Code Scout – Free Local AI Coding Agent

Code Scout provides a native workbench that lets developers run an AI coding assistant entirely on their own machine. It supports locally hosted language models through Ollama, LM Studio, or llama.cpp, and can also be configured to call hosted endpoints when desired. The tool keeps the repository context in plain files (.codescout) and gives the agent direct access to the shell, git, and the filesystem, allowing it to execute commands and apply changes rather than merely suggest code.

The architecture separates planning and execution into two roles: an orchestrator that selects tools and outlines steps, and a coder that runs those steps against the codebase. Each iteration produces concrete output such as logs, diffs, or test results, which feed back into the next cycle, maintaining a small, focused context that works well with modest‑size models. A visible heartbeat indicates when the agent is thinking, running a tool, or waiting for a model response.

Users can benchmark different local and cloud models side‑by‑side within the same interface, comparing latency, output quality, and cost. The application runs on macOS (Apple Silicon and Intel), Windows, and Linux, requires no sign‑up, and keeps all source code on the user’s hardware.