Jina is an open-source framework to build, deploy & manage machine learning applications at scale. LangChain is another open-source framework for building applications powered by language models.

langchain-serve helps you deploy your LangChain apps on Jina AI Cloud in just a matter of seconds. You can now benefit from the scalability and serverless architecture of the cloud without sacrificing the ease and convenience of local development.

1. Refactor your code to function(s) that should be served with @serving decorator.
2. Create a requirements.txt file in your app directory to ensure all necessary dependencies are installed.
3. Run lc-serve deploy local app to test your API locally.
4. Run lc-serve deploy jcloud app to deploy on Jina AI Cloud.

• 🌎 RESTful/Websocket APIs with TLS certs in just 2 lines of code change.
• 🌊 Stream LLM interactions in real-time with Websockets.
• 👥 Enable human in the loop for your agents.
• 📄 Swagger UI, and OpenAPI spec included with your APIs.
• ⚡️ Serverless apps that scales automatically with your traffic.
• 📊 Builtin logging, monitoring, and traces for your APIs.
• 🤖 No need to change your code to manage APIs, or manage dockerfiles, or worry about infrastructure!

• 🔑 Authorize API endpoints
• 🛠️ Enable Streamlit playground deployment for your apps

If you have any feature requests or faced any issue, please let us know!

Let's first install langchain-serve using pip.

pip install langchain-serve

HITL for LangChain agents on production can be challenging since the agents are typically running on servers where humans don't have direct access. langchain-serve bridges this gap by enabling websocket APIs that allow for real-time interaction and feedback between the agent and a human operator.

Check out this example to see how you can enable HITL for your agents.

Let's build a custom agent using this example taken from LangChain documentation.

# app.py
from langchain.agents import ZeroShotAgent, Tool, AgentExecutor
from langchain import OpenAI, SerpAPIWrapper, LLMChain

search = SerpAPIWrapper()
tools = [
    Tool(
        name = "Search",
        func=search.run,
        description="useful for when you need to answer questions about current events"
    )
]

prefix = """Answer the following questions as best you can, but speaking as a pirate might speak. You have access to the following tools:"""
suffix = """Begin! Remember to speak as a pirate when giving your final answer. Use lots of "Args"

Question: {input}
{agent_scratchpad}"""

prompt = ZeroShotAgent.create_prompt(
    tools, 
    prefix=prefix, 
    suffix=suffix, 
    input_variables=["input", "agent_scratchpad"]
)

llm_chain = LLMChain(llm=OpenAI(temperature=0), prompt=prompt)
tool_names = [tool.name for tool in tools]
agent = ZeroShotAgent(llm_chain=llm_chain, allowed_tools=tool_names)
agent_executor = AgentExecutor.from_agent_and_tools(agent=agent, tools=tools, verbose=True)
agent_executor.run("How many people live in canada as of 2023?")

> Entering new AgentExecutor chain...
Thought: I need to find out the population of Canada
Action: Search
Action Input: Population of Canada 2023
Observation: The current population of Canada is 38,610,447 as of Saturday, February 18, 2023, based on Worldometer elaboration of the latest United Nations data. Canada 2020 population is estimated at 37,742,154 people at mid year according to UN data.
Thought: I now know the final answer
Final Answer: Arrr, Canada be havin' 38,610,447 scallywags livin' there as of 2023!

> Finished chain.

Refactor your code to function(s) that should be served with @serving decorator

# app.py
from langchain import LLMChain, OpenAI, SerpAPIWrapper
from langchain.agents import AgentExecutor, Tool, ZeroShotAgent

from lcserve import serving


@serving
def ask(input: str) -> str:
    search = SerpAPIWrapper()
    tools = [
        Tool(
            name="Search",
            func=search.run,
            description="useful for when you need to answer questions about current events",
        )
    ]
    prefix = """Answer the following questions as best you can, but speaking as a pirate might speak. You have access to the following tools:"""
    suffix = """Begin! Remember to speak as a pirate when giving your final answer. Use lots of "Args"

    Question: {input}
    {agent_scratchpad}"""

    prompt = ZeroShotAgent.create_prompt(
        tools,
        prefix=prefix,
        suffix=suffix,
        input_variables=["input", "agent_scratchpad"],
    )

    print(prompt.template)

    llm_chain = LLMChain(llm=OpenAI(temperature=0), prompt=prompt)
    tool_names = [tool.name for tool in tools]
    agent = ZeroShotAgent(llm_chain=llm_chain, allowed_tools=tool_names)

    agent_executor = AgentExecutor.from_agent_and_tools(
        agent=agent, tools=tools, verbose=True
    )

    return agent_executor.run(input)

if __name__ == "__main__":
    ask('How many people live in canada as of 2023?')

