deepagents-azure-blob-backend 0.3.0

Azure Blob Storage filesystem backend for LangChain Deep Agents

deepagents-azure-blob-backend

Azure Blob Storage filesystem backend for LangChain Deep Agents.

Deep Agents exposes a BackendProtocol — a pluggable interface for file operations (read, write, edit, ls, glob, grep) that the agent uses as its virtual filesystem. This package provides an Azure Blob Storage implementation of that interface.

Installation

pip install deepagents-azure-blob-backend

Or with uv:

uv add deepagents-azure-blob-backend

Quick Start

import asyncio
from deepagents import create_deep_agent
from deepagents_azure_blob_backend import AzureBlobBackend, AzureBlobConfig

async def main():
    config = AzureBlobConfig(
        account_url="https://<your-account>.blob.core.windows.net",
        container_name="agent-workspace",
        prefix="session-001/",
    )
    backend = AzureBlobBackend(config)

    agent = create_deep_agent(backend=backend)

    result = await agent.ainvoke(
        {"messages": [{"role": "user", "content": "Create a hello world script at /hello.py"}]},
    )
    print(result["messages"][-1].content)
    await backend.close()

asyncio.run(main())

Configuration

from deepagents_azure_blob_backend import AzureBlobConfig

config = AzureBlobConfig(
    account_url="https://<account>.blob.core.windows.net",
    container_name="my-container",
    prefix="agent-workspace/",     # Namespace isolation for multi-agent setups
    max_concurrency=8,             # Parallel blob ops for grep/glob
    encoding="utf-8",
)

Authentication

AzureBlobConfig supports five mutually exclusive authentication methods. Set at most one credential source — if none is provided, DefaultAzureCredential is used automatically. account_url is required for all methods except connection string:

# 1. Connection string (e.g., Azurite or Azure Portal)
config = AzureBlobConfig(
    container_name="test",
    connection_string="UseDevelopmentStorage=true",
)

# 2. Account key
config = AzureBlobConfig(
    account_url="https://<account>.blob.core.windows.net",
    container_name="my-container",
    account_key="your-storage-account-key",
)

# 3. SAS token
config = AzureBlobConfig(
    account_url="https://<account>.blob.core.windows.net",
    container_name="my-container",
    sas_token="sv=2021-06-08&ss=b&srt=co&sp=rwdlacitfx&se=...",
)

# 4. Credential object (any Azure credential)
from azure.identity.aio import ClientSecretCredential
config = AzureBlobConfig(
    account_url="https://<account>.blob.core.windows.net",
    container_name="my-container",
    credential=ClientSecretCredential(tenant_id, client_id, client_secret),
)

# 5. Default (AAD) — omit all credential fields
config = AzureBlobConfig(
    account_url="https://<account>.blob.core.windows.net",
    container_name="my-container",
)

The default path uses DefaultAzureCredential, which supports az login, managed identity, workload identity federation (OIDC), and environment variables.

Supported Operations

All methods from BackendProtocol:

Method	Async	Description
ls_info(path)	als_info	List directory with synthesized subdirectories
read(path, offset, limit)	aread	Read file with line numbers
write(path, content)	awrite	Create new file (errors if exists)
edit(path, old, new)	aedit	String replacement editing
glob_info(pattern, path)	aglob_info	Glob pattern file matching
grep_raw(pattern, path, glob)	agrep_raw	Literal text search across files
upload_files(files)	aupload_files	Batch binary upload
download_files(paths)	adownload_files	Batch binary download

Development

# Install dev dependencies
uv sync --group dev

# Run unit tests
uv run pytest tests/test_backend_unit.py -v

# Run integration tests (requires Azurite)
docker run -p 10000:10000 mcr.microsoft.com/azure-storage/azurite azurite-blob --skipApiVersionCheck --blobHost 0.0.0.0
uv run pytest tests/test_backend_integration.py -v

# Lint and format
uv run ruff check .
uv run ruff format .

# Type check
uv run ty check

See the examples/ folder for runnable scripts with setup instructions.

License

MIT