KINSUGI
Build what you need from what's already proven.Gain is your AI engineering partner on Kinsugi. One conversation, decisions that stay with you, outcomes you can deliver.
Origin
Cracks rejoined with gold.
Kintsugi is the Japanese art of repairing broken pottery — the fractures aren't hidden, they're sealed with lacquered gold and made load-bearing. Kinsugi reads enterprise software the same way. Composed deliberately, the seams between systems are where the structure gets stronger.
What it is
What Kinsugi is.
KADE — the Kinsugi Agentic Development Engine — pairs two things: a trusted library of ready-made, reusable building blocks, and an AI partner, Gain, that works alongside you inside the product. It isn't tied to any one AI model.
KADE
Kinsugi Agentic Development Engine
Gain works from code that's already been tested and approved, rather than inventing something new each time. Every job has its own building block, and every new feature is assembled from blocks that have already passed our checks. KADE builds by combining what's proven — it never falls back to writing everything from a blank page.
Analogy
Imagine a workshop where every part is sorted, labelled, and already tested. You build a chair by choosing four legs that match a seat that match a back — not by carving each piece from scratch. Our library is that workshop. Gain is the partner who knows where each piece lives.
Why it matters
Most AI coding tools give you a printout from a search engine. We give you a workshop where the parts have already been tested for the job.
How we work
How Gain works with you
One conversation, all the way through
Tell Gain what you're stuck on. The conversation keeps its thread through every step, every device, every session.
Your context stays with you
Every conversation remembers the last one. Switch devices, come back tomorrow, change topic — Gain keeps track of where you are and what you've already decided.
You stay in charge
Gain proposes; you decide. Every choice is recorded, every action is auditable, and you can change direction at any turn without losing the work behind you.
Mental model
How the conversation changes.
Working with a ready-made library changes the questions you ask — and different questions get you different answers.
How you ask
Generic AI workflow
Build me a component that does X.
Kinsugi workflow
Do we already have something that does X? What does it not cover? What is the smallest new piece to add?
How it answers
Generic AI workflow
Generates code from scratch, hoping it matches your stack.
Kinsugi workflow
Every part it uses has already passed its checks. All that's left to get right is how they fit together, not the whole result.
What you trust
Generic AI workflow
The model's confidence.
Kinsugi workflow
The checks that ran afterwards. Trust is earned by the checks, not claimed by the AI.
How you verify
Generic AI workflow
Read the code. Run the tests. Hope the tests cover it.
Kinsugi workflow
Checking is built into the work, not a step tacked on afterwards. Every claim is shown next to the proof it passed.
What survives
Generic AI workflow
A snippet. Maybe a file. Maybe nothing in two weeks.
Kinsugi workflow
A building block. It does one job, combines with others, and can be removed in a few steps if you change your mind.
Disciplines
How we work, in five lines
▸ We treat every claim as unproven until we've checked it against the code.
▸ We build from what's proven. We don't invent what isn't.
▸ When the library doesn't cover what you need, we flag the gap as a task — we don't paper over it with a guess.
▸ Documentation and code are produced together, so they can't drift apart.
▸ A check either passes or it doesn't. There's no 'warning' level to quietly ignore.
▸ Approval is earned, not assumed.
Why it works
Why it works.
It starts from proven work, not guesswork.
Gain builds from blocks that have already passed their checks, so the only thing left to get right is how the pieces connect — not the whole result. Far less can go wrong.
One job per building block.
Adding a new block gives you new options without disturbing the ones already there. The library grows piece by careful piece, instead of turning into one tangled mass that's hard to change.
Documentation that can't fall out of date.
The documentation is generated straight from the working code, so the two can never tell different stories. When the code changes, its description changes with it — nobody has to remember to update it.
Constant checks, like guardrails on a road.
79 checks, 40 scanners and 5 automatic fixes run on every single change we make to Kinsugi itself — and on the code KADE builds for you. Problems are caught the moment they appear, not weeks later.
How it works
One conversation. A clear next step at the end.
