🐝 Bee Agent Framework Starter

This starter template lets you quickly start working with the Bee Agent Framework in a second.

📚 See the documentation to learn more.

✨ Key Features

• 🔒 Safely execute an arbitrary Python Code via Bee Code Interpreter.
• 🔎 Get complete visibility into agents’ decisions using our MLFlow integration thanks to Bee Observe.
• 🚀 Fully fledged TypeScript project setup with linting and formatting.

📦 Requirements

• JavaScript runtime NodeJS > 18 (ideally installed via nvm).
• Container system like Rancher Desktop, Podman (VM must be rootfull machine) or Docker.
• LLM Provider either external WatsonX (OpenAI, Groq, …) or local ollama.

🛠️ Getting started

1. Clone this repository or use it as a template.
2. Install dependencies npm ci.
3. Configure your project by filling in missing values in the .env file (default LLM provider is locally hosted Ollama).
4. Run the agent npm run start src/agent.ts

To run an agent with a custom prompt, simply do this npm run start src/agent.ts <<< 'Hello Bee!'

🧪 More examples can be found here.

[!TIP]

To use Bee agent with Python Code Interpreter refer to the Code Interpreter section.

[!TIP]

To use Bee agent with Bee Observe refer to the Observability section.

🏗 Infrastructure

[!NOTE]

Docker distribution with support for compose is required, the following are supported:

• Docker
• Rancher - macOS users may want to use VZ instead of QEMU
• Podman - requires compose and rootful machine (if your current machine is rootless, please create a new one, also ensure you have enabled Docker compatibility mode).

🔒Code interpreter

The Bee Code Interpreter is a gRPC service that an agent uses to execute an arbitrary Python code safely.

Instructions

1. Start all services related to the Code Interpreter npm run infra:start --profile=code_interpreter
2. Run the agent npm run start src/agent_code_interpreter.ts

[!NOTE]

Code Interpreter runs on http://127.0.0.1:50051.

🔎 Observability

Get complete visibility of the agent’s inner workings via our observability stack.

• The MLFlow is used as UI for observability.
• The Bee Observe is the observability service (API) for gathering traces from Bee Agent Framework.
• The Bee Observe Connector is the observability connector that sends traces from Bee Agent Framework to Bee Observe.

Instructions

1. Start all services related to Bee Observe npm run infra:start --profile=observe
2. Run the agent npm run start src/agent_observe.ts
3. Upload the final trace to the MLFlow (the agent will print instructions on how to do that).
4. See visualized trace in MLFlow web application http://127.0.0.1:8080/#/experiments/0
   • Credentials: (user: admin, password: password)

[!TIP]

Configuration file is infra/observe/.env.docker.