TonAI

Всем привет! Участвуем в хакатоне сделали инфраструктуру для легкого создания TMA c помощью AI. И подались в трек Agent Infrastructure (улучшаем вайбкодинг TMA)

Подержите нас клэпами 👏 если вам понравился проект! Спасибо.

Смотреть тут: https://identityhub.app/contests/ai-hackathon?submission=cmn6mzbcf01a901nvcowgfokd

У https://docs.ton.org/ вышло обновление и они добавили ИИ агента для поиска по докам

Сам агент не оч умный, но хорош в качестве поисковика (Показать где что найти)

Решил на базе него создать skill который можно подключить к Cursor, Codex, Claude Code и вообще куда угодно

тем самым можно использовать самые умные модели с всегда актуальной документацией ТОНА

https://github.com/ton-ai-core/docs-ton-skill

Отлично подойдёт для тех кто только разбирается в разработке под ТОН


(Ну и сам скил очень простой и его можно видоизменять под себя)

Я создаю новый канал на русском языке @ProverCoderAI
Буду пробовать развивать концепцию ИИ кодинга глобально
Пытаться описать новую парадигму для разработки с ИИ (что бы ИИ мог автономно работать по 20 часов+)

На ТОН я не забиваю
Под него фичи тоже буду доделывать вермя от времени

I am currently looking for a job.
I work on developing and optimizing systems for vibecoding.
I can integrate and configure existing projects so that Cursor, Claude code, Codex, and other tools work more efficiently.

I know how to develop compilers that simplify and accelerate AI’s work with code.
I can build complete AI development pipelines.

I have won four hackathons.
I have given 7 talks on AI-related topics.
I have commercial development experience in C# and TypeScript.

I am currently interested in functional programming (exploring new paradigms).
I am learning new languages (Rust, Gleam).

My GitHub repositories:
https://github.com/skulidropek
https://github.com/orgs/publicrust/repositories
https://github.com/orgs/ton-ai-core/repositories

Linkedin - https://www.linkedin.com/in/vladimir-chikaidze-ba87212a5/

Telegram - @skulidropek

An example of how AI independently generates a smart contract on TON.
It searches for the necessary information on its own, verifies it, and produces only the desired result.

The smart contract generation continued until 35:00. After that, deployment and verification of the contract’s compliance with the original task began using TonCenter, TonAPI, and TonViewer.

Example - https://www.youtube.com/watch?v=sc7CE2SBnM8

The installation guide is in the readme file https://github.com/ton-ai-core/contract-knowledge

I’ve been invited to give a talk about TonAI at an event in Moscow. I’ll be joining as an online speaker.

More details:
https://t.me/tonntalks/123
https://t.me/tonntalks/129

Lang: RU

The day before yesterday I gave a presentation of the TonAI project to the Georgian startup community.

I was given feedback on the project. They recommended how best to promote the project in their opinion. They said that it is better to immediately start communicating with final clients and solving their problems.

I hope that I will start using many of the tips that I was given

All the tips can be viewed in Part2

The presentation itself is in Part1

Part1 - https://www.youtube.com/watch?v=yE_0Ejohwy0
Part2 - https://www.youtube.com/watch?v=HjQUC3_MnZk

Georgia startup chat - https://t.me/unicornembassy_georgia

Thanks to @homo_habitus1 for hosting this event.

On the 21st of this month I’ll be giving a presentation in Tbilisi about TonAI - how I built it and what I plan to implement before the MVP.
The talk will be in Russian.

https://t.me/unicornembassy_georgia/1082

Parsers of all chats on TON (development of smart contracts) and a TON resear site were also implemented

https://github.com/ton-ai-core/tonresear-parser/tree/main/data/markdown
https://github.com/ton-ai-core/ton-chats/tree/main/out/markdown

Wrote an automatic indexing TON API.
It will go from time to time and parse all contracts from the blockchain. This is needed as a training dataset for AI

Put STARs on the repository so that this repository gets into the training dataset of basic AI models

https://github.com/ton-ai-core/ton-api-indexing/tree/main/data

As for the contest, I couldn't get the prize. I couldn't solve the second problem for about 8 hours, I always failed the last test. I generated about 60 of my own tests to find holes in my solution, but it still didn't work. Then someone gave me a hint that I should have pulled the 2025 check from recive to init and only then I was able to pass the second problem. With other problems, the problem was basically the same. I couldn't complete the last tests and stopped solving the contest.
Now I'm fixing the agent so that it can solve problems better

Information on Current Plans:

I've successfully completed many tasks I previously set. Currently, I've fully redesigned the Blueprint to better suit my specific needs.

The Blueprint has been made non-interactive, and additional configurations with event subscriptions have been introduced (e.g., prehook).

