Bubat AI 0.9.2

Beta. Third test release of Bubat AI for Windows, for beta testers. Expect rough edges, and please report problems to support@eabudakubat.com (Settings → Logs helps us see what happened).

📖 New to Bubat AI? Read the user guide first: https://eabudakubat.com/bubat-ai

Install: download BubatAI-Setup-0.9.2.exe below and run it (no administrator rights, Python or git needed). On first start, Bubat AI checks your PC, installs Ollama if needed and downloads AI models that fit it (about 7 to 22 GB). You need MetaTrader 5 with a demo account, logged in.

Already on 0.9.0? Run the new setup over it. Your sign-in, license, settings and history are kept, and the settings new since 0.9.0 are added to yours. The 0.9.0 app can't see beta releases by itself, so this one has to be installed by hand once; from 0.9.1 on, beta updates show up in Settings → Updates.

On 0.9.1? Settings → Updates offers 0.9.2.

Free 7-day trial: sign in with Google in the app. During the trial, the bot's trade results and its BUY/SELL decisions (pair, direction, time, confidence; never your account number, balance or money amounts) are shared to improve Bubat AI for everyone (privacy policy). For a license after the trial, email support@eabudakubat.com.

Please read: trading is risky and this bot has lost money in testing; the beta is for testing, not for income. It trades demo accounts only unless you change a setting by hand. Not financial advice.

What's new since 0.9.0

• Stop means stop (new in 0.9.2): after you press Stop, the trading agent stays stopped until you press Start. In 0.9.1 the automatic restart brought it back within seconds.
• The AI decides once an hour instead of every 5 minutes. Open trades are still checked every 15 seconds (break-even, trailing stop).
• Strategy Lab strategies can trade. A rule-based strategy that passes the Lab's tests can trade on your demo at 0.01 lot once you approve it in the Brain tab. To pass, it must beat simply holding the same direction ("market drift") on the last part of 5 years of history, which it was not tuned on, after your real spreads, swaps and the wider spreads at the daily rollover. Lab trades are tagged "Bubat Lab", keep every account safety wall, may use at most 7 of the 10 trade slots and 6% of the balance at stop loss, and drop back to paper trading by themselves if their last 20 live trades average below -0.2R.
• AI vs Strategy Lab card on the Overview: both scored the same way on your account over the same period.
• Edge study: a test of whether the trading AI does better than just following the 1-hour trend (card on the Brain page, log under Logs → Edge study).
• Daily brain review: each proposal now carries a replay computed in code from your results, proposals the replay no longer supports withdraw themselves, runs finish before the next trading hour, and one Ollama crash no longer ends the night's review.
• Faster, steadier AI calls: the trading model stays loaded on the graphics card between hourly decisions, and is reloaded if part of it ended up on the processor. It holds about 5 GB of graphics memory while the market is open; if games stutter, set "ollama_keep_alive": null in config.json.
• Recovery: the trading agent restarts by itself if it stops unexpectedly (Settings → System automation & recovery). Ollama's own log is kept in logs\ollama_server.log.
• Updates: beta (pre-release) versions now reach installed copies.

Known limits in this beta

• Results so far are negative. On our demo, the AI's 71 trades from 3 to 16 October averaged -0.14R per trade (R = the amount at risk). A study over a year of history found the AI has no edge over following the 1-hour trend, so trading is moving to tested Strategy Lab rules step by step. The first one went live on our demo on 11 October; it is too early to judge it.
• Weekly weight training of the AI model is not set up on tester PCs (it needs a large download and a strong GPU).
• PCs with less than 16 GB of memory run without the daily brain review and the Strategy Lab's web research.

"Windows protected your PC": the installer is not code-signed yet. Click More info, then Run anyway. Compare the file's SHA-256 with the .sha256 file below to verify it.

Bubat AI 0.9.1

Beta. Second test release of Bubat AI for Windows, for beta testers. Expect rough edges, and please report problems to support@eabudakubat.com (Settings → Logs helps us see what happened).

📖 New to Bubat AI? Read the user guide first: https://eabudakubat.com/bubat-ai

Install: download BubatAI-Setup-0.9.1.exe below and run it (no administrator rights, Python or git needed). On first start, Bubat AI checks your PC, installs Ollama if needed and downloads AI models that fit it (about 7 to 22 GB). You need MetaTrader 5 with a demo account, logged in.

Already on 0.9.0? Run the new setup over it. Your sign-in, license, settings and history are kept, and the settings new in 0.9.1 are added to yours. The 0.9.0 app can't see beta releases by itself, so this one has to be installed by hand once; from 0.9.1 on, beta updates show up in Settings → Updates.

Free 7-day trial: sign in with Google in the app. During the trial, the bot's trade results and its BUY/SELL decisions (pair, direction, time, confidence; never your account number, balance or money amounts) are shared to improve Bubat AI for everyone (privacy policy). For a license after the trial, email support@eabudakubat.com.

