MAC & WINDOWS NATIVE APP arrow_forward

Same plan, twice the week.

Hand the busywork to workers that cost a fraction.

Frontier model on the hard parts, cheap workers on everything else. Nothing new to subscribe to.

Installs in 30 seconds · No card required · Top up from $1

Read the getting started guide
Unlimited
Projects in parallel
99%
Cheaper on the busywork
$0
To connect Claude or Codex
$1
Minimum top-up · no subscription

* Savings depend on which model the busywork is handed to. 99% is the ceiling, not the average.

Why are you always out of tokens?

Because you're handing everything to one agent.

Run a coding agent from a terminal and the same top-tier model does all of it — reading forty files to answer one question, sweeping the codebase, writing boilerplate. You pay the highest price for work a far cheaper model does just as well. Then the hard problem arrives and there's nothing left.

  • “Resets at 5PM”

    You hit the wall mid-task. Until the clock turns over, there is nothing you can do about it.

  • Switch models and the context is gone

    You'd move the cheap work somewhere cheaper — but you'd have to explain the whole project over again.

  • You can't see where it went

    You find out it's spent. You never find out what spent it.

Agentic development environment · ADE

So your expensive model only does the thinking. The rest routes down.

Connect the Claude or Codex you already pay for — your skills, hooks and CLAUDE.md carry over unchanged — and reading, searching and boilerplate go to workers that cost a fraction and hand back only the summary. Nothing new to learn, nothing new to subscribe to.

  • The busywork routes down

    Reading, searching and boilerplate go to workers that cost a fraction and hand back only the summary. Your top model stays on judgment and design, so the limit arrives much later.

  • Every pane runs its own model

    You don't switch — you run them side by side. Change the model in one pane and the work in the others keeps its context, so nothing has to be explained twice.

  • You pay for what you use

    No subscription to add. Usage draws down a prepaid balance you top up, the balance sits in the app, and anything you haven't spent is refundable at any time.

Orchestration engine

One model orchestrates — many run the subtasks in parallel.

The top model stays on judgment, design, and final synthesis. Bulk file reads, searches, and boilerplate are delegated to worker models — a local endpoint, flow, haiku, or sonnet — that run in parallel and hand back a summary, keeping the top model's context on the hard parts.

task_alt read 40 files
task_alt grep sweep
task_alt boilerplate
task_alt routine refactor
hub
Top model
Orchestrator — judgment · design · synthesis
bolt
delegate to local · flow
your own endpoint, or a fast lightweight worker
bolt
delegate to haiku
quick bulk reads, searches, boilerplate
bolt
delegate to sonnet
heavier subtasks that still aren't the core
Parallel workers
3x
Subtasks split across workers running at the same time
Cost saved
Up to 99%
When busywork runs on a cheaper worker instead of the top model
bolt

delegate to local · flow

your own endpoint, or a fast lightweight worker

manage_search

delegate to haiku

quick bulk reads, searches, boilerplate

layers

delegate to sonnet

heavier subtasks that still aren't the core

One workspace, split your way.

Start from a layout that matches how you work — two stacked, three columns, a 2×2 grid, or a 2-plus-1 split — and run a separate project in each pane. Assign a model per pane and let them build in parallel.

meshcode
2×2 grid
~/newapp Claude
~/shop Codex
~/saju meshcode
~/docs Claude
Model per pane meshcode Claude Codex

Big Coding Brain

It remembers how you solved it — so it doesn't dig again.

Every fix, gotcha, and convention from your work gets distilled and stored automatically. Next time a similar task comes up, meshcode recalls the exact know-how instead of re-searching from scratch — saving tokens and getting it done faster.

lightbulb Why that deploy kept failing
lightbulb The auth quirk in this API
lightbulb Convention this repo follows
lightbulb
Big Coding Brain
capture → distill → recall
autorenew
Without memory
Re-reads the code, re-searches, re-derives the fix
check
With Big Coding Brain
Recalls the known fix, applies it immediately

Every task adds to the memory — the next one runs faster and costs fewer tokens.

