mirror of
https://github.com/langflow-ai/langflow.git
synced 2026-07-24 02:47:59 +08:00
70 lines
3.4 KiB
Plaintext
70 lines
3.4 KiB
Plaintext
---
|
|
title: File System
|
|
slug: /file-system
|
|
---
|
|
|
|
The **File System** component provides sandboxed filesystem access for agents.
|
|
It includes five file I/O tools for agents:
|
|
|
|
* `read_file`
|
|
* `write_file`
|
|
* `edit_file`
|
|
* `glob_search`
|
|
* `grep_search`
|
|
|
|
All operations are scoped to a base directory designated by the [`LANGFLOW_FS_TOOL_BASE_DIR`](/environment-variables) environment variable. The default location is `~/.langflow/fs_tool/fs_sandbox`.
|
|
To use a different directory, set the location in `LANGFLOW_FS_TOOL_BASE_DIR` in your `.env` file.
|
|
|
|
If [`LANGFLOW_AUTO_LOGIN`](/api-keys-and-authentication#langflow-auto-login) is set to `true`, all agents share a single workspace at `LANGFLOW_FS_TOOL_BASE_DIR/shared/`.
|
|
If `LANGFLOW_AUTO_LOGIN` is set to `false`, each authenticated user gets a private workspace at `LANGFLOW_FS_TOOL_BASE_DIR/users/HASH/`, where `HASH` is an opaque HMAC-derived identifier. Anonymous requests are refused.
|
|
`LANGFLOW_AUTO_LOGIN` defaults to `true`, so shared mode is the default.
|
|
|
|
Tool errors are not raised as an unhandled exception.
|
|
Tool errors are returned to the agent as structured JSON:
|
|
|
|
```json
|
|
{"error":"File not found: DOC.md","path":"DOC.md"}
|
|
```
|
|
|
|
## Use the File System component in a flow
|
|
|
|
1. Connect the **File System** component to an **Agent** component's **Tools** input.
|
|
2. In the **File System** component, enable **Tool Mode** to expose `read_file`, `write_file`, `edit_file`, `glob_search`, and `grep_search` to the agent.
|
|
3. Optionally, set a **Workspace Sub-path** to scope the agent to a specific sub-folder inside the sandbox.
|
|
The sub-folder is created automatically on first use.
|
|
Leave the field empty to give the agent access to the full sandbox root.
|
|
|
|
For example, in the **Workspace Sub-path** field, enter `projects/my-application` to scope all file operations to `LANGFLOW_FS_TOOL_BASE_DIR/shared/projects/my-application/`.
|
|
|
|
If `LANGFLOW_AUTO_LOGIN` is set to `false`, the scoped sub-folder for the agent is at `LANGFLOW_FS_TOOL_BASE_DIR/users/$HASH/projects/my-application`.
|
|
|
|
4. Optionally, select **Read Only** to disable `write_file` and `edit_file` operations.
|
|
5. In the **Playground**, send a message to test the connection. For example, `List all files in my workspace`.
|
|
|
|
The agent calls `glob_search` with `**/*` and returns the contents of the sandbox directory.
|
|
|
|
|
|
## File System parameters
|
|
|
|
| Name | Type | Description |
|
|
|------|------|-------------|
|
|
| `root_path` | String | Input parameter. Sub-folder inside the sandboxed workspace. Leave empty to use the workspace root. |
|
|
| `read_only` | Boolean | Input parameter. If `true`, `write_file` and `edit_file` are disabled and not registered with the agent. Default: `false`. |
|
|
| `metadata` | Data | Output parameter. A fixed-structure `JSON` object that describes the sandbox configuration at build time and is independent of any file operation. The fields are `root_path`, `read_only`, `tools_registered`, `auto_login`, `mode` (`shared`, `isolated`, or `refused`), `effective_root`, and `resolution_error`. |
|
|
|
|
## Docker deployment
|
|
|
|
When running Langflow in Docker, you must mount the sandbox directory as a persistent volume.
|
|
This ensures that the pepper file used to derive per-user namespace hashes survives container restarts.
|
|
|
|
```yaml
|
|
services:
|
|
langflow:
|
|
environment:
|
|
- LANGFLOW_FS_TOOL_BASE_DIR=/data/fs_sandbox
|
|
volumes:
|
|
- lfx-fs-data:/data/fs_sandbox
|
|
|
|
volumes:
|
|
lfx-fs-data:
|
|
``` |