• We moved our code to an ask function.
• Added type hints to the function parameters (input and output), so API definition can be generated.
• Imported from lcserve import serving and added @serving decorator to the ask function.
• Added if __name__ == "__main__": block to test the function locally.

Create a requirements.txt file in your app directory to ensure all necessary dependencies are installed.

# requirements.txt
openai
google-search-results

Run lc-serve deploy local app to test your API locally.

app is the name of the module that contains the ask function.

lc-serve deploy local app

────────────────────────────────────────────────────────────────────────────────────────────────────── 🎉 Flow is ready to serve! ───────────────────────────────────────────────────────────────────────────────────────────────────────
╭──────────────────────── 🔗 Endpoint ────────────────────────╮
│  ⛓   Protocol                                         HTTP  │
│  🏠     Local                                 0.0.0.0:8080  │
│  🔒   Private                          192.168.29.185:8080  │
│  🌍    Public  2405:201:d007:e8e7:2c33:cf8e:ed66:2018:8080  │
╰─────────────────────────────────────────────────────────────╯
╭─────────── 💎 HTTP extension ────────────╮
│  💬          Swagger UI        .../docs  │
│  📚               Redoc       .../redoc  │
╰──────────────────────────────────────────╯

Let's open the Swagger UI to test our API locally. With Try it out button, we can test our API with different inputs.

Let's test our local API with How many people live in canada as of 2023? input with a cURL command.

curl -X 'POST' \
  'http://localhost:8080/ask' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  -d '{
  "input": "How many people live in canada as of 2023?",
  "envs": {
    "OPENAI_API_KEY": "'"${OPENAI_API_KEY}"'",
    "SERPAPI_API_KEY": "'"${SERPAPI_API_KEY}"'"
  }
}'

{
  "result": "Arrr, there be 38,645,670 people livin' in Canada as of 2023!",
  "error": "",
  "stdout": "Answer the following questions as best you can, but speaking as a pirate might speak. You have access to the following tools:\n\nSearch: useful for when you need to answer questions about current events\n\nUse the following format:\n\nQuestion: the input question you must answer\nThought: you should always think about what to do\nAction: the action to take, should be one of [Search]\nAction Input: the input to the action\nObservation: the result of the action\n... (this Thought/Action/Action Input/Observation can repeat N times)\nThought: I now know the final answer\nFinal Answer: the final answer to the original input question\n\nBegin! Remember to speak as a pirate when giving your final answer. Use lots of \"Args\"\n\n    Question: {input}\n    {agent_scratchpad}\n\n\n\u001b[1m> Entering new AgentExecutor chain...\u001b[0m\n\u001b[32;1m\u001b[1;3m\nThought: I need to find out how many people live in Canada\nAction: Search\nAction Input: How many people live in Canada as of 2023\u001b[0m\nObservation: \u001b[36;1m\u001b[1;3mThe current population of Canada is 38,645,670 as of Wednesday, March 29, 2023, based on Worldometer elaboration of the latest United Nations data.\u001b[0m\nThought:\u001b[32;1m\u001b[1;3m I now know the final answer\nFinal Answer: Arrr, there be 38,645,670 people livin' in Canada as of 2023!\u001b[0m\n\n\u001b[1m> Finished chain.\u001b[0m"
}

• POST /ask is generated from ask function defined in app.py.
• input is an argrment defined in ask function.
• envs is a dictionary of environment variables that will be passed to all the functions decorated with @serving decorator.
• return type of ask function is str. So, result would carry the return value of ask function.
• If there is an error, error would carry the error message.
• stdout would carry the output of the function decorated with @serving decorator.

Run lc-serve deploy jcloud app to deploy your API to Jina AI Cloud.