Mobile remote

From your phone, put your home machine's Claude & Codex to work.

Just open the mobile app, wherever you are. mesh's flash model takes your request as the orchestrator and drives the local CLIs — Claude, Codex, GLM — installed on your home desktop or laptop, then sends only the results back to your phone. The computer just needs to be on.

smartphone mesh · mobile
Fix the Coupang parser bug
bolt flash · orchestrator
iroh relay
computer home · laptop local CLIs on your machine
claude patch the root cause
codex write & run tests
grok review the diff
glm commit & push

* An iroh relay links your phone directly to your own machine. Your code stays on your computer; the heavy execution runs on the local CLIs.

Focus music, built in

Music with taste, built in.

500+ original focus tracks, streaming right inside meshcode. Free with your account — no new tab, no ads.

Play
Midnight Compiler
Lo-fi · Deep Focus
In the queue
1 Neon Terminal · Synthwave
2 Rainy Standup · Jazz
3 Ship It · Ambient

* Free with your account — part of the native workspace, not a separate subscription.

MESHLING

A little companion that grows while you build.

Every meshcode workspace comes with a MESHLING — a small pet that levels up as you ship. It keeps its own ledger and can be traded in the in-app auction house. Purely for fun; it doesn't affect your code or your bill.

  • trending_up Levels up as you build
  • receipt_long Own MST ledger
  • gavel Tradeable in the auction house

* Open with ⌘5 in the desktop app.

MESHLING
Lv. 12 · MST 1,240

Pay as you go

Top up, then spend what you use.

No subscription, no seats, nothing that renews. Credit sits on your account until you spend it, so a month you don't build costs nothing.

  • bolt
    Top up from $1
    Add credit any time, in any amount. Unused balance stays on the account and doesn't expire while the account is active.
  • bolt
    Usage billed at cost
    Tokens are billed at provider cost. No markup added on each request.
lightbulb

Questions, answered

Will my existing Claude Code or Codex setup work? add

Yes. meshcode drives the CLI you already have installed, so your skills, hooks, MCP servers and CLAUDE.md carry over untouched. There's no second config to maintain and nothing to port. If it works in your terminal today, it works in a pane.

Doesn't quality drop if cheaper models do the work? add

It depends entirely on what you hand over, which is why you decide. The work worth delegating is the work you can check — reading files, sweeping searches, boilerplate, mechanical edits a build or a test verifies. Judgment, design and anything subtle stays on your top model. You set the routing rules in settings and change them whenever you want.

Does this remove my Claude usage limits? add

No, and no tool can — those limits belong to Anthropic and your plan. What changes is how fast you reach them. When file reading and repetitive work run on cheaper workers instead of your subscription, the same day's work draws far less from it, so the ceiling arrives much later.

Do I have to give up my terminal? add

No. meshcode runs the same CLI agents you already use and puts a window around them — several at once, each on its own project, with the output laid out instead of scrolling past. Keep using the terminal whenever you prefer it.

Where does my code live? add

On your machine. meshcode is a native desktop app rather than a browser tool: it opens your local project folders and the files stay exactly where they are.

What models can I use? add

Pick per pane. Connect the Claude, Codex, GLM or Grok you already pay for, or use meshcode's own metered tiers — Flow, Logic and Architect — for the delegated work. Different panes can run different models at the same time.

How does pricing work? add

Pay as you go, with no subscription and no seats. Your own Claude or Codex usage is billed by them and never passes through us. meshcode's own models draw down a prepaid balance you top up from $1 — nothing renews, nothing resets monthly, and unused balance is refundable at any time.

What does it run on? add

A native app for Mac and Windows. It opens in under a second, and it's free to download and start.

Set up your first workspace.

Download the native app for Mac or Windows, open a project in each pane, and give each pane its own model.

Download arrow_forward

help@meshcode.ai

New here? Read the guide first