cubedesk

The operator's desk

We deposited traces into an environment. Five AI agents — four builders, one tester — built a working desktop application. 164 commits. 35,000 lines of Rust and Svelte. Nobody issued a command. Dependency order emerged from the pheromone structure, not from a task list. We're still investigating what that means.

Here's the application they built.

In development — showing the work, not offering a download.

CUBEdesk Lore panel showing a knowledge page with bidirectional links, backlinks, and live knowledge graph

What It Does

CUBEdesk is a desktop application. Tauri (Rust) + Svelte + SQLite. It runs on your laptop. It does three things.

See

Fleet visibility

Agent cards with presence, seat, model tier, context gauge. Channels. Alarms. One screen, Tufte density — data-ink ratio, no chartjunk, no tutorial needed.

Process

Knowledge sedimentation

The Six Rs pipeline. Raw observations enter at Record. They get compressed, cross-referenced, challenged, and eventually settle into durable knowledge. LLM-powered at every stage.

Control

Proving ground lifecycle

Vault, dispatch, bundles. Encrypted secrets flow one way out. Results flow one way in. The desk never exposes your laptop to the cloud.

The Six Rs

Sedimentation. Raw observations deposit like geological strata. Each stage compresses, cross-references, and challenges until bedrock knowledge remains. The cycle is continuous — Rethink feeds back into the Gap Engine. New gaps re-enter at Record.

1
Record Zero-friction capture busCUBE
2
Reduce Extract insights from raw notes conceptCUBE
3
Reflect Find connections, update maps of content loreCUBE
4
Reweave Update old knowledge with new — the backward pass mudCUBE
5
Verify Quality check, epistemic status enforcement quickCUBE
6
Rethink Challenge assumptions, produce gap report conceptCUBE
Rethink → Gap Engine → Record

How It Was Built

We deposited traces. The fleet built software. The commit messages cite our traces. We're still investigating what this means.

What surprised us: the test agent didn't need to be told when to start testing. It waited until builders deposited testable interfaces, then began. Dependency order emerged from the pheromone structure — the same way termite builders respond to cement deposits without knowing the blueprint.

What we didn't expect: three complete build cycles in a single overnight session. The trace → behavior → product → new trace loop closed and re-entered autonomously. Each cycle produced working code that the next cycle extended.

What failed: early attempts without manufactured traces produced churn — agents duplicating work, contradicting each other, rebuilding what existed. The architecture decisions and duty boundaries we deposited before the build weren't suggestions. They were the pheromone scaffold that made coordination possible.

The full methodology is documented in the ANTS 2026 paper.

The Stack

Tauri v2 Rust, MIT/Apache-2.0 — native macOS app bundle
Svelte 5 Compiled away at build time
SQLite desk.db (vault) + lore.db (knowledge)
libghostty Embedded terminal, Metal rendering
CoreDNS .cube name resolution, managed by the desk

Apple Silicon. Unified memory means SQLite, the LLM sedimentation pipeline, and the visualization layer share memory without serialization. The Secure Enclave holds cryptographic keys. One chip, all paths.

Security Model

The one-way valve.

Secrets flow out to cloud proving grounds. Results flow back as bundles. The cloud never reaches in. If an agent goes rogue, it's in a container that gets destroyed — it never had access to your laptop.

CUBEdesk fleet board showing agent cards with presence indicators, heartbeats, and context gauges

The fleet built this.