agent-builder-api

Agent Builder API

🚀 Quick Start

You can start the API server using docker containers or manually cloning and building this repository.

Manual setup: Clone this repository in your local machine and start the python FAST API server. Optionally, install and set up Mongo DB.

Dev Container Setup: Dev Containers allow you to automate the environment setup.
Use this setup to install and run container services in an isolated environment with extensions preinstalled.
You can also open GitHub Codespaces in a remote environment/browser using secrets to pass Model API keys.

Manual Setup

1. Set up a Python virtualenv and install dependencies

   python -m venv --prompt agent-builder-api venv
   source venv/Scripts/activate
   # venv/Scripts/activate (Windows PS)
   pip install -r requirements.txt
   

2. Set the model name and API key in .env file

   OPENAI_API_KEY="sk----"
   MODEL_NAME="openai"
   

3. Start the server in the new terminal

   python -m agentbuilder.main
   

[Optional] Start using Poetry

For fine-grained dependency management, use Poetry to pick and choose dependency packs based on your LLM model provider and tool features.

1. Follow the Offical Instruction Guide to install Poetry.

2. Pick and choose dependency packs to install.

   poetry install --extras "mcp openai gemini cohere anthropic mongodb togetherai ollama vectordb langgraph guardrails ui"
   

3. Set the model name and API key in .env file

   OPENAI_API_KEY="sk----"
   MODEL_NAME="openai"
   

4. Start the server in the new terminal

   poetry run start-server
   

[!NOTE]
Poetry will create a virtual environment for us.

MCP servers

• Local MCP servers can be added in MCP server folder

• Multiple MCP clients can be configured in MPC client file

  mcp_servers={
        "math": {
            "command": sys.executable,
            "args": [str(current_dir / "servers" / "mcp_math.py")],
            "transport": "stdio",
          },
     }
  

Vibe coding Agent

Projects are run inside a docker container (alpine_runner).
Pre-defined docker compose file can be used to bind host volumes and application ports.

• Start the docker container using the compose file.

• All commands executed by the model can be viewed using docker logs.

  docker logs alpine_runner -f

• Project related information can be configured in openvibe.py

  @mcp.resource("config://app")
  def get_config() -> any:
     """Static configuration data"""
     return  {
        "project_path": "~/projects/appname",
        "project_docs_path": "~/projects/appname/docs"
     }
  

CLI mode

• Add alias to chat.sh & api_client.sh script in shell config (/.zshrc,/.bashrc)

  alias chat=/pathto/chat.sh
  alias api=/pathto/api_client.sh
  

• Use the help command option to identify different options

  chat -h

Custom UI (Gradio)

• Start gradio ui

  poetry run start-ui
  # OR
  gradio ./agentbuilder/ui/app.py
  

• More agents can be enabled by adding it in the dropdown choices - app.py

[Optional] Enable MongoDB

By default, data is stored as JSON files. Enable storage in Mongo DB by setting url using the environment variable.

MONGODB_URL="mongodb://localhost:27017/llmdb"

Dev Containers Setup

1. Enable Dev Containers in vscode by following the steps in official documentation.

2. Click on the badge below to run the services in an isolated container environment in a local machine.

   This will clone the repo and start the API and Mongo DB container services.

   

[!TIP]
Use URL mongodb://mongodb:27017/llmdb in Mongo DB vscode extension to view storage data.

4. Execute F1 > Dev Containers: Attach to Running Container.. and select agent-builder-container.

5. Set the model name and API key in .env file

   OPENAI_API_KEY="sk----"
   MODEL_NAME="openai"
   

Customization

Dependency packs and environment configuration

Dependency packs allow fine-grained package installations based on your requirements.
Use the environment variable EXTRA_DEPS in the docker-compose file to update.

install-extra-deps.sh script can be used in dev container mode if docker-compose is not available.

For example, the below environment configuration will install dependencies for the Gemini model,
Mongo DB, Langchain Graph, and VectorDB

EXTRA_DEPS: "gemini,mongodb,langgraph,vectordb"

[!TIP]
Start with a basic dependency pack to support your model and add other features incrementally

The following models are supported by its dependency pack

Model	Dependency pack	ENV key name
OPEN AI	openai	OPENAI_API_KEY
GEMINI	gemini	GOOGLE_API_KEY
COHERE	cohere	COHERE_API_KEY
ANTHROPIC	anthropic	ANTHROPIC_API_KEY

Some pre-configured tools require extra dependencies or API keys to get enabled.

