freeact

I recently experimented using Gemini 2 Flash and Gemini 2 Flash Thinking with unrestricted code execution in a local sandbox and open sourced the results: https://github.com/gradion-ai/freeact. Here's an example that uses Gemini 2 Flash Thinking as agent that acts via code, executed in a sandbox based on IPython and Docker: https://gist.github.com/krasserm/dcdae47f85ee9922e3284953d07... Gemini's code execution environment is restricted to selected Python libraries like NumPy or SymPy and prevents the model from installing new packages, besides other limitations. While this may be useful for agentic applications in restricted environments, it may prevent agents from adapting to new environments, especially agents that write their actions in code (see CodeAct paper https://arxiv.org/abs/2402.01030). Has anyone else experimented with Gemini

The main example lies in our core architecture, which uses just three fundamental abstractions: a code executor, a model, and the agent itself (see https://github.com/gradion-ai/freeact/blob/main/freeact/exam...). All actions are expressed through executable Python code (that you can easily inspect) rather than through multiple abstraction layers. While other libraries or frameworks may offer convenience through extensive abstractions, this often creates black boxes that complicate debugging and customization when you go beyond standard use cases. For example, when you need to trace what prompts are being sent to the LLM or how tool outputs are being processed, multiple layers of abstraction can make this difficult to inspect. With our approach you can easily trace the flow from agent decision to code execution. HTH!

Hello! We just released freeact (https://github.com/gradion-ai/freeact), a lightweight agent library that empowers language models to act as autonomous agents through executable code actions. By enabling agents to express their actions directly in code rather than through constrained formats like JSON, freeact provides a flexible and powerful approach to solving complex, open-ended problems that require dynamic solution paths. * Supports dynamic installation and utilization of Python packages at runtime * Agents learn from feedback and store successful code actions as reusable skills in long-term memory * Skills can be interactively developed and refined in collaboration with freeact agents * Agents compose skills and any other Python modules to build increasingly sophisticated capabilities * Code actions are executed in ipybox (https://g