Skip to main content
Use a Tensorlake file system when many agents need the same files at a stable path. Put operating manuals, skills, configs, test fixtures, or binary tools in a repository. Update the repository from a laptop, CI job, or backend service. Agents mount a branch read-only. When the branch moves, following mounts refresh changed paths automatically.

​
Pattern

  1. Store shared files in a Tensorlake repository.
  2. Publish updates from outside the sandbox with Git or the Repository SDKs.
  3. Mount the repository into agents as a read-only directory.
  4. Follow a branch for automatic distribution, or pin a commit for fixed releases.

​
Create an Asset Repository

$ tl git create agent-assets --default-branch main
created https://git.tensorlake.ai/project_9f3c2a1b/agent-assets
Keep the layout simple and stable: 
agent-assets/
  manuals/
  skills/
  bin/
  configs/
Agents can refer to paths like /opt/agent-assets/manuals/operator.md, /opt/agent-assets/skills/research/SKILL.md, or /opt/agent-assets/bin/validator.

​
Publish From Outside a Sandbox

The producer does not need a sandbox. It can be a developer machine, CI job, release service, or control plane process. 
TOKEN=$(tl git token --repo agent-assets --json | jq -r .token)
git clone https://t:$TOKEN@git.tensorlake.ai/project_9f3c2a1b/agent-assets

cp -R ./manuals ./skills ./bin ./configs agent-assets/
chmod +x agent-assets/bin/validator

git -C agent-assets add .
git -C agent-assets commit -m "publish agent assets"
git -C agent-assets push origin main
push_worktree and pushWorktree skip .git, honor .gitignore, preserve symlinks, and preserve executable bits on regular files.

​
Mount Into Agents

Mount the asset branch read-only at a predictable path: 
$ tl fs mount agent-assets:main /opt/agent-assets --mode ro
Mounted agent-assets:main at /opt/agent-assets (workspace 7c2e9f0a1b3d, read-only, follows the branch)
New sandboxes read the current branch tip. Running following read-only mounts refresh as main moves, so updated manuals, skills, configs, and tools appear without rebuilding sandbox images. Agents should write outputs to a separate writable workspace. Keep shared assets read-only so every agent sees the same source files.

​
Version Releases

Use branches as channels: 
tl fs mount agent-assets:stable /opt/agent-assets --mode ro
Move stable when you want following mounts to receive a release. Branch activity records who moved the branch and which commit became the new tip. Use a commit when a run must be reproducible: 
tl fs mount agent-assets:9f2a1c8e4d6b1a0f3c7e9d2b8a4f6c1e0d3b7a99 /opt/agent-assets --mode ro
Pinned mounts stay fixed for the lifetime of the sandbox.

​
Binary Tools

Put tools under a stable directory such as bin/ and commit them with the executable bit set: 
chmod +x agent-assets/bin/validator
git -C agent-assets add bin/validator
git -C agent-assets commit -m "add validator tool"
git -C agent-assets push origin main
Agents can call the tool directly: 
/opt/agent-assets/bin/validator --input /work/result.json

​
Choose a Mount

NeedUse
Roll out the latest manuals, skills, or tools to many agentsFollowing read-only mount
Keep an eval, benchmark, or release run fixedPinned read-only mount
Publish assets from CI or an external serviceGit push or Repository SDK push_worktree
Let an agent create or modify filesWritable workspace

​
Next Steps

Read-only Mounts

Choose between following branches and pinned commits.

Repository SDKs

Push local files and inspect repository state from Python or TypeScript.

Git Workflows

Publish the same assets with ordinary Git commands.

Writable Workspaces

Give agents a separate place to write outputs.