TonAI

Maybe I didn't convey the idea completely. A system is incomplete until it can test itself.
For example:
There was once AlphaGo, which dominated its opponents, but learned on human data.
Then, over time, there was AlphaZero, which learned exclusively by playing against itself.
The main goal is to allow the agent to learn by playing with itself, solving problems using only the tools at its disposal.
It must be able to write tests that will allow it to self-optimize.

TonAI pinned «Current Task List (it will be constantly updated). 1. Blueprint 1.1 Access blueprint from CLI Description: Provide full access to blueprint from the command line with a single command. Related: (issue#179) 1.1.1 Suggest available contracts or scripts…»

Probably everyone knows that AI is quite bad at coding tasks compared to humans.
Why is that?
Why do people write code better?
What do humans have that AI doesn't?

These questions can be answered.
One of the main problems is that AI doesn’t write code — it writes a poem with syntax highlighting.
The second problem is that AI doesn’t do just one iteration of code generation.
It doesn’t try to look for existing solutions, or think before writing.

Most of these problems have already been solved.
We now have new AI programmers like Cursor.
For some reason, not everyone uses them.

But okay.
Can Cursor solve all problems with code?
No, it can’t.
Cursor’s environment is pretty good.
It can run terminal commands, see linter errors, invoke tools like search_web, search_codebase, file_edit, and other file tools.
Cursor can also enter into an agent chain.

All of this is super cool and levels up AI significantly — but still not enough.

Right now, if we want to bring AI to human-level performance in product-level coding tasks, we need to give AI more environment.

We need a Senior developer in this area, someone who knows the nuances and can fill in the AI’s gaps.

Current issues that I’ve identified so far:
https://t.me/ton_ai_core/7

(This list will be expanded)
If you want to accelerate AI’s adoption in smart contract development on TON — I’m waiting for your contributions.

Current Task List (it will be constantly updated).

1. Blueprint

1.1 Access blueprint from CLI
Description: Provide full access to blueprint from the command line with a single command.
Related: (issue#179)
1.1.1 Suggest available contracts or scripts when name is invalid (issue#3)

1.2 Instant transaction validation via Tonapi
Description: Implement a mechanism in blueprint for instant transaction validation via Tonapi to immediately detect deployment errors. This will help avoid unnecessary AI validation requests, since not all errors are caught at the deployment stage. (issue#8)

1.3 Execution order: build tests before running any script
Description: Change execution order in blueprint: first run test builds, and only then execute any script. Script execution must not occur before tests have successfully run. (issue#6) (need delete)

1.4 Connect eslint and eslint-plugin-suggest-members (issue#26)

1.5 Add Misti (issue#3)

1.6 It is not possible to use the "npx blueprint create" command until the "npm create ton-ai@latest" command has been written. (issue#4)

1.7 Make blueprint misti non-interactive (issue#1)

1.8 Support pre/post hooks by first argument (issue#12)

2. Prompts
Right now I’m trying to build a prompt system where AI always fact-checks (searches the internet by activating Deep Research mode, can clone repositories to search, and looks into the codebase).
Project link – https://github.com/ton-ai-core/contract-knowlenge
I also plan to make a hacker who will try to hack all your contracts. There are already some developments for this.
https://github.com/ton-ai-core/contract-knowlenge/blob/main/.cursor/rules/ton-security-testing.mdc

3. Static Analyzers

3.1 TypeScript
Description: Develop an updated analyzer for TypeScript that detects attempts to call non-existent methods. The logic should mirror the one described for Tact (issue#1697).

3.2 Smart Contracts
Description: Build analyzers for smart contract languages that:
– Detect unused methods and variables.
– Catch empty code blocks (including cases where a comment is considered an empty block).

Goal: Strengthen code quality control so AI can “punish” any code misbehavior and promote better development practices.

4. Tact
Task: Complete the analyzer-related task for Tact described in (issue#1697).
tact-language-server:
issue#582
issue#438
issue#3408

5. TON API Indexing
Description: Fully index the TON API for further AI training and to improve its performance in working with smart contracts.

6. CLI Tools with connect old chats
6.1 Two active Cursor window crash Cursor

Channel name was changed to «TonAI»

Created MCP server for interaction of the AI ​​agent with the TON API
This is the first version
Here is the command for installation in Cursor MCP

Github - https://github.com/ton-ai-core/ton-api-mcp
NPM - https://www.npmjs.com/package/ton-api-mcp?activeTab=readme

npx ton-api-mcp --api-key AFPJTKEBPOX3AIYAAAAKA2HWOTRNJP5MUCV5DMDCZAAOCPSAYEYS3CILNQVLF2HWKED6USY --modules blockchain

Channel created