AI coding agent · transfers between models

One model conducts.
Every model does the work.

Noritsugi lets any model take the lead. The lead model splits the job into subtasks and sends each one to the model that fits it best. A local Qwen leads, Claude designs, DeepSeek writes the tests, ChatGPT reviews. You see where every subtask went and what it cost.

Auto or manual routingLive cost per subtaskOne click to stop
Lead
Routing
Total for this task $0.00 Subtasks 0 / 0 Demo data. Prices are illustrative
Model routing

Every task transfers to the model that fits

Every model has its strengths and its price. Noritsugi puts routing at the core, and every other feature exists to make routing fast and safe to use. No more betting everything on a single model.

1 · Lead

Any model can lead

A small local model, a low-cost model or a frontier flagship. The lead understands the request, splits it into subtasks and assembles the result.

2 · Route

The right model, picked for you

Chosen by task type, context length, unit price and your budget. Or pick by hand in the panel and reassign at any time.

3 · Run

Subtasks run in parallel

See what is running, for how long and how many tokens it used. If one drifts off course, stop just that subtask.

4 · Merge

The lead merges and checks

The lead merges the changes, asks another model for a cross-review when needed, and you approve the result.

Transfer plans: set once, switch per project

Plan · Save

Local first

Anything that can run locally stays local. The cloud only gets the hardest step.

LeadQwen local
Code / testsDeepSeek
Hard design callsClaude
Typical feature ≈ $0.06
Plan · Balanced

Low-cost workhorses

Flat-rate Coding Plan quota does most of the work. Frontier models only review.

LeadDeepSeek
Reading the repoKimi
Cross-reviewChatGPT
Typical feature ≈ $0.17
Plan · Best

Quality first

A flagship leads and key changes get two reviewers. For refactors before a release.

LeadClaude
ImplementationClaude / GLM
Cross-reviewChatGPT
Typical feature ≈ $0.92
Cost

Where every cent went,
down to the subtask

Live pricing

Cost ticks up while each subtask runs, priced at each provider's published rates. Calls covered by a flat-rate plan show as $0.

Budget caps

Cap a task or a whole project. Near the limit, it switches to a cheaper model or stops and asks you.

Compared with “all flagship”

After each task, see what the same work would have cost on one flagship model, and how much routing saved.

Task #A31 · Add refunds to the order service11 min 42 s
SubtaskModelTokensCost
Split the task and planQwen local18.4k$0.00
Read the repo, find the modulesKimi182k$0.05
Design the refund state machineClaude24.1k$0.08
Build the API and migrationDeepSeek61.0k$0.01
Write 34 unit testsGLM47.3k$0.00 in plan
Cross-review the concurrency logicChatGPT15.8k$0.02
Total348.6k$0.16
All on one flagship: about $1.18Saved 86%Sample data
Models

Low-cost models and flagships, connected in a minute

Paste an API key or sign in to a Coding Plan. Local models connect through Ollama or LM Studio and never leave your network.

Qwen local / cloud DeepSeek Kimi GLM MiniMax Claude ChatGPT Gemini Any OpenAI-compatible API
Low-cost models, usable on flat-rate Coding PlansPay-per-use models, at published rates
Platforms

Native on Windows. In the terminal, on the desktop, on your phone

No WSL, no setup. The desktop app and your phone share the same task, so you can check routing, approve changes and stop a drifting subtask from anywhere.

Windows desktop

Routing, cost and diff review in one window

A native installer with PowerShell and Git Bash support. Non-ASCII paths and legacy encodings open as they are.

PS C:\work\shop> noritsugi route --preset local-first ● lead Qwen 3.8 27B (local) ● route design → Claude · code → DeepSeek · review → ChatGPT ● budget $0.40 per task, ask before going over
Mobile · iOS / Android

The routing panel in your pocket

Push alerts, progress and cost per subtask, approve or reject changes, stop with one tap.

Windows CLI

Everything, in the terminal

The same engine as the desktop app. Works in scripts and CI.

Linux CLI

Servers, containers, remote dev boxes

One-line install. Runs headless and over SSH.

Sync across devices

Pick up the same task anywhere

Start on your work PC, check on it from your phone on the way home, continue on a server the next day.

Pricing

Free for individuals. Upgrade when your team needs it

You pay model providers directly. Noritsugi adds no markup.

Personal

$0

Individual developers, students, open source

  • Linux / Windows CLI
  • Auto and manual routing
  • Live cost panel
  • Bring your own keys for every model
Download free

Pro

$ — Pricing soon

For people who code with AI every day

  • Everything in Personal
  • Windows desktop
  • Control and approve from your phone
  • Unlimited transfer plans and budget rules
  • One-click Coding Plan sign-in
Join the waitlist

Team

Contact us

For teams that need central control and compliance

  • Everything in Pro
  • Self-hosted. Your code stays inside
  • Member roles and model allowlists
  • Full audit log
  • One bill, budgets per team
Book a demo
Teams

Runs on your servers. Every routing decision logged

Admins decide who can use which models, which repositories may reach external APIs, and each member's monthly budget. Every routing decision, every charge and every blocked attempt lands in the audit log, ready for security and finance.

2026-11-17 14:02:11 alex.k route tests → DeepSeek allow $0.01 2026-11-17 14:02:40 alex.k route read repo → Kimi allow $0.05 2026-11-17 14:05:03 sam.r route payments → ChatGPT deny repo may not leave network 2026-11-17 14:05:04 sam.r route payments → Qwen local allow $0.00 2026-11-17 14:09:52 admin policy monthly budget backend $110 updated