Install the CLI, create your first persistent volume, and reuse the same workspace across runs, machines, and teams.