LOCAL TASK CONTROL FOR AI CODING

AI coding,under control.

A clear scope. Purposeful checks. Progress that stays.TaskBelay keeps long-running work connected, even across sessions.

Open source. On your machine. In your coding tool.

TaskBelaySample task

Codex · task-auth-example

Fix the login retry bug.

Verifying
Agreed file scope
  • src/auth/retry.ts
  • tests/auth/retry.test.ts
Verification budget2 / 4

2 commands remaining

A scope decision

The agent wants to edit a file outside the plan http/client.ts

Needs attention
Your plan. Every check. The next step. All in one task.
Made for the tools you already use
  • Codex
  • Claude Code
  • DeepSeek
  • ZCode

HOW IT WORKS

Long tasks.
A little more clarity.

Keep the important decisions with the work, from start to finish.

A clear scope

Keep changes on purpose.

Agree on the files first. When the work needs to go further, record the scope decision.

Purposeful checks

Check what matters.

Set a verification budget. Additional checks come with a specific reason and a recorded increase.

Lasting progress

New session. Same task.

Your plan, stage and progress stay on your machine. Reopen the workspace and pick up where you left off.

The whole task, in view.

Explore the scope, checks and recovery state of one example task.

Control Center guide
TaskBelay Control CenterInteractive example

Codex · LOCAL TASK

Fix the login retry bug.

task-auth-example
Current stage

TestTEST

Implementation is complete. Run the planned checks against the current repository content.

Acceptance
Lock after 5 failed login attempts.

Retained timeline

  1. 01Plan approved
  2. 02Implementation completed
  3. 03Targeted verification

The local Control Center lets you handle pending scope, verification and recovery decisions for the selected task.

These views illustrate product capabilities; they do not connect to your local task data.

The workflow you already know

  1. You approve the plan
  2. Your agent edits and checks
  3. Resume from saved progress

Return to decisions when the plan changes or work needs your attention.

From real development

From an idea to a native app.

The Notepad-- macOS migration. A real, long-running task.

MIGRATION RESULTS

Notepad--

Qt macOS

Native macOS project
AppKit + SwiftUI
Universal installer
Universal DMG

From the author’s recorded task

Author’s own project

A native macOS migration with TaskBelay

From an approved plan through testing and rework to a native installer, TaskBelay connected each decision with the progress that followed.

THE RESULT

Native project and universal installer produced

Full feature migration and compatibility acceptance are still in progress.

Read the full story

One task, through every turn

  1. Plan
  2. Build & check
  3. Repair & retest
  4. Repackage

Install

Your next task
starts here.

Install locally. Choose your coding tool. Get back to work.

View on GitHub

Apache-2.0 · Local-first · No telemetry

  1. 1

    Install

    Requires Node.js 24 or later and Git.

    npm install -g taskbelay@latest
  2. 2

    Run

    taskbelay
  3. 3

    Choose your coding agent

    Run taskbelay and choose your coding tool’s install option in the terminal menu. Follow its guide to enable the integration and complete trust setup.

After installation and setup, tell your coding tool:

Start this task with TaskBelay. Prepare the plan first and wait for my approval before implementing it.

A FEW MORE THINGS

Before you get started.

What kind of work is TaskBelay for?

Long-running repository work, tasks that span sessions or days, and changes that may need rework. One-off questions, explanations, status checks and small edits that need no saved progress usually do not need it.

Will I have to keep approving things?

After plan approval, the agent works within the current task. Plan changes, extra files and blockers may need decisions. More verification needs a recorded reason and specific checks, not your approval for every test command.

Does it block every change outside the scope?

Supported integrations can check planned structured writes against the saved task before execution and require a scope decision. Manual, external and unsupported changes are observed and reconciled with the task, not blocked at the OS level.

Where are the plan and progress saved?

Task state stays on your machine: the plan, file scope, current stage, progress and verification results. It survives the chat. TaskBelay collects no telemetry.

Do I need to switch coding tools?

No. TaskBelay works with Codex, Claude Code, DeepSeek and ZCode. Set up the integration, then keep using your usual coding tool and repository.