ollama-workspace-agent

OwA (Ollama Workspace Agent) is a coding agent that runs in the terminal with Ollama. It reads the workspace, finds relevant code, makes reviewed changes, and checks the result with real tools. The project is described as local by design, using your Ollama model, your workspace, your approval, and verified output.

The workflow keeps work visible. The model gets focused context and tool results, while the user remains in control of actions that change files or run commands. It can read files directly, inspect Git, or build a local index for code search, with evidence trimmed to fit smaller model contexts. For edits, it asks for a focused patch, checks workspace paths, and asks before writing. For verification, the user approves a command, sees its output and exit code, and OwA reports whether the requested check actually passed.

OwA uses concrete tools to inspect, edit, test, and explain work in a repository, and optional integrations stay explicit. It is built around local models with native Ollama tool calls, bounded context, and explicit verification to help 7–9B models handle focused tasks. Control features include approval gates for writes, commands, and MCP tools; path checks before writes; and a constrained Docker command sandbox. MCP servers run with host privileges, and host command execution is an explicit opt-in.