Everyone prompts ChatGPT their own way and results vary every time
We build one assistant with a clear instruction, examples of good answers and a fixed output format. Everyone gets the same quality.
An AI assistant that knows your rules and handles your team's routine work
We build a Custom GPT or Claude Project for one specific job: sales proposals, support replies, document checks, onboarding. It runs on your own documents and rules, not on generic model knowledge.
Quote and schedule after your brief
Who it's for
- Sales teams that keep rewriting similar proposals and emails
- Support teams answering the same questions from an internal knowledge base
- Marketing and content teams that need a consistent tone and templates
- Managers who want new hires working to company standards within a week
Problems we solve
Company knowledge is scattered across PDFs, chats and people's heads
We collect the material into a knowledge base (the files the assistant pulls facts from) and define what it does when the answer isn't there.
I worry it will invent facts or leak confidential data
We limit its sources, add an "if unsure, ask a human" rule, and configure accounts so data is not used for model training where your plan allows it.
We tried it and the team stopped using it
We fit the assistant into the workflow people already follow, run a short training session and hand over a cheat sheet of common prompts.
What's included
- Task review and platform choice: Custom GPT (OpenAI) or Claude Project (Anthropic)
- System instruction defining the assistant's role, tone and limits
- Knowledge base preparation and cleanup: policies, FAQs, templates, examples
- A test set of prompts to check answers before launch
- Rules for edge cases: when the assistant declines or hands off to a human
- Access and privacy setup for your team
- API actions to CRM or spreadsheets where needed, in higher packages
- Team guide and a training session
- Pilot with a measurable KPI, such as time to prepare a proposal
- Accounts and all files handed over to you under the agreed terms
Choose the scope
Use a package to describe what you need. We agree on the quote and schedule after reviewing your brief and requirements.
01
Single assistant
One clear task in one department
- Task review
- System instruction
- Knowledge base up to 20 documents
- Testing on 20 prompts
02
Assistant with data
Teams with a large set of policies and templates
- Everything in Single assistant
- Knowledge base structuring up to 100 documents
- Rules for complex and disputed cases
- Team training
- 2-week pilot with a KPI
03
Assistant suite
Companies where several departments want their own assistants
- Up to 4 assistants with a shared style
- CRM or spreadsheet integration via API
- Documentation and update guide
- Pilot with a KPI and a report
How we work
Brief and requirements
You describe the task and current workflow. We clarify requirements, choose a first stage and prepare a quote.
Task review
Together we pick one routine task and agree how to measure the result.
Knowledge base
You provide documents, we clean and structure them. Outdated and conflicting files get dropped.
Setup and testing
We write the instruction, run test prompts and fix weak spots.
Pilot with a KPI
Your team uses the assistant for a few weeks. We compare the outcome to the agreed metric.
Handover
You get the accounts, files and guides. Payment is in stages and everything belongs to you.
Scope and limits
- We don't promise the assistant will never be wrong: important documents and numbers still need human review
- We don't replace staff: the assistant speeds up routine work, decisions stay with people
- We won't build on a knowledge base that doesn't exist: if nobody has written the procedures down, that comes first
FAQ
What's the difference between a Custom GPT and a Claude Project?
A Custom GPT lives in ChatGPT, a Claude Project lives in Claude. Both let you set instructions and upload files. We pick the platform based on the task, the languages you work in and the plans you already pay for.
Do we need to buy subscriptions?
Yes, your team needs paid plans on the chosen platform. Pricing depends on seat count and changes, so check the vendor's current rates. We set the accounts up in your name.
Is it safe to upload internal documents?
It depends on the plan and settings. We check that data isn't used for model training and advise against uploading passwords or customers' personal data. For sensitive cases we'll discuss other options.
How many documents do we need to start?
For a simple task, 10–20 files are enough: a policy, templates and a few examples of good answers. Quality matters more than volume, and outdated files hurt results.
How do you measure the result?
Before we start, we agree on one metric: for example, time to prepare a proposal or the number of edits from a manager. We measure before and after the pilot.
What if we want to change things later?
You can update the instruction and knowledge base yourself, and we leave a guide on how. If you want help, support is billed separately at a published rate.
Start with your task
Describe your business, the outcome you need and your current workflow. We will clarify the requirements and prepare a quote after your brief.