NativeSoft AI
NativeSoft AI is our own assistance engine, built in-house. It gives people fast, contextual help inside a product — answering from that product's own knowledge and pointing to the exact page, function or step. It runs on ordinary hardware, on the product's own machine, with nothing sent to an outside service.
Already in our products
NativeSoft AI runs today in three live products, each adapted to its own knowledge and language. Turn through them.
A database workbench for every JDBC database — SQL editor, result grids, careful data editing, exports and comparison. Its in-app assistant answers from Ananke's own documentation and links to the exact page.
On-device · English-only NativeHelpy model
Visit Ananke
A smart cookbook that works offline: hundreds of recipes, smart search and step-by-step cooking. Its assistant finds recipes by ingredients and preferences and helps with the app, in the language you read it in.
On-device · Multilingual NativeHelpy (23 languages)
Visit Gustivela
Play backgammon online against real people or practise against the AI opponent, with a learning academy. Its "Cleaner AI" assistant explains the rules and features and guides learning, with a link to the right page.
On-device · Multilingual NativeHelpy · a dedicated gammon AI engine, 177,000+ downloads on Google Play
Visit Clean BackgammonOn-device, powered by NativeHelpy
NativeSoft AI is on-device AI. The optimized NativeHelpy model does the work locally — on the product’s own machine — so questions and answers stay there. It is tested and live today across three websites and two mobile apps.
A 23-language model powers Gustivela and Clean Backgammon — on their sites and in their mobile apps — answering in the language the user reads.
A focused English model powers Ananke’s documentation assistant, where the product and its docs are in English.
Clean Backgammon’s opponent is a specialized on-device gammon AI engine, built on the same NativeHelpy technology and proven in play across 177,000+ downloads on Google Play.
A bigger general model — a sibling to NativeHelpy — is in the works, to widen what the engine can answer while staying true to the same on-device approach.
What it is
The manual, the guides, the catalogue or the help — an answer is a passage of your own material, not a guess from the open web.
Every answer points to the page, function or step it came from, so a person can read on and check it.
On the CPU, on your own server or the device, offline where the implementation allows — no GPU, no outside service, no telemetry.
Where a product is translated, it replies in the language of the page; the question may come in another.
Where it can be adapted
The three products above are live. The rest are directions the same engine fits — what a deployment looks like depends on your knowledge, your tools and your constraints.
Answer from a manual, a wiki or an internal handbook, with a link to the exact passage.
Guide a new user through a product from inside it, in their own language.
A help assistant that ships with the product and works without a network where it can.
Explain rules, mechanics and lessons, as in Clean Backgammon's academy.
Precise, source-linked answers for software like Ananke, where a wrong guess is worse than none.
Help with a task and, where integrated with your tools, carry out actions — bounded by your permissions and approvals. The engine does not act on its own.
Search internal documents, explain procedures and guide staff and customers. Sensitive deployments need specific validation, access control and oversight; NativeSoft AI does not make lending decisions, give personalised financial advice or carry out transactions.
Tell us what you are building and the knowledge it should draw on. We will tell you honestly what fits.
Discuss an integrationor write to [email protected]
