Asdes Studio - complex work, made simple

Four products. Each one handles something for you.

Four products. Each one takes something you'd normally have to manage and handles it.

A personal assistant that actually knows you

It remembers your preferences, context, and patterns - and gets more useful every day you use it.

Most assistants start from zero every time. This one carries your context forward - your preferences, your routines, how you like things handled. Over time it anticipates what you need before you ask. It does the work too: acts on your phone, moves things forward, follows tasks through to done. The more you use it, the less you have to explain.

In private use

Coach

The trainer that programs for you, not a logbook you fill in.

Most fitness apps are a notebook with a suggestion engine. Coach builds your actual training week, adjusts it as you go, and checks in daily to see how you moved. You show up. It handles the rest.

In private use

Spotworth

Point your camera at something and know what it is worth.

Find something at a garage sale or the back of a wardrobe and know what it resells for before you walk away. Spotworth identifies the item from a photo and gives you a resale range and where it sells, in seconds. Market estimates, not appraisals.

In development

Codebase Reader

Any codebase in plain English, with a safety check on every commit.

Feed it a codebase - your own, one an AI built, one you inherited - and it produces a plain-English map of what the software actually does. It also runs a safety check on every commit so you know what changed and whether anything load-bearing moved.

Internal tool

The engineering underneath.

We build our products on production-grade systems, not demos. That is the difference between something that works once and something you can rely on. Five pieces of the engineering underneath the products above.

01Orchestration

Orchestration that cannot drift from its definition

Complex multi-step work - building, reviewing, shipping, researching - runs on an engine where the running process is the single source of truth, not a pile of scripts that quietly drift from what they were meant to do. Every action a process can take has to be a declared capability before it can run, so nothing gets bolted on ad hoc. What executes and what was specified stay in lockstep, enforced by the system rather than by someone remembering to keep them in sync.

A flow of connected stages - plan, build, review, ship - each running only its declared capabilities

02Durability

Work that survives the gap between sessions

Most automation breaks the moment a task cannot finish in one run. Ours does not. A long-running job can pause itself against a real condition - an event, a dependency, a human decision, a deadline - hand back its resources, survive a restart, and resume exactly where it left off, without ever running twice. It is the least flashy and most load-bearing thing we have built.

A timeline that pauses against a condition, hands back resources, then resumes exactly where it left off

03Verification

The thing that builds the work never signs it off

Nothing certifies its own completion. A separate, independent check verifies the work against reality and gives a clear verdict before anything ships. Higher-risk work - payments, security, production - escalates to a stricter cross-check, often on a different model so it is not blind to the same mistakes. Every completion claim has to point to real evidence; a claim with none fails automatically.

The builder hands its work to a separate independent check that verifies it against reality before anything ships

04Authority

Autonomy with limits the system enforces

Every automated action carries an authority tier. Low-risk work runs freely; anything touching money, external messages, or production waits for a person to approve it - and that limit is enforced at the execution layer, not requested in an instruction. A human is always accountable at the gate; the system routes the decision and holds the line.

Four authority tiers - autonomous, notify after, approval required, and human only

05Memory

Memory, so the work compounds instead of resetting

We built a layered memory model so our systems do not start from zero every run and do not confuse hard facts with recalled context. Authoritative facts, live working state, and searchable past decisions each live in their own layer with a defined precedence. The payoff is products that get sharper with use instead of forgetting.

Four memory layers - authoritative facts, working state, searchable recall, and archive - each with a defined precedence

The systems accelerate the work. Accountable people set the direction, approve the judgment calls, and stand behind the result.

Complex work, made simple.

Not a wrapper on a chatbot. A self, assembled.

First, a self.