Question 1

Can I see how it works before buying?

Accepted Answer

The demo is available immediately after registration, with no payment required. Sign in or create an account, then download the demo build from your account. Registration includes 500 trial coins for the AI assistant; you can buy the full version later if you need every capability.
You can get a feel for it from the video demos and screenshots on the product page: they show real scenarios — the assistant building a query against a configuration, refining an existing one, and answering questions about the result.

Question 2

How does the demo differ from the full edition?

Accepted Answer

The demo is free, can be downloaded from your account immediately after registration, and has no expiry date. Its AI assistant reads your configuration metadata and builds and explains a query; the Console then runs it against your database and shows the complete result. The connectable Standard Subsystems Library report and data processor builders are also available in the Demo.
Queries run in both the Demo and full editions. The Demo includes the standard query editor, conversation history, and basic and balanced AI models. Their current roster appears in the edition comparison and may be updated by the server.
What the Demo does not include: the full edition's AI agent inspects result data once you confirm it — the Demo has no such capability at all. The full edition also adds advanced and flagship AI models, a larger context allowance, the professional editor with autocomplete and syntax highlighting, the package and temporary-table navigator, background execution, and extra themes.
Trial coins for the AI assistant are granted at registration. When they run out, you can buy coin packages while the Console and the Demo itself remain usable without coins.

Question 3

Is this a subscription? How many times do I pay?

Accepted Answer

There is no subscription or recurring fee. The full edition is a one-time purchase and remains yours.
The free demo has no expiry date either and does not need to be renewed.
Two things are paid separately with coins: AI assistant usage and building a finished file — an external data processor (.epf) or an external report (.erf). Coins are spent as you go and topped up in packages when needed. AI model rates and the build price are published on the separate pricing page.

Question 4

How do I install the Console and connect the AI assistant?

Accepted Answer

After signing in, your account lets you download the .epf build available to you and issue an API key — a personal access code to the AI service, much like a long password. The demo requires no payment; after purchase, the full version appears on the same page.
The steps are:
Download the data processor from your account and open it in 1C via "File → Open" (or add it to additional data processors).
In the chat header, click the connect button and paste the API key.
Click "Sign in" — the key is verified on the server, the list of available models and your balance load, and the chat unlocks.
The key is stored in the information base where you entered it, so you won't need to retype it there. In a different base the console will ask for the key again: enter the same one — a single key covers all your workstations and bases.

Question 5

What exactly am I buying: the data processor or AI access?

Accepted Answer

The demo is available without payment. Buying the full version gives you a perpetual licence for the complete query console; AI is an additional working layer on top of it.
The .epf data processor is the query console itself: the query builder, the editor with syntax highlighting, parameters, temporary tables, result drill-down, background execution and export to Excel. It is yours permanently and works without an internet connection and without coins.
The API key is access to the AI assistant that writes and refines queries. Assistant requests are paid for in coins.
Only the AI assistant and separate paid services consume coins. Registration grants trial coins, while buying the full version includes a separate starting-coin package; its current amount is stated on the product page.
The terms for returning a purchase or unused coins are covered in a separate answer.

Question 6

What are coins, and why do models cost differently?

Accepted Answer

Coins are the single currency for the console's paid operations: AI assistant requests and building finished files. Each request spends coins at the rate of the model you selected in the chat; a build is priced per file rather than per volume of text — see the answer on paid services.
Registration includes 500 trial coins. Buying the full version includes a separate package of starting coins; the current amount is stated on the product page. Additional coin packages can be purchased independently of the full Console version.
Orders of magnitude matter more than theory. A small query on a basic model costs about 100–200 coins (≈ $0.25–$0.50), while an average Claude Sonnet query costs about 800 coins (≈ $2). In a real case, preparing a complete report through several AI iterations cost about $19 for the finished task; see the detailed breakdown on the economics page.
Why there is no flat price per request. Models are billed for the volume of text they read and wrote, and that depends on the task: "add a manufacturer field" and "build a full-year settlements report" differ by orders of magnitude. That's why pricing lists rates rather than a price per item.
Why models cost differently. These are the AI providers' own prices: higher-tier models reason longer and cost more for the same volume. The spread between the entry-level and the top model is several-fold, and it is the main lever on your spending — far bigger than the length of your question.
What is charged and what is not. Coins go out per request to the AI model and per finished file built. Follow-up turns in the conversation — "add the manufacturer", "drop that filter" — are separate requests and are charged each time. Building an .epf or an .erf is not a neural network request: it is server-side work priced per file. A network failure, a missing model response and a failed build all charge nothing.
A complex task may need several refinements: a conversation in which you improve the result step by step is a normal working scenario. The cost remains visible for every turn — price and date on hover; context-window fill, input and output tokens, cache share and speed on click. The figures stay in the conversation history, so the spend can be reviewed later.
Practical advice. To avoid overpaying, start a task on smaller models and move up only when the selected model cannot cope. The full model table with rates is in pricing.
Your balance is visible in the chat header and in your account, where packs are also purchased.