Please read: trading is risky and this bot has lost money in testing; the beta is for testing, not for income. It trades demo accounts only unless you change a setting by hand. Not financial advice.

What's new in 0.9.1

• The AI decides once an hour instead of every 5 minutes. Open trades are still checked every 15 seconds (break-even, trailing stop).
• Strategy Lab strategies can trade. A rule-based strategy that passes the Lab's tests can trade on your demo at 0.01 lot once you approve it in the Brain tab. To pass, it must beat simply holding the same direction ("market drift") on the last part of 5 years of history, which it was not tuned on, after your real spreads, swaps and the wider spreads at the daily rollover. Lab trades are tagged "Bubat Lab", keep every account safety wall, may use at most 7 of the 10 trade slots and 6% of the balance at stop loss, and drop back to paper trading by themselves if their last 20 live trades average below -0.2R.
• AI vs Strategy Lab card on the Overview: both scored the same way on your account over the same period.
• Edge study: a test of whether the trading AI does better than just following the 1-hour trend (card on the Brain page, log under Logs → Edge study).
• Daily brain review: each proposal now carries a replay computed in code from your results, proposals the replay no longer supports withdraw themselves, runs finish before the next trading hour, and one Ollama crash no longer ends the night's review.
• Faster, steadier AI calls: the trading model stays loaded on the graphics card between hourly decisions, and is reloaded if part of it ended up on the processor. It holds about 5 GB of graphics memory while the market is open; if games stutter, set "ollama_keep_alive": null in config.json.
• Recovery: the trading agent restarts by itself if it stops unexpectedly (Settings → System automation & recovery). Ollama's own log is kept in logs\ollama_server.log.
• Updates: beta (pre-release) versions now reach installed copies.

Known limits in this beta

• Results so far are negative. On our demo, the AI's 71 trades from 3 to 16 October averaged -0.14R per trade (R = the amount at risk). A study over a year of history found the AI has no edge over following the 1-hour trend, so trading is moving to tested Strategy Lab rules step by step. The first one went live on our demo on 11 October; it is too early to judge it.
• Weekly weight training of the AI model is not set up on tester PCs (it needs a large download and a strong GPU).
• PCs with less than 16 GB of memory run without the daily brain review and the Strategy Lab's web research.

"Windows protected your PC": the installer is not code-signed yet. Click More info, then Run anyway. Compare the file's SHA-256 with the .sha256 file below to verify it.

Bubat AI 0.9.0

Beta. First test release of Bubat AI for Windows, for beta testers. Expect rough edges, and please report problems to support@eabudakubat.com (Settings → Logs helps us see what happened).

📖 New to Bubat AI? Read the user guide first: https://eabudakubat.com/bubat-ai

Install: download BubatAI-Setup-0.9.0.exe below and run it (no administrator rights, Python or git needed). On first start, Bubat AI checks your PC, installs Ollama if needed and downloads AI models that fit it (about 7 to 22 GB). You need MetaTrader 5 with a demo account, logged in.

Free 7-day trial: sign in with Google in the app. During the trial, the bot's trade results and its BUY/SELL decisions (pair, direction, time, confidence; never your account number, balance or money amounts) are shared to improve Bubat AI for everyone (privacy policy). For a license after the trial, email support@eabudakubat.com.

Please read: trading is risky and this bot has lost money in testing; the beta is for testing, not for income. It trades demo accounts only unless you change a setting by hand. Not financial advice.

What's inside

• Desktop app: overview, positions, health check, AI chat, brain proposals, logs, one-click updates with rollback.
• A local AI picks BUY / SELL / WAIT for up to 29 symbols every 5 minutes; code sets the stop, target and lot, and 17 risk walls (spread, news blackout, H1 trend, daily loss stop, demo-only, license, ...) can veto any trade.
• Forex knowledge: the AI is taught how its trades really work (stop size, duration, costs) and what to look at (daily / 4-hour / hourly trend, stretched entries, levels in the way), and this is measured against the old prompt on random pair-hours.
• Learns from its own results: every signal is scored against what the market really did, after costs; the AI sees its own track record before each decision.
• Shared learning: installs share their AI's decisions and every copy learns from everyone's, re-scored on its own prices (results cannot be faked).
• Strategy Lab: backtests rule-based strategies on years of history with your real spread and paper-trades the survivors; nothing trades live without your approval.
• Daily brain review (16 GB PCs): finds what is going wrong and proposes changes for you to approve.

Known limits in this beta

• Results so far are slightly negative; the learning features are new and still being measured.
• Weekly weight training of the AI model is not set up on tester PCs (it needs a large download and a strong GPU).
• PCs with less than 16 GB of memory run without the daily brain review and the Strategy Lab's web research.

"Windows protected your PC": the installer is not code-signed yet. Click More info, then Run anyway. Compare the file's SHA-256 with the .sha256 file below to verify it.