local-coder
Builds a local coding-agent team on top of Ollama and OpenCode by detecting the machine's hardware, recommending models for seven specialist roles, and testing whether each assigned model can actually use tools, edit a repository and pass verification.
local-coder builds a local coding-agent setup on top of Ollama and OpenCode, detecting what a machine's memory, GPU, disk and context budget can support, recommending models that fit, and assigning them to seven specialist roles such as exploring, planning, implementing and reviewing. It generates a permission-restricted workflow for the assigned models and then tests whether the result can actually use tools, delegate work, edit a repository, and verify that the edit succeeded, rather than assuming a downloaded model will behave correctly.
Its stated completion rule is that an agent's own claim of success is not evidence; the generated workflow includes a separate verifier that inspects the repository and validation output before accepting a result. A single model can fill more than one role when the hardware and available model catalogue support it, downloads are deduplicated by tag, and a dry-run mode shows the planned recommendation and downloads without writing or downloading anything.