MaxLabs for developers

An agent that can
do the work.

Open a project, describe the result you want, and let MaxLabs inspect the code, make changes, run commands, and show you exactly what changed.

01 · WORK IN YOUR CODEBASE

It does more than answer questions.

MaxLabs can read, search, create, and edit files. It can run terminal commands, fetch web pages, and delegate a focused task to a subagent. It keeps working through multi-step jobs instead of stopping after one response.

  • Read, write, edit, glob, grep, shell, and web tools
  • Up to 100 tool-calling steps in one turn
  • Focused subagents for research and isolated work
maxlabs · storefrontACT
❯ Fix the checkout race and run the tests.
Searched for checkout locks
Edited CheckoutService.php
Ran the focused test suite
step 7/100 · cache active · complete

02 · STAY IN CONTROL

See the plan. Approve risky actions. Review every edit.

Switch to plan mode before the agent changes anything. Read-only work can continue without interruption, while file writes and other risky actions wait for your approval. Every edit appears as a clear diff.

  • Plan and act modes with a visible keyboard switch
  • Allow once, allow for the session, or deny
  • Exact file diffs with added and removed lines
approval requiredWAITING

MaxLabs wants to edit:

app/Services/CheckoutService.php

− return $order->save();
+ return DB::transaction(fn () => $order->save());
Allow onceAlways allowDeny

03 · BUILT FOR TOKEN EFFICIENCY

Reach 99.9% cache hits in long sessions.

We built MaxLabs Desktop and MaxLabs CLI around every token they send. Each model request comes from an append-only session log with byte-identical stable instructions. As a session grows, the model can reuse almost all of the repeated context instead of charging it as fresh input.

  • Deterministic requests from an append-only history
  • Stable tool definitions and session instructions
  • Lower cost and quota usage as sessions grow
long session99.9% CACHE HIT
S0Toolsfixed order
S1Agent identitysession stable
S2Instructionssession stable
S3–5Project + conversationappend only
S6Current runtimechanges at the tail

04 · CLI, DESKTOP, AND API

Use MaxLabs where the work happens.

Work from the terminal or use the desktop app for project history, persistent conversations, branches, usage, model choice, and reasoning effort. Turn off the five-hour limit when you need one long run. Building your own product? Use the same models through a separate pay-as-you-go API.

  • Desktop projects, saved conversations, and branches
  • OAuth sign-in for people; API keys for applications
  • OpenAI-compatible chat completions and streaming
POST /v1/chat/completions200
// Keep your existing SDK.
client.chat.completions.create({
  model: "pro",
  messages,
  tools,
  stream: true
});
SHORT WINDOW5 hours
LONG WINDOW7 days

TWO JOBS. TWO MODELS.

Choose by the work, not by a version number.

Both models can understand a codebase, plan, and use tools. Pro is better at deep investigation, difficult debugging, and complex planning. Worker stays efficient for everyday implementation. Both keep stable names while we improve them.

PRO

Find the right answer.

Use Pro to understand unfamiliar code, find bugs, compare approaches, and create a careful plan.

WORKER

Plan and execute efficiently.

Worker can plan, solve problems, and carry out implementation work. Choose Pro when the task needs deeper investigation or more careful reasoning.

Give MaxLabs
a real codebase.

Start with MaxLabs Desktop or MaxLabs CLI, or connect your application through the API.

Read the quickstart Compare access options