Question 7

How can I pay, and will a CIS bank card work?

Accepted Answer

Payments are processed by CloudPayments and Robokassa. The payment form accepts Visa, Mastercard and Mir cards; Russian banks can also use the Faster Payments System. Cards issued by banks in Kazakhstan and other CIS countries are accepted, although the issuing bank makes the final decision on each transaction.
If a payment fails, try another card or payment method. If it is declined again, contact support@digitalmechanics.dev and we will help investigate.

Question 8

Can I return my purchase or unused coins?

Accepted Answer

Full version. Once the data processor file has been downloaded, the licence transfer is considered complete: the file stays with the buyer and the purchase cannot be returned. This does not limit mandatory consumer rights, including where the service is defective or has not been rendered.
Coins. When you close your account, the remaining paid coins are refunded at a rate of 5 coins per ₽1, less the payment provider's fee, within 30 days of the request. Spent, starting, bonus and promo-code coins are not refunded.
If you believe you have grounds for a refund, write to support@digitalmechanics.dev — we review every case.

Question 9

Is the purchase permanent, and how many databases and workstations does it cover?

Accepted Answer

Yes, and here is what "permanently" stands for: you buy a working copy of the product and it stays with you. There is no limit on the number of your own databases or workstations. If the price goes up later, the licence you bought stays yours — there is nothing to pay again.
The API key is personal and tied to its account owner's wallet. Wherever the key is used, every AI assistant request is charged to that wallet, and the key owner is responsible for requests made with it.
Console updates are included with the purchase and do not require another licence payment — you receive them for as long as the company exists. That is an honest boundary, and we prefer to state it ourselves. New versions go to the buyer, so a copy passed on without access to the account falls behind over time.

Question 10

Can I work in my clients' databases and charge for it?

Accepted Answer

Yes. The licence sets no limit on the number of databases, and your clients' databases are part of that number — the licence agreement states it in a line of its own. What you earn for your work stays yours: consulting, customisations and reports produced with the Console are your ordinary service.
The boundary runs along access to the software itself. You may not sell it, rent it out, transfer the licence for a fee, or open paid access to your key to outsiders. The short form: being paid for your work is fine, being paid for access to the software is not.
The wallet is one per account: however many databases you serve, requests to the AI assistant are charged to your balance.

Question 11

Which 1C versions and configurations are supported?

Accepted Answer

Any configuration — standard or custom — on the 1C platform 8.3.16 or newer in managed forms mode, with a file-based or client-server database. The 1C client must run on Windows, 32- or 64-bit; Linux is not supported yet, it is on the roadmap.
Background query execution relies on the Standard Subsystems Library or a configuration extension — the extension installs from the link in the form header.
National configurations work on the same basis: the assistant relies on your database's metadata rather than on knowing a specific standard configuration, so versions for Kazakhstan, Belarus and other countries built on the Standard Subsystems Library read the same way Russian ones do. One honest caveat: it sees structure and object names exactly, but knows the local accounting specifics less well than the Russian ones — advice of that kind is worth double-checking.
The full list with versions and addresses is on the technical requirements page.

Question 12

Do I need MCP, a VPN, or a configuration export?

