The problem

Good harness, or good workflow. Pick one.

Most tools force a choice. You wire raw CLI tools into scripts, or accept opinionated wrappers that bury settings and collapse planning into execution.

What you have today

One mode for everything

Settings buried in implementation

Locked to one provider

Prompts submitted as you typed them

What you actually need

Distinct modes for distinct intents

A first-class configuration surface

Harness choice per project, per session

Prompts refined to your intent level

See it work

Four modes. One intention.

You should not have to think, plan, execute, and validate in the same mental mode.

Kodo Code Alpha

Fix session reconnect my-app

Can we add real-time sync without breaking the mobile layout?

Yes — the existing ResponsiveGrid component already handles breakpoints at 768px and 1024px. Adding WebSocket sync would not affect layout rendering because state updates flow through the existing useDataStore hook.

The main risk is re-render frequency on mobile. Consider debouncing sync events to 200ms on viewports below 768px.

No side effects. No surprises. Just answers.

Break down the session reconnect bug into a safe fix plan.

Plan · 4 steps

Reproduce reconnect failure after provider restart
Isolate session state lost during websocket rebind
Patch the server resume path
Verify reconnect and partial-stream behavior

Plan is now an artifact. It carries forward into Code mode as structured context.

Implement the approved plan.

Carrying forward approved plan (4 steps)…

Msrc/server/session.ts— added rebind state check

Msrc/server/websocket.ts— preserve session on reconnect

Msrc/server/resume.ts— restore partial stream state

Asrc/server/__tests__/reconnect.test.ts

3 files modified, 1 file added · +142 −23

Check for regressions and edge cases.

Scanning 4 changed files against full test suite…

⚠session.ts:47 — Race condition if two reconnects arrive within 50ms. Add a guard.

✓All 23 existing tests pass

✓New reconnect test covers the happy path

✓No regressions in streaming behavior

One issue found. The fix is a 3-line guard. Recommend addressing before merge.

Ask anything, @tag files/folders, or use / …

gpt-5.4

Slash commands

Type / to take control.

Switch modes, swap models, and manage presets without leaving the composer.

Commands

/model Switch response model for this thread

/presets Switch model preset or create a new one

/ask Switch this thread into ask mode

/plan Switch this thread into plan mode

/code Switch this thread into code mode

/review Switch this thread into review mode

/usage Toggle active provider usage panel

/

GPT-5.4

Prompt enhancement

Refine, then send.

Sharpen your request before it reaches the harness. Three levels of preparation.

gpt-5.4

Features

What the orchestration layer gives you.

Dual Harness

Codex CLI and Claude Agent SDK, same workflow semantics. Switch per-project or per-session.

Auto Model Switching

Different models for thinking, doing, and reviewing. A higher-reasoning model for planning, a faster one for execution.

Plans as Artifacts

Plans persist with state. Created in Plan Mode, tracked through implementation, carried forward as structured context.

Git Worktrees

Each project runs in its own git worktree. Agents get a clean, isolated environment without interfering with your working tree.

Architecture

Harnesses execute. Kōdō orchestrates.

Kōdō handles model switching, workflow state, prompt enhancement, and settings. Then hands execution to Codex or Claude.

UI Layer

Kōdō Desktop
Ask · Plan · Code · Review

Core

Kōdō Orchestrator
Model switching · Enhance pipeline
Workflow state · Settings

Codex CLI

OpenAI

Claude Agent SDK

Anthropic

Your Environment

Repo · Terminal · Files

"Harnesses execute. Kōdō orchestrates."

How it compares

Kōdō sits between the harnesses and you.

Not competing with any of them directly.

	Kōdō Code	Cline	Roo Code	Claude Code	Codex CLI
Harness	Codex + Claude	API-based	API-based	Native Claude	Native OpenAI
Workflow Modes	✓ 4 modes	✗	✓	✗	✗
Prompt Enhancer	✓ 3 levels	✓	✓	✗	✗
Auto Model Switch	✓ per mode	✗	✓ per-task	✗	✗
Dual Harness	✓	✗	✗	—	—
Desktop UI	✓	VS Code	VS Code	✗	✗
Settings Surface	✓ first-class	✓	✓	Flags	Flags
Philosophy	Orchestrator	Editor-first	Feature-rich	Raw harness	Raw harness

Who it's for

You'll like Kōdō Code if you…

This is for you.

Love tools like Cline but want a stronger harness underneath
Want different models for thinking vs. doing vs. reviewing
Care about token efficiency and cost-per-task
Believe prompts deserve to be refined, not just submitted as-is
Want a desktop experience today, editor integration soon, and self-hosting on the horizon

What it isn't.

Not a new harness — Codex and Claude already do that well
Not a reimplementation of runtime behavior
Not a generic chat app with coding bolted on
Not a terminal emulator pretending to be an IDE

AI coding workflow desktop app for Codex CLI and Claude Agent SDK