Sachi AI
Get updates

AI agents for business that act in your systems and wait for approval

For the client projects we take on, there are two ways to start. An AI harness gives your staff one AI workspace, connected to your systems, with access set by role. A person asks and a person sends. An AI agent does a whole task, such as matching an invoice to its order, and stops for a person to approve each step that matters. Both act only within their permissions, and both record every step.

Each offer stops for a person at a set point

Each card names the offer and its first approval point. Each offer page walks through one example, then the safeguards and what you receive.

Compare the six offers in detail
OfferSystems it touchesWhere a person approvesWhat you receive
AI harness for your teams

An AI workspace for your teams, connected to your systems through permissioned tools, with access set by role, a record of every action and training for staff.

  • Your ERP
  • SharePoint and document stores
  • Outlook and Teams
  • CRM platforms
  • Before anything the AI drafts is sent or posted
  • Each role reaches only the systems and data it is already allowed to see
  • Your system owners approve each new connector and skill before staff get it
  • The AI workspace, connected to your systems
  • Shared instructions and skills for the tasks your teams repeat
  • Access rules, an audit log, usage reports and training
Agents for operations

The agent reads each request, checks your ERP, mailbox and documents, and drafts the reply or the transaction. A named person approves the steps that matter, then it acts and logs each one.

  • Your ERP
  • Outlook and shared mailboxes
  • SharePoint and document stores
  • Ticket and request systems
  • Before anything leaves the organisation
  • Before any posting or payment in the ERP
  • When the agent is unsure, it hands the case to a person and explains why
  • A working agent
  • Test cases built from your real work
  • Operating guides
Connecting agents to your systems

Secure connectors (MCP) that bring AI agents into your ERP, Microsoft 365, your CRM, databases and internal APIs, each with set permissions.

  • Your ERP
  • Microsoft 365: Outlook, SharePoint, Teams
  • CRM platforms
  • SQL databases and internal APIs
  • Actions that change data can need approval, tool by tool
  • Your system owner reviews new tools and wider access before they go live
  • MCP connectors
  • A permission plan, role by role
  • An audit log of every tool request
Document AI and knowledge assistants

Document AI that answers from your own documents and data with the source shown for every answer, and that reads scans, invoices and forms into structured data for a person to check.

  • SharePoint and file shares
  • Policies, contracts and manuals
  • Scans, invoices, forms and handwritten answers
  • Structured data from your databases
  • Document owners check answers on their topics before launch
  • If no source supports an answer, it gives none rather than guess
  • Extracted fields it is unsure of go to a person before they are used
  • A search index of your documents
  • The assistant, or the extraction pipeline
  • Tests built from your own staff’s questions
Assistants on web and WhatsApp

An AI assistant that answers and acts on WhatsApp, email and your website, checking your systems first, with staff approving its work until you trust it.

  • WhatsApp Business Platform
  • Shared mailboxes
  • Website chat and forms
  • CRM, order and booking systems
  • Approval first: staff send every reply
  • Rules for passing a message to a person, agreed with you by topic and by customer
  • Complaints and sensitive topics always go to a person
  • Connections to your channels
  • Reply templates in your customers’ languages
  • Rules for passing messages to a person
Agent evaluation and QA

Evaluation and QA for AI agents, ours or anyone’s: simulated users test the agent before launch, then run records, regular tests and cost control keep it right once it is live.

  • Your agent’s endpoint, whoever built it
  • A record of every run
  • Tests built from your real work
  • Cost and usage reports for each agent
  • Your team signs off the test results before launch
  • Changes to instructions, models or tools go live only after the tests pass
  • Agreed limits send an alert to a person
  • A set of tests and simulated users
  • A report of what broke
  • Dashboards and a monthly review

Every offer is built on the same base:Security and where your data livesPrivate AI on your own GPUsAI strategy and governance

The harness and the agents run on one base

Your staff work in one place, with access set for each role. Agents act in your systems within your rules, and every step is recorded.

  1. Your teams

    • Finance
    • Sales
    • Operations
    • HR

  2. The harness

    • Access by role
    • Shared skills
    • Connectors
    • Approvals
    • Usage and cost

  3. Models and your systems

    • AI models
    • Your ERP
    • Microsoft 365
    • CRM
    • Email
  4. Every action recorded
Diagram of the AI harness: Finance, Sales, Operations and HR teams use the harness, which holds access by role, shared skills, connectors, approvals, and usage and cost. It connects to AI models and to your systems: your ERP, Microsoft 365, CRM and email, and records every action.

Each stage is agreed in writing before it starts

We price each stage before it begins. You can stop at the end of any stage.

  1. 01Discovery

    We choose the task with you and list the systems it touches. We agree what success looks like and where a person approves. The scope is fixed in advance.

    Ends withA written yes or no on going ahead, with the reasons.

  2. 02Trial agent

    One agent does one task on your real data, in live use. A person approves its actions. The scope is fixed in advance.

    Ends withA working agent in your code repository, with its test cases and operating guides.

  3. 03Grow and run

    We add tasks one at a time. A support contract keeps each agent tested and monitored, and we fix it as your cases change.

    Ends withA monthly review of each agent’s results and cost.

Questions buyers ask first

What drives the cost of an agent?

Mostly the number of systems the agent connects to and how much testing the task needs. Our FAQ page lists all five factors. We price each stage in writing before it starts, and you sign off a fixed figure.

Where does our data stay?

Where you choose: your servers, your cloud region, or our private GPUs. Our security page sets out where each part runs and what is recorded.

Which models do you use?

The one that passes your tests at the level of risk you accept. For sensitive data, that is usually an open model running privately. For work that is not sensitive, a model from a large provider can be the better choice. We build the agent so the model can be changed without rebuilding it.

What happens when an agent gets something wrong?

Important actions wait for a person, so a wrong draft is caught before it goes out. Permissions limit what any mistake can touch. Every run is recorded, so the cause can be traced to one step. The failed case joins the test cases, and every later update is checked against it.

Who owns the code?

You do. The agent, its connectors, its tests and its guides live in your repositories. Our approach page explains the handover.

Send us one task, and we will name the offer that fits

Describe the task and the systems it touches. We will reply with the offer that fits and the first stage we would propose.

Start a project