"scripts": {
  "bp": "blueprint",
  "prebuild": "npx @ton-ai-core/blueprint-linter",
  "postbuild": "blueprint misti",
  "build": "blueprint build",
  "prerun:deploy\\w+": "blueprint generate-specs && blueprint test",
  "prerun:\\w+": "blueprint generate-specs && blueprint test ./tests/$BLUEPRINT_SCRIPT_NAME.spec.ts",
  "start": "blueprint run",
  "pretest": "npx blueprint build --all && eslint . --ext .ts",
  "test": "jest --verbose",
  "lint": "eslint . --ext .ts",
  "misti": "blueprint misti"
}

I've created a simple linter for Blueprint (https://github.com/ton-ai-core/blueprint-linter) which checks the correctness of contract file names and directory structures.

Additionally, I've implemented automated test generation for scripts in the script folder (https://github.com/ton-ai-core/blueprint-sandbox). This generates nearly functional tests automatically for scripts to ensure each script is tested locally before deployment to the blockchain, significantly reducing issues when working with AI.

Work continues on the https://github.com/ton-ai-core/contract-knowlenge project. Although my customized version with additional insights isn't ready yet, I plan to release it shortly.

Upcoming Plans:

1. Develop a Telegram user bot to run on GitHub Actions (scheduled every 6 hours). This bot will parse Telegram channels related to TON (e.g., https://t.me/tactlang, https://t.me/tondev_eng). Data collected from these channels will serve as a dataset for AI in the contract-knowledge project.

2. Ensure the prompt system updates automatically when any command is executed. There will be a dedicated .mdc file continuously updated via a special command, which will generate the subproject structure for contract-knowledge. For example:

### SbtProject `projects/SbtProject`
- **contracts/** (1 file)
  - `contracts/Proposal.tact`
- **script/** (1 file)
  - `script/deployProposal.ts`
- **wrappers/** (2 files)
  - `wrappers/Proposal.ts`
  - `wrappers/Proposal.compile.ts`
- **tests/** (2 files)
  - `tests/proposal.spec.ts`
  - `tests/public1.spec.ts`

### TonVoting `projects/TonVoting`

This approach will help AI recognize existing working projects. I'll also experiment with chat functionality to allow AI to reference recent conversation history, further enhancing context understanding.

3. Introduce a new Blueprint plugin focused on contract formatting. For Tact contracts, for instance, the command npx tact-fmt -w ./contracts will be utilized.

4. Refactor and reorganize all documentation stored at https://github.com/ton-ai-core/contract-knowlenge/tree/main/.knowlenge. The recent update in Cursor allows the agent to view all project folders. To leverage this effectively, a thorough cleanup and documentation restructuring is essential.

5. Continue refining prompts for .mdc and .cursorrules files.

A little later I will report on the work done on the template.

But for now I will tell you about a new tool that I developed specifically for Cursor. I call it "unlimite-context". What is its essence? You upload all the found chats/changes of the agent's AI to the project and when the context ends, the agent can go to them and restore the lost information.

Command to generate all available composers from this project in the file system.

There is only one drawback so far, that I can not accurately receive chats from the project because of this I search by the project path, so if this project was mentioned somewhere in other projects, then I will parse the composers of other projects.

github - https://github.com/ton-ai-core/unlimite-context
npm - https://www.npmjs.com/package/@ton-ai-core/unlimite-context

Example:

npx @ton-ai-core/unlimite-context /home/user/contract-knowlenge

Not long ago I participated in TON Contests from DivaTech (https://t.me/diva_tech/20)

I was the first to solve all 5 problems.
Now the results have been announced
https://www.toncontests.com/contests/1?tab=leaderboard

I took 18th place in the TOP

Of course, everything was solved with the help of AI and I did not write a single line of code. But progress is already visible that AI can take prize places

The next step is https://t.me/toncontests/194
I will have to enter the TOP 128 to get any prize place

I will also soon publish news on the work done on AI tools

Approvals take a very long time to get to Blueprint. I think it makes sense to start running your own Blueprint branch, which should be tailored to AI.

New command for creating templates.

npm create ton-ai@latest

Soon I will add to the template so that eslint is included by default and https://t.me/ton_ai_core/11

Blueprint2 repo - https://github.com/ton-ai-core/blueprint2

I am waiting for my MR approval and after that the task will be fully completed.
https://github.com/ton-org/blueprint/pull/186

Made eslint analyzer for TypeScript. Works with any non-existent types.

Github - https://github.com/ton-ai-core/eslint-plugin-suggest-members
NPM - https://www.npmjs.com/package/@ton-ai-core/eslint-plugin-suggest-members