Skip to content
English
Русский

Tasks to agents.
Control to you.

GrigDesk brings together AI agents, tasks, and development tools. You set the direction and approve the result.

Open source. Runs on Linux with your AI subscriptions and keys.

See a task move through the pipeline.

Agents move through the stages. The pipeline stops when it needs your decision. Try the example below.

PipelineLearning example
FEATURE

Add a product availability filter

Stage by stage

Show products in stock at the selected warehouse.

0 / 6Explore the project

AI agent runs this step

Select a node to explore its stageFrom task to acceptance
  1. Return for rework: Write a plan
  2. Return for rework: Implement
Revise the plan
Revise the result
1 / 6 · CURRENT EXAMPLE STAGE

Explore the project

The agent locates the code and examines how the change relates to the rest of the project.

AI agent runs this step
InputTask and project repositories
Change map
  • The catalog already receives stock by warehouse
  • Add the filter next to sorting
  • Use the selected warehouse from current context
A prepared example without running AI or commands. Stages depend on the project. Agent Human Command
Stage 1: Explore the project. The agent locates the code and examines how the change relates to the rest of the project.
You accept the work or send it back for changes.Try the loop

Runners for your kind of work.

Choose a runner and model for each stage. Discuss tasks in chats and build a workflow for your project.

Coding

  • Claude Code
  • OpenAI Codex
  • Kimi Code
  • DeepSeek
  • GLM · Z.ai

3D model generation

  • Tripo

Planned integrations

  • Grok
  • GigaCode
  • Gemini
  • Qwen

One project. Different roles.

PlanningImplementationReview
An example lineup. You choose your team.

Work from anywhere.

At your desk, on the move, or on the sofa. Your server runs the tasks. You connect from the device that suits you.

  • ComputerYour full workspace
  • PhoneTasks at your fingertips
  • TabletMore room for context
The real interface with sample data. Select a screen to explore it. Phone and tablet use Android.

Tasks run on your Linux machine. Set up a direct connection or relay for remote access.

Connect your devices

Two ways to run a task.

Define the sequence in advance or let a lead agent manage iterations. Final acceptance is always yours.

Pipeline

When the path is clear. You define the stages, runners, and approval points.

PlanCodeReviewYou
  • A route that follows your project rules
  • Checks and returns to the right stage
Set up a pipeline

Loop

When the result needs iteration. A lead agent assigns tasks, reviews the work, and guides the runner.

LeadWorkerYou
  • Revisions guided by the lead agent
  • Round limits and human acceptance
Set up a loop
Try the loop in actionWalk through a handoff, review, and acceptance in a learning example.
LEAD ↔ WORKER

Product availability filter

The lead assigns work and reviews every worker result.

  • LeadAssignments, feedback, stages
    Now
  • WorkerChanges and reports
  • YouAcceptance and decisions at stops

Lead ⇄ Worker
Lead → Human acceptance

Handoffs to worker 0 / 4

Task stage 1 / 2

LEAD

The lead prepares the first assignment

The task is a product availability filter for the selected warehouse. The lead splits work into implementation and verification.

  • Stage 1: filter using the selected warehouse.
  • Stage 2: checks and a result for acceptance.
2
4
6

Choose 2 to see a stop. The GrigDesk profile default is 20.

Example history · 1
  1. The lead received the task. Handoffs to worker: 0.

A prepared example without running AI or commands. Buttons advance the story; in GrigDesk the pair works automatically until acceptance or a stop.

The lead prepares the first assignment. Handoffs to worker: 0 of 4.

Check the result in a preview environment. The project profile defines the working copy, containers, and data.

About previews

Your tools, alongside your task.

From reviewing changes to working with servers. Your tools stay alongside the task and your conversation with an agent.

GrigDesk
Demo tasks and chats in the Russian application interface. Grok, GigaCode, Gemini and Qwen are planned integrations.

GrigDesk pipelines

Select the route, acceptance, preview or runners to explore details.

The real GrigDesk interface with demo data. Open the screenshot to explore the details.

Additional features depend on enabled plugins and configured connections.

Start with your task.

Choose a pipeline

Before you begin.

Find more details
in the documentation.

Who is GrigDesk for?

Developers running their projects on Linux with AI agents. GrigDesk brings tasks, pipelines, chats and development tools into one application window.

What is loop mode?

In GrigDesk documentation this is called a loop. A lead agent assigns work to a worker, reviews results and manages stages until human acceptance. Handoffs are limited; lack of progress, unavailable environments and decisions outside agent authority require human intervention.

What is free?

GrigDesk is open source under MIT and runs on your machine. Cloud AI access follows your providers’ terms, using your own keys or subscriptions.

Can different stages use different agents?

Yes. Claude Code, OpenAI Codex, Kimi Code, DeepSeek and GLM (Z.ai) are available for code. Runners and models are configured per project stage and require installation and authentication. Tripo is used separately for 3D generation. Grok, GigaCode, Gemini and Qwen integrations are planned.

Does everything run locally?

GrigDesk and project environments run on your machine. Cloud runners process the context you send at their provider. Chats can use local Ollama models.

How much setup does the first task need?

Install GrigDesk and a runner, connect repositories, define check commands and prepare a preview environment. Then create a route in the editor. The guides take you through these steps in order.

Build your own workspace.

Connect a project, choose a runner, and go from idea to acceptance.

Version 0.1. Project setup requires technical preparation. What to know →