Don't keep the memory inside the AI.
Smart AI starts from zero every morning because its memory lives inside the AI. We turn your company's knowledge into a record AI can read — we build the Company Brain.
Just want the material? — AI Readiness Checklist (free PDF)
NGraph provides FDE (Forward Deployed Engineer)-style support.
See the work, the decisions, the exceptions, and the data in use
ABC sorting × Status sorting brings scattered information into the Company Brain
Install Claude Code / Codex on work PCs and connect them to the record
Build the first use case, then refine it in use until it sticks
The result isn't a delivered AI. It's a system used on site that stays with the company.
Ordering an app and receiving it is one way. NGraph's way is the opposite on three points.
You hope it fits the work. Every time the work changes, it fits a little less
Someone on the inside learns the work and owns the results. We build AI tools on top of the tools you already use
It's newest the day it's delivered. From then on it ages, and its value falls
Procedures and rules kept in a form AI can read. The more it's used, the more employees refine it, and its value grows. Designs and data are yours too
A change request goes from staff to vendor to developers and back: days to weeks. The smaller the fix, the further it slips
A procedure is one page of plain text. Employees fix it with AI on the spot, and from the next day every AI works the new way
Write how the work is done as one page of text (SKILL.md, a document that teaches the AI a procedure, in plain language), and let staff on site fix it as they go. From the day after a fix, every AI works the new way.
We don't deliver AI. We build an organization that can fix its own AI.
Work is done per Project. We set the scope, agree on a one-line definition of done, and build it to completion at a fixed price. Every Project sits on the same Foundation.
Time saved ＋ hires avoided ＋ results from reallocated time
Effects are written in only after we measure them. The record organized in the first Project makes the second one faster.
Measure on the first task, then extend the same Company Brain to the next.
Like "a new hire can complete onboarding over LINE": stated so that done or not can be judged in a single sentence.
Only three sizes: small, medium, large. Added scope goes into the next Project. No itemized billing.
The record built in the first Project becomes speed in the second. We don't lower the price; we return it as speed.
Fees are decided by the target task and the integration scope. We define the scope in a free consultation and present an estimate before any formal order.
Pick your industry and answer 8 questions. Your position on 5 axes, shown on the spot — no sign-up.
Shingo Takahashi, CEO
So we make three promises.
We define the scope in a free consultation and present an estimate before any formal order.
Before the price, decide what will remain.
In the first meeting, we pick one target task and map out the entry point for AI and the records it needs, together.
The Company Brain itself, plus maintenance and update rights. It grows thicker with every Project.
We set the scope, agree on a one-line definition of done, and build it to completion at a fixed price. Only three sizes: small (about 6 weeks), medium (about 3 months), large (about 4 months). Added scope goes into the next Project.
The first decision is just one target task. "I only want a ballpark" is fine.
The effect can be estimated as "hours of work removed × labor cost".
For example, "30 minutes of re-entering data every day" is about 120 hours a year — at ¥2,000 an hour, about ¥240,000 worth of work a year.
Our thinking on cost and available subsidies is in this article (Japanese).
Starting from 30 minutes online. We can show you something working over screen share.
Tick the boxes and 20 items show where your company stands and what to do next. Free to download.
Answer "Is our company ready for AI?" — no technical knowledge required.
Yes — that is exactly what this service is for. We explain without jargon, no documents needed, and work alongside your team on your actual tasks.
An engineer who embeds within a client company and builds AI into each team's actual work on the spot. OpenAI, Microsoft, and AWS are all reported to be investing heavily in this model.
We do not publish amounts. Work is priced per unit: a Foundation (one per company) plus Projects, each at a fixed price with a capped duration — small (about 6 weeks), medium (about 3 months) or large (about 4 months). We define the scope in a free consultation and present an estimate before any formal order.
Yes. Online-first is our default: we build live over screen share with your real data, from anywhere in Japan. On-site visits are also available.
Mainly small and mid-sized companies, in any industry. You can start from a single department.
Our own AI products (including AI customer service for restaurants) run live in real restaurants in Fukui and Kanazawa, and we are currently embedded in the HQ of a 12-location restaurant group in Fukui, wiring AI between their existing ordering and accounting systems. We only publish outcome figures we have actually measured — no inflated percentages.
No. We use what you already have (ChatGPT, Claude, Copilot, etc.), or help you select the right one.
No. We design around the systems, Excel files, and chat tools you already use.
Yes. We organize your business data, workflows, and prompts in a form that does not depend on any specific tool, so when a better AI appears you can swap it in. Not being locked into a single AI vendor is a core design principle for us.
We can sign an NDA before starting. Data-handling scope is agreed during the assessment stage.
Think of the work inside your company and tick the items that apply.
If even one applies, talk to NGraph.
In the first meeting, we pick one target task and map out the entry point for AI and the records it needs, together.
"I don't even know what to ask" is a fine place to start. We usually reply within 2–3 business days.
