Guide

Everything you need to run meshcode day to day.

A practical walkthrough: install the desktop app, add the Claude / Grok / Codex CLI, sign in with /login, then configure how work is delegated.

01 · Getting started

Install, sign in, open a workspace.

Download the app, open a workspace, then install a vendor CLI and sign in.

  1. 01

    Download the desktop app

    Grab the build for macOS or Windows and install it like any other app.

  2. 02

    Sign up and top up

    No subscription — create an account, top up from $1, and you only pay for what you use.

  3. 03

    Open a workspace

    Point meshcode at a project folder. Next: install Claude, Grok, or Codex and type /login — details below.

02 · Install a CLI

Claude, Grok, or Codex — then type /login.

meshcode talks to the real vendor CLIs. After the desktop app is installed, pick a model, get its CLI onto your machine, and sign in once.

In the app

Above the chat input, each provider chip shows (Install) if that CLI is missing. Click it — meshcode runs the official installer. You can do this later from the same chips.

Then type /login

Select the model you just installed (the chip above the input must be active). In the chat box, type /login and send. meshcode opens that vendor's login window — finish signing in there. Once per CLI, until you /logout.

Or in your own terminal

Same installers the companies publish. Paste into Terminal on macOS, or PowerShell on Windows:

Claude Code

Anthropic

macOS / Linux

curl -fsSL https://claude.ai/install.sh | bash

Windows (PowerShell)

irm https://claude.ai/install.ps1 | iex
Claude Code setup →

Grok Build

xAI

macOS / Linux

curl -fsSL https://x.ai/cli/install.sh | bash

Windows (PowerShell)

irm https://x.ai/cli/install.ps1 | iex
Grok Build docs →

Codex

OpenAI · ChatGPT

macOS / Linux

curl -fsSL https://chatgpt.com/codex/install.sh | sh

Windows (PowerShell)

irm https://chatgpt.com/codex/install.ps1 | iex
Codex CLI docs →

03 · Auto-delegation

One model leads, workers handle the busywork.

Inside a session, the top model you're chatting with stays focused on judgment, design, and synthesis. Bulk file reads, searches, and boilerplate get delegated automatically to worker models running in parallel — you can see and adjust which backend each kind of work goes to in session settings.

Auto-delegation (this section) picks a worker backend per subtask. The Flow / Logic / Architect tiers below are a separate setting — they control which real model answers when a subagent or slash command targets one of those three named tiers.

04 · Model tiers

Flow, Logic, Architect — and the real model behind each.

meshcode ships three named tiers so tools, subagents, and workflows can request 'a fast model' or 'a careful model' without hardcoding a vendor. By default each tier maps to a sensible built-in model — you can override any of them with a specific model from the catalog.

bolt

flow

Fast, lightweight tasks

Bulk reads, greps, boilerplate, quick edits — high volume, low latency.

psychology

logic

Multi-step reasoning

Debugging, multi-file changes, anything that needs a few steps of reasoning to get right.

architecture

architect

Complex design & refactors

Architecture calls, large refactors, and anything you'd want a second opinion on.

Where to set it

Go to Account → Routing. Each tier is its own card — pick a model from the catalog (or leave it on the default), then tune it:

  • check Model — search the catalog and pick the exact model that should answer for this tier.
  • check Temperature — how deterministic vs. creative responses are (0–2, defaults to a low value for coding).
  • check Thinking — when the picked model supports it, choose how much reasoning budget it gets.
  • check Exclude data-training providers — restrict the pool to providers that don't train on your requests.
Open Routing settings arrow_forward

05 · Account

Everything else lives under Account.

Balance, keys, usage, and the model catalog are all one click away.

Overview Balance at a glance and quick links to every settings page. arrow_forward Routing Configure the Flow / Logic / Architect tiers (see above). arrow_forward Models Browse the full model catalog with live pricing. arrow_forward API keys Create and revoke keys for programmatic access. arrow_forward Usage Daily spend and request counts, broken down by day. arrow_forward Billing Top up your balance — unused balance is always refundable. arrow_forward

Frequently asked

Do I have to install every CLI? add

No. Install the one you want to chat with. (Install) and /login only apply to that provider.

Is /login the same as signing into meshcode? add

No. A meshcode account is billing/routing/catalog. /login is the vendor CLI sign-in (Claude, Grok, or ChatGPT).

The in-app install button didn't work. add

Run the official command for your OS in your own terminal, then restart meshcode so it picks up PATH. Or use the vendor setup page linked above.

What happens if I don't touch the tier settings? add

Each tier uses meshcode's default model — you only need Routing settings if you want to swap one in for a specific reason (cost, latency, capability).

Is auto-delegation the same thing as picking a tier model? add

No — auto-delegation decides which subtasks get handed off to a worker at all. The tier settings decide which real model answers when something targets Flow, Logic, or Architect by name. They compose: a delegated subtask can itself be routed through a tier.

Can I change a tier's model later? add

Yes, anytime — changes apply to your next request, no restart needed.

Where do I see pricing for a specific model? add

The Models page (in-app or the public catalog) shows input/output/cached pricing per million tokens for every model.