Accepted Answer

None of them. The console is an external .epf data processor: download it, open it in 1C via "File → Open", paste your API key. No extra service appears next to your database.
No VPN. All model requests go through a single address — the LenSA API; the console never calls foreign providers directly. Details for your administrator are on the technical requirements page.
No configuration export. The assistant reads the metadata of your live database as the conversation goes — selectively, only what the current task needs. There is no upfront XML export and no re-sync afterwards, so its knowledge of the database cannot go stale: change the configuration and it sees the change immediately. For the same reason there is no cap on the number of databases — see what the purchase covers.
What you do need is full rights in the infobase: with restricted rights the assistant sees only part of it. A 1C specialist usually has them already.

Question 13

Will it work in a closed network or behind a proxy?

Accepted Answer

The console needs outbound HTTPS to a single address — the LenSA API: model requests, key validation, balance and updates all go through it. Proxies are supported: platform settings, your own HTTP proxy, or a direct connection.
If your network is closed, hand your administrator the technical requirements page — it states the exact address and provides a ready-made PowerShell command that verifies access from the user's workstation. One detail that matters for firewall rules: allow the domain name, not a fixed IP — the infrastructure address may change.
We impose no country restrictions: there is no geo-blocking on our side, and the workstation still needs only that single outbound address. This is a working setup for anyone maintaining Russian databases from outside Russia. On payment, see payment methods.
The AI assistant cannot work fully offline: the models live on the providers' side. The console itself works without an internet connection.

Question 14

1C shows a "Security warning" when opening — is that normal?

Accepted Answer

Yes, that's the platform's standard "Protection against dangerous actions", not a sign of trouble.
There will be two warnings: when the console opens — the built-in code editor is a full web editor, and 1C asks permission for its service files — and on your first message to the AI chat. Click "Yes": subsequent openings of the same version won't ask again.
If the editor fails to load for some reason, the query can be typed in a plain text field — your work isn't blocked.

Question 15

How is this different from just asking a neural network directly?

Accepted Answer

The assistant sees your configuration; a chat does not.
Exact names instead of guesses. The assistant explores the structure of your specific database — catalogs, documents, registers, their attributes and composition — and looks up concrete types and metadata names in it. An external chat can't see your configuration and invents names: you get a plausible query that doesn't run.
The platform validates, not the neural network. Before a finished query reaches the editor, 1C's own query validator parses it. The check is external to the model: whatever wrote the query is not what approves it.
The work is visible and verifiable. The chat shows which objects the assistant inspects, what it changes and what it checks. For complex tasks it first proposes a plan, which you can confirm or adjust before it is applied.
Diagnostics from actual metrics. For questions like "why is the selection empty" or "why is it slow", the assistant answers while seeing the row count, temporary tables, joins, console output settings and execution time. An external chat has no access to any of this.
Without separate confirmation, rows from your database are not sent to the neural network — see what is sent to the neural network.

Question 16

Will LenSA explain someone else's query?

Accepted Answer

Yes. Paste a finished query and ask what it does — the assistant breaks it apart: which tables are joined, where the filters come from, what each temporary table contributes, where the bottleneck is. For a query from your own database it leans on the real metadata, so it names the actual objects and rates the risk against your data volumes.
A query from a different database is explained too, but the assistant flags it: without metadata the explanation stays theoretical — based on the structure of the text rather than on what stands behind it. That is deliberate: a tool that pretended to see an unfamiliar database would be more dangerous than one that admits it.

Question 17

What does the dialogue fill indicator show?

Accepted Answer

The indicator shows how much of the context window the dialogue already occupies. It is not task progress: 31% means roughly one third of the available context is in use, not that the task is 31% complete.
Context is the entire dialogue history that the model re-reads on every request. The fuller it is, the more each subsequent request costs. That's why history is worth compacting: the assistant retells the gist of the conversation to itself briefly and drops the long transcript.
Compaction works in two modes: automatically as the context fills, and manually — the button becomes available at 30% fill. The context limit is 120,000 tokens in the demo and 200,000 tokens in the full version. If you've moved to an entirely different task, it's cleaner to start a new dialogue than to compact the old one.

Question 18

Does the cost of a request depend on the size of my configuration?

