Find a rule
An employee asks a question about an instruction and gets an answer with the source, in a supported, agreed format.
These are examples of possible tasks, not completed client projects. For the first launch, we choose one main scenario.
An employee asks a question about an instruction and gets an answer with the source, in a supported, agreed format.
The assistant assembles a draft from approved materials; a manager checks the facts before sending.
The assistant fills in an agreed template from the source information; an employee reviews and approves the result.
How wework
An example for discussion. We define the sources, actions and human involvement for your task before development.
An authorized user contacts the assistant through the chosen private interface.
The assistant looks for information in the agreed set of available materials.
Prepares the result and cites the source where the chosen solution supports it.
An employee checks the facts, clarifies contradictions and makes the final decision.
The employee uses the approved result. Sending emails or changing systems requires a separate scenario.
02 / 04
You decide which instructions, product descriptions, regulations and templates are permitted. After reviewing a sample, we'll fix the volume, formats, freshness and update method. We don't set a universal limit on documents or users.
Scans, complex tables and contradictory instructions may need separate preparation. New sources and regular knowledge base updates are estimated separately. Automatic updating of materials is not implied: if needed, we'll include it in the estimate as a specific feature.
03 / 04
The first launch covers reading materials and preparing drafts. The assistant helps find and compile information, and an employee checks it before use. The model can make mistakes; we don't promise error-free operation, full replacement of an employee or a specific number of hours saved.
Changing a CRM, sending emails, creating orders and other actions are separate scenarios. For each, we agree on permissions, employee confirmation and action logging. We check what ready-made integrations can do before including them in the scope.
We define test cases and expected behavior in advance. At handover we check them on the finished solution; these are criteria for the future work, not results already achieved.
from $200
Tell us what problem you need to solve and what result you want. After we discuss it, we'll agree on the scope of work, a fixed price and a deadline. A detailed technical specification isn't required for your first inquiry.
If we need to study the process in detail, review sample materials or check connections, we'll agree in advance on a paid assessment and its deliverable: a solution outline, the boundaries of the work and an estimate. The assessment is paid in full before it starts. You can receive it as a standalone deliverable; if you order the implementation, the payment is credited toward the total price.

We subtract the already paid assessment from the total agreed price. The remaining amount is split: 25% before implementation and 75% after the result is checked. The assessment is not paid for twice. We estimate additional scenarios and scope changes before work begins.
Platform subscriptions, the AI model, API, hosting, storage and paid connections are billed separately. Before the start, we'll show the expected costs for the agreed number of requests or operations and state our assumptions. Costs may change as the load grows; we'll set up limits and notifications wherever the service allows it.
We'll hand over the working agreed scenario, client access and short instructions. Where possible, accounts are registered in your name. Acceptance is recorded by your dated confirmation in the agreed channel. For 30 days from confirmed acceptance, we fix errors in our implementation free of charge, within the agreed scope.
New tasks, material updates, adapting to changes in third-party services and ongoing support are estimated separately. If the planned failure handling was implemented incorrectly by us, that is our error. Website support plans do not automatically apply to bots and automation.
If the project is handed over without further maintenance, you revoke our permissions and change the passwords and keys known to us, following a list. We record what was handed over, ask for confirmation and remove our own access where we can. For agreed ongoing support, we keep only the necessary permissions.
The internal assistant works on an employee's task and the team's approved documents. The website customer consultant is a separate service.
First we review a sample, quality, formats and rights. We define the contents of the knowledge base and its preparation for each order, with no blanket promise to handle any volume.
We confirm support and format before work starts. If the chosen solution doesn't support them, we agree on the limitation and an alternative way to verify answers in advance.
The basic mode is reading and drafts. Actions are estimated separately, with permissions, employee confirmation and a record of the result.
Only if this update method is included in the agreed scope. Regular manual updates and new sources are discussed separately.
That depends on the chosen architecture and services. We agree on storage location, external processing and deletion before starting; we don't promise local-only storage by default.
Describe the process, website or channel, the systems you use and the result you need. We'll discuss sample documents and access later in an agreed channel. Don't send passwords, keys or confidential materials through a public form.
Discuss the assistant's task