# Login to Jina AI Cloud
jina auth login

# Deploy your app to Jina AI Cloud
lc-serve deploy jcloud app

⠇ Pushing `/tmp/tmp7kt5qqrn` ...🔐 You are logged in to Jina AI as ***. To log out, use jina auth logout.
╭────────────────────────── Published ───────────────────────────╮
│                                                                │
│   📛 Name           n-64a15                                    │
│   🔗 Jina Hub URL   https://cloud.jina.ai/executor/6p1zio87/   │
│   👀 Visibility     public                                     │
│                                                                │
╰────────────────────────────────────────────────────────────────╯
╭─────────────────────── 🎉 Flow is available! ───────────────────────╮
│                                                                     │
│   ID               langchain-ee4aef57d9                             │
│   Gateway (Http)   https://langchain-ee4aef57d9-http.wolf.jina.ai   │
│   Dashboard        https://dashboard.wolf.jina.ai/flow/ee4aef57d9   │
│                                                                     │
╰─────────────────────────────────────────────────────────────────────╯

╭──────────────┬─────────────────────────────────────────────────────────────╮
│ AppID        │                    langchain-ee4aef57d9                     │
├──────────────┼─────────────────────────────────────────────────────────────┤
│ Phase        │                           Serving                           │
├──────────────┼─────────────────────────────────────────────────────────────┤
│ Endpoint     │       https://langchain-ee4aef57d9-http.wolf.jina.ai        │
├──────────────┼─────────────────────────────────────────────────────────────┤
│ Swagger UI   │     https://langchain-ee4aef57d9-http.wolf.jina.ai/docs     │
├──────────────┼─────────────────────────────────────────────────────────────┤
│ OpenAPI JSON │ https://langchain-ee4aef57d9-http.wolf.jina.ai/openapi.json │
╰──────────────┴─────────────────────────────────────────────────────────────╯

Let's open the Swagger UI to test our API on Jina AI Cloud. With Try it out button, we can test our API with different inputs.

Let's test the API on JCloud with How many people live in canada as of 2023? input with a cURL command (Replace the Hostname with your own hostname):

curl -X 'POST' \
  'https://langchain-ee4aef57d9-http.wolf.jina.ai/ask' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  -d '{
  "input": "How many people live in canada as of 2023?",
  "envs": {
    "OPENAI_API_KEY": "'"${OPENAI_API_KEY}"'",
    "SERPAPI_API_KEY": "'"${SERPAPI_API_KEY}"'"
  }
}'

{
  "result": "Arrr, there be 38,645,670 people livin' in Canada as of 2023!",
  "error": "",
  "stdout": "Answer the following questions as best you can, but speaking as a pirate might speak. You have access to the following tools:\n\nSearch: useful for when you need to answer questions about current events\n\nUse the following format:\n\nQuestion: the input question you must answer\nThought: you should always think about what to do\nAction: the action to take, should be one of [Search]\nAction Input: the input to the action\nObservation: the result of the action\n... (this Thought/Action/Action Input/Observation can repeat N times)\nThought: I now know the final answer\nFinal Answer: the final answer to the original input question\n\nBegin! Remember to speak as a pirate when giving your final answer. Use lots of \"Args\"\n\n    Question: {input}\n    {agent_scratchpad}\n\n\n\u001b[1m> Entering new AgentExecutor chain...\u001b[0m\n\u001b[32;1m\u001b[1;3m\nThought: I need to find out how many people live in Canada\nAction: Search\nAction Input: How many people live in Canada as of 2023\u001b[0m\nObservation: \u001b[36;1m\u001b[1;3mThe current population of Canada is 38,645,670 as of Wednesday, March 29, 2023, based on Worldometer elaboration of the latest United Nations data.\u001b[0m\nThought:\u001b[32;1m\u001b[1;3m I now know the final answer\nFinal Answer: Arrr, there be 38,645,670 people livin' in Canada as of 2023!\u001b[0m\n\n\u001b[1m> Finished chain.\u001b[0m"
}

• In a matter of few seconds, we've deployed our API on Jina AI Cloud 🎉
• The API is serverless and scalable, so we can scale up the API to handle more requests.
• You might observe a delay in the first request, that's due to the warm-up time of the API. Subsequent requests will be faster.
• The API includes a Swagger UI and the OpenAPI specification, so it can be easily integrated with other services.
• Now, other agents can integrate with your agents on Jina AI Cloud thanks to the OpenAPI Agent 💡

• Serverless is not your thing?
• Do you want larger instances for your API?
• Looking for file uploads, or other data-in, data-out features?
• Do you want to setup custom authorization for your API?

📣 Got your attention? Join us on Slack and we'd be happy to help you out.

lc-serve is a CLI tool that helps you to deploy your agents on Jina AI Cloud.

LangChain agents use LLMs to determine the actions to be taken in what order. An action can either be using a tool and observing its output, or returning to the user. We've hosted a Streamlit Playground on Jina AI Cloud to interact with the agents, which accepts with following inputs:

• Agent Types: Choose from different agent types that Langchain supports.