Accepted Answer

Almost not at all. LenSA does not read all configuration metadata at once — it requests only what the task needs.
The sequence is: the assistant looks at metadata statistics, checks likely object names against the task — sales documents, for example — and only then expands the attribute composition of what it found. Metadata is read selectively as needed rather than exported all at once.
What drives the cost is not the overall size of the configuration but the complexity of the task: how many objects it has to touch. One sales document or five — that is what shows up in the spend.
Re-reading the accumulated context within one dialogue is billed at the cheaper cache rate. A model provider's cache has a limited lifetime, so after a long pause part of the context may once again be billed at the standard input rate. The practical takeaway: finishing a task in one sitting costs less than coming back to it a day later. The cache also resets when you switch models mid-task.

Question 19

Can I switch models mid-task?

Accepted Answer

Yes, the model can be changed at any moment from the dropdown in the chat, and the dialogue context carries over — you don't have to restate the task.
One detail about money: switching resets the cache, so the first request to the new model costs more than the ones that follow — reading the entire accumulated history is paid for again. A spike in spending at that moment is normal.
The practical conclusion is the same as for coins: start with a smaller model and switch up when you see the task is beyond it. The reverse order — the most expensive model from the start — is almost always overpaying.

Question 20

Paid services: generating a data processor or a report

Accepted Answer

Beyond assistant requests, the console offers services that assemble a finished artifact from your query — an external data processor or an external report. They are billed per delivery rather than by text volume; prices are in pricing.
You are charged only for a successful build: if generation fails, no coins are deducted. Regenerating with the same request within 24 hours is free.

Question 21

Where do I write if something is not working?

Accepted Answer

Write to support@digitalmechanics.dev. Requests are accepted both by email and through the feedback form on the site — all of them are taken into work. A reply goes to those who wrote by email, within 5 business days.

Question 22

What happens when I run out of coins?

Accepted Answer

Coins are needed only for the AI assistant and paid services. Running queries, the query builder, drill-down, background mode and export to Excel are standalone tools and do not consume coins.
When you need the assistant again, buy a pack in your account, whether or not you have purchased the full Console version. Your key and settings don't change — the balance is simply topped up.
Paid coins do not expire on their own. After one year with no account activity, we may write them off — until then, your balance simply waits.

Question 23

I use a different query console — is migrating easy?

Accepted Answer

Yes. LenSA saves and loads individual queries in the .q1c format, compatible with other popular consoles, and stores the whole query list in its own .q1cs format.
The familiar tools are all there — the query builder, parameters, temporary tables, drill-down — plus the AI assistant on top. There's nothing to relearn.

Question 24

Can LenSA change data in my database?

Accepted Answer

The Console itself is read-only: it builds and runs queries but does not write changes to the database.
Changes are possible only in a generated data processor that you explicitly asked LenSA to build and then run yourself. Its actions are named directly — write, change fields, mark for deletion, post or unpost — and you select the affected rows. Immediate deletion is placed behind a separate red button.

Question 25

Is my data sent to the neural network?

Accepted Answer

Three levels, so you never have to guess.
Always sent: the query text, plus the names and types of metadata objects — as needed, in the parts the task requires, and never the whole configuration tree. That is what the assistant runs on: without the object names from your database it would build queries from memory of how typical configurations are arranged.
Sent on separate confirmation: query parameter values and a data sample for diagnostics — up to ten result rows. Before it is sent you see the exact query text that will run, and you can decline; declining does not break the conversation. Reference values travel as technical tokens rather than the data itself, and other values are marked redacted. The free Demo edition has none of this.
Without your confirmation not a single result row leaves your machine: the assistant sees only the fact of execution and the query metrics.
If your database holds personal data. The levels above apply whatever the data is: the names of metadata objects describe the configuration's structure, not the contents of your tables. The models Lensa calls are foreign ones, and we do not hide that — so the decision whether to send a sample of rows for diagnostics stays yours every time, and the free Demo edition does not send rows at all.
Details of data processing are in the privacy policy.

Question 26

How does the console update?

Accepted Answer

The data processor checks for updates on its own and notifies you about a new version. Download the fresh .epf directly from the data processor with its update button, or from your account, then open the downloaded file instead of the previous one. Settings and saved queries are preserved, and you don't need to re-enter the key.
What changed in each version is listed in the "What's new" section on the product page.

Frequently asked questions

Buying and installing