GEKTORCODING AGENT

OPEN SOURCE / MIT LICENSE

Code and workflows.

Use an agent to read and edit files, or build workflows you can run again.

Download GEKTOR Documentation

WINDOWS / LINUX · WRITTEN IN GO

ABOUT GEKTOR
01 / OVERVIEWAGENT + WORKFLOWS

Two ways to work.

Agent and Workflows share a workspace, but run differently.

In Agent, the model chooses tools as the task unfolds. In Workflows, you define the steps and their inputs before running them.

01 / AGENT

Work through a task.

Ask the agent to inspect files in Plan mode. Switch to Act for edits and commands. Reopen a saved session to continue the conversation later.

PLAN → ACT → RESUME
↗
02 / WORKFLOWS

Run a saved recipe.

Edit a recipe, run it with new inputs, and inspect each step’s output. Model-generation steps need a model connection; the other built-in steps do not.

DEFINE → RUN → INSPECT
↗

02 / SETUP

Connect a model.
Choose a workspace.

Use llama.cpp, KoboldCPP, an OpenAI-compatible API, or an eligible ChatGPT account. Local models need a separately running server and tool-call support.

For ChatGPT, sign in first, then select it under Settings → Model connection. Choose your project folder under GEKTOR → Change workspace.

Read the setup guide
START AGENT
  1. 01 Download and extract a release.
  2. 02 Open a terminal in the extracted folder.
  3. 03 Launch Agent with the command below.
.\start-gektor-agent.cmd

Change the connection in
Settings → Model connection.

DEFAULT / LLAMA.CPP127.0.0.1:9931

WINDOWS / LINUX

[ Download GEKTOR ]

Experimental software. Agent results depend on the model, connection settings, and task.