Tool	Dependency pack	ENV key name
internet_search	-	TAVILY_API_KEY
vectorstore_search	vectordb	EMBED_MODEL_NAME

Adding Tools

Add custom Tools or Toolkits using tool factory module (agentbuilder/factory/tool_factory).

1. Create your tool

   agentbuilder/tools/my_custom_tool.py

   
   from pathlib import Path
   from langchain_core.tools import tool
   from pydantic import BaseModel, Field
   
   @tool
   def my_custom_tool(a: int, b: int):
   """Custom Tool Description"""
   return a + b
   
   my_custom_tool.name="custom_tool_name"
   my_custom_tool.description="Custom Tool Description"
   
   class field_inputs(BaseModel):
   a: int = Field(description="First input")
   b: int = Field(description="Second input")
   
   my_custom_tool.args_schema = sum_inputs
   my_custom_tool.metadata= {"file_path": str(Path(__file__).absolute())}
   

2. Add your tool in get_all_tools method in tool_factory module.

   agentbuilder/factory/tool_factory.py

   def get_all_tools()->Sequence[BaseTool]:
   
       return get_vectordb_tools()+ get_websearch_tools() + json_tools + [
               directly_answer_tool,
               weather_clothing_tool,
               temperature_tool,
               temperature_sensor_tool,
               sum_tool,greeting_tool,
               git_diff_tool,
               repl_tool,
   +           my_custom_tool
               ]
   

Adding Agents

Agents can be created using Extension UI or declared in code.
Add your agents using the Agent Factory Module (agentbuilder/factory/agent_factory).

1. Create your agent

   def my_agent():
     return AgentParams(
           name="my_agent",
           preamble= "You are a powerful agent that uses tools to answer Human questions",
           tools=  ["my_custom_tool"],
           agent_type= 'tool_calling'
     )
   

2. Add your agent in get_all_agents method.

   def get_all_agents():
   return [
           default_agent(),
           weather_agent(),
           python_agent(),
           git_agent(),
   +       my_agent()
   ]
   

Custom Agent Builder and Graphs

Customize your Agent Workflow using custom prompts and graphs.
Filter the agent using the agent name to apply customizations.

For example, the following code applies graph builder workflow for the Agent named "graph_agent"

def get_agent_builder(params:AgentBuilderParams):
    agent_name= params.name
    match agent_name:
        case "graph_agent":
            from agentbuilder.agents.BaseGraphAgentBuilder import BaseGraphAgentBuilder
            return BaseGraphAgentBuilder(params)
        case _:
            return BaseAgentBuilder(params)

[!IMPORTANT]
Dependency pack "langgraph" needs to be installed for BaseGraphAgentBuilder.

Configuring models

Update model configuration using environment variables.

Supports {Provider}/{ModelName} format

• Gemini

  Create API keys https://aistudio.google.com/app/apikey

   MODEL_NAME="gemini/gemini-pro"
   EMBED_MODEL_NAME="gemini/embedding-001"
   GOOGLE_API_KEY=<GOOGLE_API_KEY>
  

• Cohere

  Create API keys https://dashboard.cohere.com/api-keys

   MODEL_NAME="cohere/command"
   EMBED_MODEL_NAME="cohere/embed-english-v3.0"
   COHERE_API_KEY=<COHERE_API_KEY>
  

• Open AI

  Create API keys https://platform.openai.com/docs/quickstart/account-setup

   MODEL_NAME="openai/gpt-4o"
   EMBED_MODEL_NAME="openai/text-embedding-3-large"
   OPENAI_API_KEY=<OPENAI_API_KEY>
  

• Anthropic

  Create API keys https://www.anthropic.com/ and https://www.voyageai.com/

   MODEL_NAME="anthropic/claude-3-opus-20240229"
   EMBED_MODEL_NAME="voyageai/voyage-2"
   ANTHROPIC_API_KEY=<ANTHROPIC_API_KEY>
   VOYAGE_API_KEY=<VOYAGE_API_KEY>
  

• Ollama

  Use local models for function calls.

  [!TIP]
  Use JSON chat agent type for better compatibility with local models.

  Install Ollama and pull the model.

  ollama pull mistral:v0.3
  

  Set environment variable.

  MODEL_NAME="ollama/mistral:v0.3"