Questions, answered directly

Our own AI products, in the same line as the work we know: AI agents that do real work in the systems people already use, stop for a person to approve what matters, and keep data private. They are in development and not named yet. We will not describe a feature as finished before it is.

Leave your email under Get updates on the Products page, and we write when there is something real to show. To talk about a partnership, write to hello@sachiai.com.

Yes, selected ones. Our products come first, so we take on a project when it fits the work we know: AI agents, private AI and the software around them. Send a few lines through the contact form and we say honestly whether it fits. The Work with us page sets out what we build for clients.

An AI agent is software that handles a set task from start to finish. It reads the request, checks your systems through connections we build, and prepares the next step. It stops for a person before anything important. It works inside the systems you already use, such as your ERP, Microsoft 365, your CRM, email or WhatsApp. Every step it takes is recorded.

Approval mode means the agent prepares an action, and a named person approves it before it happens. The reply is written but not sent. The credit note is prepared but not issued. We agree with you who approves what before we start. A trial agent works on live tasks in approval mode. A step leaves approval mode only when you decide its track record is good enough.

Where you choose: your servers, your cloud region, or our private GPUs. Each connection has fixed limits, so an agent can reach only the systems and actions its role allows. We pick the AI model at the start, based on where your data must stay. The Security page sets out each option.

Five things drive it. The first is how many systems the agent connects to, and how well they are documented. Then come how many steps need a person to approve, how many cases it handles and where the model runs. The last is how much testing the task needs before you trust it.

Important actions wait for a person, so the approver catches a wrong draft before anything happens. The run log shows the step that went wrong. We add that case to the agent’s tests and fix it. The case is then tested again after every later change.

Yes, as a client project. Evaluation and QA of AI agents is one of the things we take on. We hold real multi-turn conversations with your agent as the kinds of users you describe, score them and report what broke. We test through the same endpoint your users reach, so you do not need to change your agent to have it tested.

We do not publish a price list, because the work varies too much for one price to mean anything. Each phase (assessment, plan, pilot or build) is defined and priced in writing before it starts. You agree a fixed price for that phase.

It changes where the time goes. AI coding agents do much of the typing, so more of the time we charge for goes on design, review and testing. We do not promise a saving we cannot measure on your project. The fixed price for each phase is what you compare.

Yes. You can stop between any two phases and keep everything made up to that point: code, documents and server setup. Nothing is held back to keep you with us.

No. We scope each engagement as a fixed, agreed piece of work: an assessment, a plan, a pilot, or a set build. For larger change and platform work, an ongoing contract is available if a client wants one.

Any change to the agreed work is priced again in writing before we start on it. We do not squeeze extra work into a fixed price, and we do not charge for it without telling you first.

No. We also build web apps and platforms, Flutter mobile apps, backends and APIs in FastAPI and Node, integrations with the systems you already run, and automation. An agent depends on the systems and data it works in, so we build those as well. One team builds both, to the same standard.

Yes. We connect agents and apps to the ERP you run, through its own interfaces, with set permissions, and build the integrations and automation around it. We are not a reseller or a certified partner of any software maker. Where the maker’s own partner is the better choice for part of the work, such as setting up the ERP itself, we say so.

Yes. We build mobile apps in Flutter for iOS and Android, with the web apps and APIs behind them. One codebase serves both platforms, and the app talks to the same backend as your web app.

AI coding agents work inside a harness we maintain, and a person on our side owns every piece of work and approves every change. Every change passes tests, type checks and a build gate, and a reviewer in a fresh context tries to break it before it is merged. When several agents work at once, no file has two writers. The How we build page sets out the method.

Yes. We start by reviewing it. We read the code, how it is deployed and any documents that exist. Then we write an assessment of what is sound, what is weak but important, and what it costs to keep or to replace. If replacing it is cheaper over the life of the system, we say so and price that as a separate decision.

No. We do not train AI models on client data. When we host on our private GPUs, your data stays there only for as long as the contract says. Otherwise we keep no copy after handover.

In the planning phase, we map the rules that apply to you onto your system, whether it is an app that handles personal data or an AI model: the data protection law where you operate, your sector’s rules and the security standards you follow. We hold no certifications and claim none. We build so your own compliance team and auditors can check the controls.

Yes. Private AI on open models, including systems fully cut off from the internet, is one of our services. The apps around the AI model are built to run inside the same closed network.

You do, whether the work is an ERP connection, a mobile app or an AI agent. The code lives in your own repositories, under your organisation’s accounts, from the first commit. The How we build page lists everything you own.

No. If you stop working with us, nothing stops running. The code, the server setup files, the models and the documents are already yours.

Yes. We write the documents for your engineers, and the handover sessions end when they can change the system on their own. Any support after that is a separate contract with set hours.

We work remotely, across time zones, and we work in English. On a client project we agree meeting times that suit your working day, and most of the work happens in your own code repositories and accounts, so you can see it as it moves.

Write to hello@sachiai.com or use the contact form with a few lines about the work and the systems it touches. A person reads it and replies with a first view of the work, and, if it fits, a call to go further.

Yes, in the first conversation and before any proposal. Often the better answer is ordinary software, such as a script, a report or a link between two systems. The How we build page lists the usual reasons an agent does not fit.

We publish a case study only after the client approves it, and we will not invent one. What you can judge before you sign is the method: the How we build page shows how we plan, build and check the work, and where a person approves. In the first conversation an engineer walks you through how your own case would work.