aisberg 0.4.0

Aisberg SDK for Python - A simple and powerful SDK to interact with the Aisberg API

Aisberg Python SDK

Aisberg SDK for Python is a robust and easy-to-use library for interacting with the Aisberg API.
It provides synchronous and asynchronous clients, advanced module abstractions, and built-in support for conversational LLM workflows, collections, embeddings, and more.

---

Features

• Sync & Async clients: Use with regular scripts or async frameworks
• Auto tool execution for LLM flows (tool calls, results integration)
• Modular architecture: Collections, chat, models, workflows, embeddings, and more
• Strong typing via Pydantic models
• Environment-based configuration (supports .env files and system environment variables)
• Context manager support for easy resource management
• Custom tool registration: Easily extend LLM capabilities with your own functions
• Document Parsing: Parse documents into structured data (e.g., JSON, CSV, PNG, PDF, etc.)

---

Installation

pip install aisberg

Or, for local development:

git clone https://your.git.repo/aisberg.git
cd aisberg
pip install -e .

---

Quickstart

1. Configure your API key and base URL

You can set them as environment variables, or in a .env file:

AISBERG_API_KEY=...
AISBERG_BASE_URL=https://url
AISBERG_TIMEOUT=...  # Optional, default is 180 seconds (3 minutes)

In order to use the Document Parsing feature, you also need to set the S3 credentials - ask the FreePro team for these:

S3_ACCESS_KEY_ID=...
S3_SECRET_ACCESS_KEY=...
S3_ENDPOINT=https://s3.endpoint

2. Synchronous Usage

from aisberg import AisbergClient

with AisbergClient() as client:
    me = client.me.info()
    print(me)

    chat_response = client.chat.complete(
        input="Bonjour, qui es-tu ?",
        model="llm-aisberg",
    )
    print(chat_response.choices[0].message.content)

3. Asynchronous Usage

import asyncio
from aisberg import AisbergAsyncClient


async def main():
    async with AisbergAsyncClient() as client:
        await client.initialize()
        me = await client.me.info()
        print(me)

        chat_response = await client.chat.complete(
            input="Hello, who are you?",
            model="llm-aisberg",
        )
        print(chat_response.choices[0].message.content)


asyncio.run(main())

---

Modules

• client.me — User/account info
• client.chat — Conversational LLM completions and streaming
• client.collections — Manage data collections
• client.embeddings — Encode texts to embeddings
• client.models — Model discovery & info
• client.workflows — Workflow management & execution
• client.tools — Register and execute tools for LLM tool calls
• client.documents — Document parsing and management

Each module is available both in the sync and async clients with similar APIs.

---

Tool Calls and Automatic Execution

The SDK supports tool-augmented LLM workflows. Register your own functions for use in chat:

def my_tool(args):
    # Custom business logic
    return {"result": "tool output"}


client.tools.register("my_tool", my_tool)
response = client.chat.complete(
    input="Use the tool please.",
    model="llm-aisberg",
    tools=[{"name": "my_tool", ...}],
    auto_execute_tools=True,
)

---

Advanced Usage

Custom Configuration

You can override configuration when instantiating the client:

client = AisbergClient(
    api_key="...",
    base_url="https://url",
    timeout=60,
)

Environment Variables Supported

• AISBERG_API_KEY
• AISBERG_BASE_URL
• AISBERG_TIMEOUT (optional)
• S3_ACCESS_KEY_ID (for document parsing)(optional)
• S3_SECRET_ACCESS_KEY (for document parsing)(optional)
• S3_ENDPOINT (for document parsing)(optional)

Using in a Context Manager

Both clients are context manager compatible:

with AisbergClient() as client:
    # Sync usage
    ...

async with AisbergAsyncClient() as client:
    # Async usage
    ...

---

License

Private License — Not for public distribution or resale.

For enterprise/commercial use, please contact Mathis Lambert or Free Pro.

---

Authors

• Mathis Lambert
• Free Pro

---

Support

For support, bug reports, or feature requests, please contact your technical representative.