• Tools: Choose from different tools that Langchain supports. Some tools may require an API token or other related arguments.

To use the playground, simply type your input in the text box provided to get the agent's output and chain of thought. Enjoy exploring Langchain's capabilities! In addition to streamlit, you can also use our RESTful APIs on the playground to interact with the agents.

export OPENAI_API_KEY=sk-***
export SERPAPI_API_KEY=***

curl -sX POST 'https://langchain.wolf.jina.ai/api/run' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  --data-raw '{
    "text": "Who is Leo DiCaprios girlfriend? What is her current age raised to the 0.43 power?",
    "parameters": {
        "tools": {
            "tool_names": ["serpapi", "llm-math"]
        },
        "agent": "zero-shot-react-description",
        "verbose": true
    },
    "envs": {
        "OPENAI_API_KEY": "'"${OPENAI_API_KEY}"'",
        "SERPAPI_API_KEY": "'"${SERPAPI_API_KEY}"'"
    }
}' | jq

{
  "result": "Camila Morrone is Leo DiCaprio's girlfriend, and her current age raised to the 0.43 power is 3.6261260611529527.",
  "chain_of_thought": "\u001b[1m> Entering new AgentExecutor chain...\u001b[0m\u001b[32;1m\u001b[1;3m I need to find out the name of Leo's girlfriend and then use the calculator to calculate her age to the 0.43 power.Action: SearchAction Input: Leo DiCaprio girlfriend\u001b[0mObservation: \u001b[36;1m\u001b[1;3mDiCaprio met actor Camila Morrone in December 2017, when she was 20 and he was 43. They were spotted at Coachella and went on multiple vacations together. Some reports suggested that DiCaprio was ready to ask Morrone to marry him. The couple made their red carpet debut at the 2020 Academy Awards.\u001b[0mThought:\u001b[32;1m\u001b[1;3m I need to use the calculator to calculate her age to the 0.43 powerAction: CalculatorAction Input: 20^0.43\u001b[0mObservation: \u001b[33;1m\u001b[1;3mAnswer: 3.6261260611529527\u001b[0mThought:\u001b[32;1m\u001b[1;3m I now know the final answerFinal Answer: Camila Morrone is Leo DiCaprio's girlfriend, and her current age raised to the 0.43 power is 3.6261260611529527.\u001b[0m\u001b[1m> Finished chain.\u001b[0m"
}

export OPENAI_API_KEY=sk-***
export SERPAPI_API_KEY=***

curl -sX POST 'https://langchain.wolf.jina.ai/api/run' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  --data-raw '{
    "text": "What is the hometown of the reigning mens U.S. Open champion?",
    "parameters": {
        "tools": {
            "tool_names": ["serpapi"]
        },
        "agent": "self-ask-with-search",
        "verbose": true
    },
    "envs": {
        "OPENAI_API_KEY": "'"${OPENAI_API_KEY}"'",
        "SERPAPI_API_KEY": "'"${SERPAPI_API_KEY}"'"
    }
}' | jq

{
  "result": "El Palmar, Murcia, Spain",
  "chain_of_thought": "\u001b[1m> Entering new AgentExecutor chain...\u001b[0m\u001b[32;1m\u001b[1;3m Yes.Follow up: Who is the reigning mens U.S. Open champion?\u001b[0mIntermediate answer: \u001b[36;1m\u001b[1;3mCarlos Alcaraz Garfia\u001b[0m\u001b[32;1m\u001b[1;3mFollow up: What is Carlos Alcaraz Garfia's hometown?\u001b[0mIntermediate answer: \u001b[36;1m\u001b[1;3mCarlos Alcaraz Garfia was born on May 5, 2003, in El Palmar, Murcia, Spain to parents Carlos Alcaraz González and Virginia Garfia Escandón. He has three siblings.\u001b[0m\u001b[32;1m\u001b[1;3mSo the final answer is: El Palmar, Murcia, Spain\u001b[0m\u001b[1m> Finished chain.\u001b[0m"
}

Chains in LangChain allow users to combine components to create a single, coherent application. With Jina,

• You can expose your Chain as RESTful/gRPC/WebSocket API.
• Enable Chains to deploy & scale separately from the rest of your application with the help of Executors.
• Deploy your Chain on Jina AI Cloud and get exclusive access to Agents on Jina AI Cloud (coming soon)