◎

hydracoder

A local-AI development orchestrator. Enter a project goal, watch local models plan it, build it, and review it. No corporate AI in the loop.

What it is

hydracoder is the layer that turns a goal into shipped code using only the models on your own machine. It plans the work into a task graph, routes each task to a right-sized local model, reviews every result, and keeps an append-only journal so a crash or a full context window never loses progress. It is built to keep development moving on local hardware when hosted AI becomes too expensive to use.

Plan

A planner model decomposes the goal into tasks with explicit interfaces and acceptance conditions.

Build

Each task runs through a local model in an isolated context, streaming its work to a live terminal.

Recover

An append-only journal makes every run resumable. Interrupt it, restart it, it continues.

Built on

hydra-llm runs and manages the local models in Docker.
lillycoder is the per-task agent loop that reads, writes, and edits files.

Install

curl -fsSL https://ra-yavuz.github.io/apt/KEY.gpg | sudo gpg --dearmor -o /usr/share/keyrings/ra-yavuz.gpg
echo "deb [signed-by=/usr/share/keyrings/ra-yavuz.gpg] https://ra-yavuz.github.io/apt stable main" | sudo tee /etc/apt/sources.list.d/ra-yavuz.list
sudo apt update && sudo apt install hydracoder

hydracoder serve   # then open http://127.0.0.1:8765/

Links

source releases install

hydracoder is provided AS IS, WITHOUT WARRANTY OF ANY KIND. It runs local models that read, write, and delete files in a workspace and run shell commands on your machine. You alone are responsible for any damage to your data, hardware, or system. By running it you accept all risk.