You ask
Gain listens
Gain proposes
You decide
Gain remembers
Each turn is yours. Each decision is on the record. Come back any time — Gain picks up where you left off.
Pain points
How we handle the hard parts.
The usual complaints about AI-assisted development each have a specific answer.
It hallucinates code that doesn't exist.
Working from fact, not guesswork.
The partner only assembles from parts that have already cleared their checks. When the library doesn't cover what you need, you get an honest "we don't have that yet" — not a convincing wrong answer.
Analogy
A librarian who tells you 'we don't have that book yet' is more useful than one who hands you a book with the right title and the wrong contents.
The docs always lie about what the code does.
Documentation can't drift from what authors it.
Code and documentation are produced together as one thing. A change that updates one but not the other doesn't pass.
I have to re-explain everything every session.
Built-in context, not chat history.
The library and its structure are always on hand. The partner reads them directly instead of asking you to paste the same explanation again.
The output is brittle and breaks the moment I touch it.
Constant checks as part of the work, not after it.
Automatic checks and fixes run alongside the AI as it works. Output that breaks one of your rules never reaches you in the first place.
It writes ten files when I asked for one.
Assembled from parts, not generated from scratch.
Features are built from existing blocks plus a little connecting code. Writing everything from scratch isn't an option, so there's nowhere for the bloat to come from.
Analogy
A sculptor working with marble removes; a 3D printer adds. We sculpt.
I can't tell what the agent actually did or why.
Every decision on the record.
The AI's reasoning and the result of the checks are shown together. You see what was claimed and what was verified. A full trail is the default, not an add-on.
Analogy
A surgeon explains the procedure before the cut and writes the notes after. The trail is the trust.
The human
Where the human stays.
Before any agent runs, the question is: where does the human stay in the loop?
Gain suggests; you decide. Every decision is written down, and every action can be traced back. Your rules, your scope and your limits are set once and applied everywhere — you don't have to explain them again each time.
Automation handles the back office. Judgment, taste, and accountability stay with the people who own the outcome.
What changes
What changes about how you deliver.
Three things change once the library and the AI partner are in place.
Less rework.
Work starts from code that already passes, not from a prompt and a hope.
Less drift.
The documentation is generated, not hand-kept. It and the code come from one source, so they can't disagree.
Less retraining.
What the team has learned is carried across sessions and devices, so engineers stop re-arguing the same decisions.
The time saved shows up wherever a team has been fixing AI output instead of delivering. We measure it, and we'll publish the results when the benchmark is ready.
Timing
Why now.
The cost of running AI has fallen roughly a hundredfold in two years. Work that was too expensive to automate in 2024 now runs around the clock. Analysts forecast a US$170B market by 2030 — yet a reported 95% of enterprise AI projects never earn their money back. Most fail for two reasons Kinsugi is built to avoid: AI that invents everything from scratch, and AI that forgets what you told it last time. The chance to shape how this is done is open now, and it won't stay open long. Kinsugi is built for teams that want AI to work on its own — without handing over control.
Proof
Already in production.
Kinsugi was built to run Beasr.world — a marketplace that makes moving home simpler, so people aren't left juggling endless choices, suppliers and costs. Gain is the engineering partner you meet here; on Beasr.world, KADE powers three customer-facing assistants that share a single memory across moving in, styling a home and settling in.
Moving assistant
Aida
Plans the move end to end — budgeting, timelines, the rent-or-buy decision — and remembers what the customer already chose at every step.
Plan a move on BeasrStyle and space
Livia
Recommends furniture, décor, and home solutions that sit together. Style and practicality, sized to the room and the budget.
Style a room on BeasrLocal knowledge
Chinwag
Surfaces the community context — neighbourhood tips, local services, the small things that make a new place feel like home.
Join ChinwagFrom a single live deployment, Kinsugi is in active commercial discussions across six sectors: travel, finance, fashion, office supplies, cyber security, banking. Chair Capital is partnered for the next stage of scale.
Visit Beasr.world