Execute Agent

An interactive execute agent with a simple planning loop, tool registry, built-in tools (shell, filesystem, http, python eval), and pluggable model providers (echo, OpenAI, Ollama). Supports streaming and an optional web UI over Server-Sent Events (SSE).

Install

pip install -e .

For OpenAI provider:

pip install -e .[openai]
export OPENAI_API_KEY=... 

Usage

• One-shot (streaming):

execute-agent --provider echo --stream "Use math.calc to compute 2*(3+4)/5"

• REPL:

execute-agent --provider echo --stream

OpenAI:

export OPENAI_API_KEY=...
execute-agent --provider openai --model gpt-4o-mini --stream "Fetch https://example.com via web.get"

Ollama:

execute-agent --provider ollama --model llama3.1 --stream "Hello"

Web (SSE)

Install extras and run:

pip install -e .[web,openai,ollama]
uvicorn agent.runtime.web:app --reload --host 0.0.0.0 --port 8000

Open http://localhost:8000

Type exit to quit.