Skip to content

MABOS-Code / From intent to implementation

Your intent.
An entire
coding system.

Set the direction. Choose your models. Bring agents, tools, and lasting project memory into the work. Code, Plan, and Orchestrate carry your intent from the repository to a running application.

Sign in with your authorized KI account. Get desktop access

01 / THE WORKSPACEWEB + DESKTOPONE SHARED RUNTIME
01 / The working environment

The work.
In full view.

Move from the conversation into files, changes, commands, and a live preview. The repository stays in reach. The task keeps its history. You keep the thread.

MABOS-Code / Workspace
Current MABOS-Code workspace with the Orchestrate composer, team selection, voice input, New Atlas, and an unsent example prompt
Current MABOS-Code workspace, with an unsent example prompt.

02 / Inside the harness

Follow the intent.
See the work.

Take one request through the system. Open the plan, inspect an agent’s assignment, then use the application preview. Every stage belongs to the same project.

▱   example / field-notesInteractive product tour
Choose a repository

Start where the work lives.

Bring the repository, branch, and project instructions. Give Code a goal and the conditions that matter.

Repository contextFiles & instructionsLocal or cloud execution

A working example you can explore. Tour controls do not change your account or run a coding task.

Add a searchable notes view. Keep the existing architecture.

Auto · Balanced

Repository instructions, files, and project context travel with the task. Model choice and the machine that runs the work are separate controls.

03 / Your models. Your decisions.

Bring your models.
Direct the work.

Local when it fits. Cloud when it helps. Your choice when you know what you want. Auto weighs eligible models, tool support, and benchmark evidence—and shows you why.

Explore model routingTour controls
Local first · cloud when eligibleAuto routing

Balanced prefers eligible local models, then eligible cloud models. The task, tool support, context, account access, and your settings shape the route.

Frontier models in Auto

Higher-cost frontier models join Auto only when you opt in.

Off by default. This includes automatic fallback and agent selection. Manual selection does not enable Auto opt-in.

The model is a choice.
The workspace stays yours.

Work with supported coding models from OpenAI, Anthropic / Claude, Google / Gemini, xAI / Grok, Together, and your local machine. Your connected accounts and each model’s supported tools determine the available catalog.

  • Bring your keysConnect provider accounts and use the KI / Array credential wallet.
  • Separate model from machineChoose inference and local or cloud execution independently.
  • Inspect the decisionSee the selected model, routing reason, context capacity, and connection health.
  • Keep costs visibleActual API/token usage and estimated charges have their own accounting view.

Model and provider availability follow your connections. Product access and customer-funded inference are separate.

04 / Continuity without starting over

The conversation moves.
The knowledge stays.

A project learns as you build it. The Living Record keeps the task’s experience. The Meaning Lattice connects decisions, discoveries, and evidence so the next step starts with understanding.

01

Living Record

Remember what happened.

Retain the request, conditions, actions, and evidence. Inspect the task’s context and compact it without losing what matters.

02

Meaning Lattice

Connect what it means.

Link summaries, code changes, and discoveries. Continue a task or begin the next with connected project knowledge.

Primary model contextIllustrated example
38%

of the selected primary
model’s context window

Instructions & toolsConversation & resultsAvailable context

The persistent percentage follows the primary conversation and the selected model’s capacity. Child agents have their own context windows.

Inspect the breakdown. Keep the task’s intent and conditions through compaction.

✓ Intent✓ Conditions✓ Useful evidence
Separate usage accounting

Primary calls · sub-agents · retries · tools · Cold Lane
Token/API consumption and monetary cost are measured separately.

05 / A second perspective on the task

Cold Lane.
Keep the goal in view.

A small-budget supervisor checks in when the work needs it. Cold Lane helps route models, coordinate workers, identify drift, and check the requested outcome.

It resolves suitable routine reviews within your configured authority. Your instructions, conditions, and Stop control stay in force.

Cold Lane

Supervision is on.

Read the goalCheck the workGuide the next step

Default-on, and yours to disable. Invoked as needed with a deliberately small budget; its usage appears in accounting.

Tour example · No workspace settings change

06 / Connected by design

Your workspace.
A wider system behind it.

A KI account connects the experience. Add the services and knowledge your project needs, with clear scope and visible connection status.

Connect the services you build with.

GitHub for repositories. Supabase for data. Vercel for delivery. Local Environment detects dependencies and guides setup, including Docker where needed.

Add search on your terms.

Tavily is an optional, additional connection using your own key. Public KI Search is included without another search-provider account.

Explore web search

Web search is on in this example.

Use public web results to find references and verify current information. This tour does not send a search or change your workspace.

Tour controls · Public information only

07 / One project, wherever you work

Open in the browser.
Make it at home.

The web workspace and desktop application share the same Rust runtime. Get access in the KI Store, then install and update Code through Array. Desktop keeps your sign-in through updates and prepares the local environment for your work.

  • KI sign-in, carried throughUse your KI account or sign in with Array. Your project and account stay distinct.
  • Independent conversations, shared workSeparate chats collaborate on one project. Supervised agents coordinate their work, with concurrency governed by available machine capacity.
  • Atlas and a soundtrackPurposeful recorded cues, captions, and your music player make the workspace feel alive.
The Code soundtrackOwner recordings

Set the rhythm.

Six original tracks, on a continuous loop. Press Play and explore. Music softens while Atlas speaks.

Recursive Presence

Ready to play
0:000:00
65%

Music starts when you press Play.

MABOS-Code / Kuykendall Industries

Build what
you have in mind.

Your models. Your project. A workspace that carries the work forward.

Authorized accounts can enter the workspace. Selected KI Fieldwork testing partners receive the product rights listed in their invitation. Inference and execution use your configured resources.

Explore Code’s ecosystem connections ↗

Inside the MABOS-Code workspace

Full current MABOS-Code workspace with an unsent example prompt

Current MABOS-Code workspace, with an unsent example prompt.