rebyte-langchain 0.0.12

Python langchain-like client library for the Rebyte API

rebyte-langchain

rebyte-langchain is a Python library to access the ReByte Agent REST API. This library enables you to use rebyte in a langchain-like style.

Features

1. Generate streaming or non-streaming output.
2. Use async or sync method.
3. Compatiable with langchain callback functions.
4. Support stateful agent memory with session_id.

Install

pip install rebyte-langchain

Prerequisites

ReByte API Key

To get your ReByte API key, follow these steps:

1. Go to the ReByte website and sign up for an account if you haven't already.
2. Once you're logged in, go to Settings > API Keys.
3. Generate a new API key by clicking on the "Generate" button.
4. Prepare to use it during configuration.

ReByte Agent Creation

1. You can follow the tutorial in the webpage ReByte Quick Start to create your own agent. Or you can use agents in ReByte Community. Open an agent and clone to your personal project.

2. Deploy the agent and get the project_id and agent_id in the following format: https://rebyte.ai/api/sdk/p/{project_id}/a/{agent_id}/r.

Simple Demo

# import packages
from rebyte_langchain.rebyte_langchain import RebyteEndpoint
import os
import asyncio
from langchain.schema.messages import HumanMessage
from langchain.callbacks.streaming_stdout import StreamingStdOutCallbackHandler

# required keys & config
# You can set REBYTE_API_KEY as an environment variable 
os.environ['REBYTE_API_KEY'] = "<rebyte_api_key>"
# Or pass it as an argument to the RebyteEndpoint constructor
test_rebyte_api_key:str = "<rebyte_api_key>"

# create an agent on rebyte platform and get the project_id and agent_id
# https://rebyte.ai/api/sdk/p/{test_project_id}/a/{test_agent_id}/r
# An example agent for Mark Zuckerberg: https://rebyte.ai/p/d4e521a67bb8189c2189/callable/a38ec8c60c3925696385/editor
test_project_id:str = "d4e521a67bb8189c2189"
test_agent_id:str = "a38ec8c60c3925696385"

# You may use any string as session_id and try to avoid duplicate session_ids
# Note that you must set session_id if you want to enable stateful actions, such as threads (aka, memory), in your agent.
# Or you can leave it as None when the agent has no stateful actions.
test_session_id:str = None

# test input
human_messages = [HumanMessage(content="Who are you")]

def generate(stream = False):
  model = RebyteEndpoint(
    rebyte_api_key=test_rebyte_api_key,
    project_id=test_project_id,
    agent_id=test_agent_id,
    session_id=test_session_id,
    streaming=stream
  )
  response = model.generate([human_messages],
                            callbacks=[StreamingStdOutCallbackHandler()]
                            )
  return response

async def agenerate(stream = False):
  model = RebyteEndpoint(
    rebyte_api_key=test_rebyte_api_key,
    project_id=test_project_id,
    agent_id=test_agent_id,
    session_id=test_session_id,
    streaming=stream
  )
  response = await model.agenerate(messages=
                                   [human_messages],
                                   callbacks=[StreamingStdOutCallbackHandler()]
                                   )
  return response

if __name__ == "__main__":
  print("\n\nTEST GENERATE\n\n")
  response = generate(stream=True)

  print("\n\nTEST AGENERATE\n\n")
  loop = asyncio.get_event_loop()
  loop.run_until_complete(agenerate(stream=True))

Please see more examples in main.py and example.ipynb.

Documentation

More information can be found on the ReByte Documentation Site.