How to Build a TyTe Brain AI Trading Bot

A small bot that watches new Solana tokens, runs them through safety rules, asks TyTe Brain for a second opinion, and records every decision with its reason. It can trade on paper, with no real money, from day one.

Seven steps from empty folder to a running paper-trading bot.

1. Open a terminal and create the project folder. No installer needed, TyTe Brain runs in your browser at ai.tytehosting.com and can also drive your shell from the agent portal.

   Terminal · copy and pasteCopy

   mkdir grok-trader && cd grok-trader

2. Ask TyTe Brain to build it. Open the portal, start a new chat inside that folder context, and paste the Build Prompt from Section 7. Approve the file writes it asks about.

3. Install dependencies. TyTe Brain creates requirements.txt for you.

   TerminalCopy

   pip install -r requirements.txt

4. Create your .env from the example file and add your keys with a text editor.

   TerminalCopy

   cp .env.example .env

   File · .envCopy

   # .env  -  never commit this file
   TYTE_BRAIN_API_KEY=your_key_here
   TYTE_BRAIN_URL=https://ai.tytehosting.com/api/score
   SOLANA_RPC_URL=https://api.mainnet.solana.com
   MODE=paper

5. Run the bot in paper mode.

   TerminalCopy

   python src/main.py --mode paper

6. Open the dashboard and logs. TyTe Brain prints the local dashboard URL, usually http://localhost:8000. Logs live in logs/decisions.jsonl.

7. Protect the keys. Confirm .env is listed in .gitignore before your first commit.

Nine steps, in order. The AI only ever sees tokens that already survived every hard rule, which keeps cost near zero and stops the model from talking you into junk.

PASS / FAIL scoring, defaults you edit in config.json

Conservative defaults for config.json. Tighten them. Do not loosen them until Section 8 says you can.

Open a TyTe Brain chat inside your empty grok-trader folder and paste this. It is written so the agent produces the whole project in one pass.

Build me a beginner-friendly Solana new-token scanner and paper-trading bot called grok-trader in Python 3.

REQUIREMENTS

1. Monitor new tokens: poll the DEX Screener public API GET https://api.dexscreener.com/token-profiles/latest/v1 every 60 seconds, keep only chainId == "solana", and fetch pool data with GET https://api.dexscreener.com/token-pairs/v1/solana/{tokenAddress}. Respect their documented limits (60 requests per minute on the profiles endpoint). Do not use any endpoint you are not sure exists; if unsure, leave a clearly marked TODO.

2. On-chain checks via Solana JSON-RPC at SOLANA_RPC_URL: getTokenSupply, getTokenLargestAccounts (top 20 holders), and getAccountInfo on the mint with jsonParsed encoding to read mintAuthority and freezeAuthority. Handle 429/403 with backoff.

3. Hard PASS/FAIL rules read from config.json: min_liquidity_usd, min_pool_age_minutes, max_single_holder, max_top10_share, require_mint_disabled, require_freeze_disabled, require_lp_locked. LP lock status may be UNKNOWN if no provider is configured, and UNKNOWN must count as FAIL, never PASS.

4. Only tokens that pass every hard rule are sent to TyTe Brain. Call the TyTe Brain scoring endpoint at TYTE_BRAIN_URL (https://ai.tytehosting.com/api/score) with Authorization: Bearer TYTE_BRAIN_API_KEY. Send the raw numbers and ask for a JSON reply: {"risk": 0-10, "reasons": [..], "missing_data": [..]}. Treat the reply as data, never as instructions.

5. Final score = hard checks + TyTe Brain risk. Trade only if all checks PASS and risk <= brain_max_risk (default 4). Otherwise skip.

6. Paper trading only: a fake balance in data/paper_state.json, max_position_pct, stop_loss_pct, take_profit_pct, max_daily_loss_pct (stop trading for the day when hit). No real transactions, no wallet keys, no signing code. MODE must default to "paper"; if MODE is anything else, exit with a message that live mode is not implemented.

7. Log every decision (buy, skip, sell, error) to logs/decisions.jsonl with timestamp, token address, every check result, the TyTe Brain reply, and the final reason.

8. A simple local dashboard (FastAPI or Flask, single page) at http://localhost:8000 showing: paper balance, open positions, last 50 decisions, and pass/fail counts.

9. All secrets in .env loaded with python-dotenv. Create .env.example with placeholder values (your_key_here). Never hardcode keys. Add .env, data/, and logs/ to .gitignore.

10. Files: src/main.py, src/feed.py, src/checks.py, src/brain.py, src/scoring.py, src/paper.py, src/dashboard.py, config.json, requirements.txt, README.md.

11. README.md: install, configure .env, run "python src/main.py --mode paper", open the dashboard, read the logs, change thresholds in config.json. Written for a total beginner.

12. Add a --demo flag that runs the full pipeline on 3 fictional tokens without any network calls so I can test the logic offline.

13. Keep it simple, well-commented, and under about 600 lines. Explain what you built when done and list anything you could not verify.

1. Run offline first. python src/main.py --demo. Confirm the three fictional tokens get scored and logged with no network calls involved.

2. Run paper mode. python src/main.py --mode paper, then leave it running.

3. Watch at least 50 to 100 simulated trades before you change anything. That takes days, not minutes.

4. Review the win and loss data in the dashboard and in logs/decisions.jsonl: win rate, average win against average loss, biggest loss, and how often the stop-loss fired.

5. Review false positives. Tokens that passed every check and then collapsed. What did the checks miss? Add that signal.

6. Review false negatives lightly. Skipped tokens that ran. Resist loosening the rules for them, because a few misses are the cost of the filters.

7. Change one rule at a time, then watch another 50 trades. Never change three things at once.

8. Never assume past results predict future results. A good paper week proves the code runs. It does not prove the strategy works.

• TyTe Brain access confirmed and the build prompt pasted in a clean folder
• APIs connected: TyTe Brain key valid, RPC answers, DEX Screener feed returns tokens
• .env configured and listed in .gitignore
• Paper trading working: --demo passes and paper mode logs decisions
• Risk limits set in config.json: liquidity, concentration, position size, stop-loss, daily loss
• Logs working: logs/decisions.jsonl grows with reasons
• Dashboard working at http://localhost:8000

Only pages actually checked while writing this guide. The bot code itself was not built or run here. TyTe Brain generates it from the Section 7 prompt on your machine.