Human-in-the-loop coding agents (e.g., Claude Code, OpenCode, Aider) are currently mutually exclusive with programmatic agent frameworks (e.g., smolagents, PydanticAI).
A key goal of this project is to marry both: use the same AI agent interactively in your terminal (`patchpal`) or in Python scripts (`agent.run("task")`), plus an autopilot mode for fully autonomous runs.
While it's true that the agentic coding assistant space is crowded (Claude Code, Aider, Opencode, Codex, etc.), we needed something supporting both local and cloud models that we could easily modify/extend for custom workflows.
Being able to access the agent through both a terminal and a Python API/REPL has come in handy.
Recent releases includes custom tools, agent skills, and built-in support for Ralph Wiggum loops.
A key goal of this project is to marry both: use the same AI agent interactively in your terminal (`patchpal`) or in Python scripts (`agent.run("task")`), plus an autopilot mode for fully autonomous runs.
Supports both local and cloud tool-calling LLMs.
Code: https://github.com/wiseprobe/patchpal
Documentation: https://ai.wiseprobe.io/patchpal/