aser

Aser is a modular AI agent framework that lets you assemble an agent with a few lines of code. It supports wiring together various tools, memories, knowledge sources, and optional tracing to build sophisticated agents capable of reasoning, tool usage, and task execution via the MCP (Model Context Protocol) workflow. The framework can be instantiated directly in Python, and you can customize the agent with different models (e.g., GPT-4 variants) and a library of built-in or user-defined components to suit your use case. The MCP integration enables structured task planning and execution across multiple modules, enabling the agent to context-switch between tools, fetch knowledge, and maintain state across conversations.

Prerequisites:

• Python 3.8+ and pip
• Optional: a Python virtual environment

Install from PyPI:

pip install aser

Or clone the repository and install locally:

git clone https://github.com/AmeNetwork/aser.git
cd aser
pip install -r requirements.txt

If you clone the repo and plan to run examples from source, install in editable mode:

pip install -e .

Reference environment setup (example):

# Create a local .env with your credentials/settings
cp .env.example .env
# Edit .env to include MODEL_BASE_URL, MODEL_KEY, etc.

Running a basic example (from Python):

python - <<'PY'
from aser.agent import Agent
agent = Agent(name="aser agent", model="gpt-4.1-mini")
response = agent.chat("what's bitcoin?")
print(response)
PY

Environment variables: Aser relies on your chosen language model provider. Set MODEL_BASE_URL and MODEL_KEY (or equivalent) in a .env file as described in the README. If you plan to use tools, memory, or knowledge retrieval modules, ensure you configure their respective settings in your environment or via the Python code. When running from source, use pip install -e . to enable import of the local aser package. If you encounter model or network errors, verify your API keys, base URLs, and HTTP access. For MCP-specific usage, you can enable tracing and instrument your agent to visualize decision steps and tool usage for debugging.