SeymourBETA

Aggiornamento 2.230

13/08/2026

v2.230.0.0BETAMINOR

Seymour13 ago 2026

verboo_router.py — Ollama provider: commands ollama (chat+vision via -i) and ollamamodels. Registered models: qwen3-vl:8b (main vision slave — 12/12 on the local battery, ~4s/question on RTX 4090) and gemma4:26b (reserve, 11/12). Auto power-on formula: _ollama_ensure() detects the server (loopback:11434) and starts ollama serve in background if offline.
Validated local vision battery: 12 objective questions on 2 real FFX Mod Studio screenshots. qwen3-vl:8b = 12/12 (49.2s); gemma4:26b = 11/12 (80.4s).
llama3.2-vision dropped on Windows (evidence): mllama architecture unsupported by the Ollama 0.32.9 runner — real error in server.log ("unknown model architecture"). Await Ollama >= 0.4.0.
AGENTS.md — vision fallback rule: (1) local Ollama first; (2) gpt-5.6-luna on failure/bad result; (3) Verboo as backup. Structured QA stays on qwen3.6-27b. ### PATCH

v2.229.0.0

v2.230.1.0BETAPATCH

Seymour13 ago 2026

scripts/windows_clocks_deepseek.ps1 (new): -Install writes the 2 extra taskbar clocks via HKCU (Clock1=UTC (DeepSeek), Clock2=China (Beijing); local clock stays); -Status shows current zone + countdown to the next transition + prices; -Watch = live panel with alert when ≤30min remain before ENTERING peak; -Uninstall; -RestartExplorer. Installed and validated on the user's machine (NumClocks=2, registry read-back OK, full uninstall/reinstall cycle OK).
scripts/tests_windows_clocks_deepseek.ps1 (new): 18/18 PASS (window edges, Chinese lunch gap, day rollover).
Data source (checked 2026-08-13): DeepSeek pricing switches on 2026-08-16 16:00 UTC to peak/off-peak billing (off-peak = half of peak); PEAK windows = 01:00-04:00 and 06:00-10:00 UTC (= 09:00-12:00 and 14:00-18:00 Beijing; 22:00-01:00 and 03:00-07:00 Brasília). Current models: deepseek-v4-flash/deepseek-v4-pro.
2 bugs fixed during validation: DateTime.Parse('...Z') becoming Kind=Local (showed 13:00 instead of 16:00 UTC) and UTF-8 without BOM (PS 5.1 read it as ANSI — corrupted accents).

v2.230.0.0

SeymourBETA

13/08/2026

v2.230.0.0BETAMINOR

Seymour13 ago 2026

verboo_router.py — Ollama provider: commands ollama (chat+vision via -i) and ollamamodels. Registered models: qwen3-vl:8b (main vision slave — 12/12 on the local battery, ~4s/question on RTX 4090) and gemma4:26b (reserve, 11/12). Auto power-on formula: _ollama_ensure() detects the server (loopback:11434) and starts ollama serve in background if offline.
Validated local vision battery: 12 objective questions on 2 real FFX Mod Studio screenshots. qwen3-vl:8b = 12/12 (49.2s); gemma4:26b = 11/12 (80.4s).
llama3.2-vision dropped on Windows (evidence): mllama architecture unsupported by the Ollama 0.32.9 runner — real error in server.log ("unknown model architecture"). Await Ollama >= 0.4.0.
AGENTS.md — vision fallback rule: (1) local Ollama first; (2) gpt-5.6-luna on failure/bad result; (3) Verboo as backup. Structured QA stays on qwen3.6-27b. ### PATCH

v2.229.0.0

v2.230.1.0BETAPATCH

Seymour13 ago 2026

scripts/windows_clocks_deepseek.ps1 (new): -Install writes the 2 extra taskbar clocks via HKCU (Clock1=UTC (DeepSeek), Clock2=China (Beijing); local clock stays); -Status shows current zone + countdown to the next transition + prices; -Watch = live panel with alert when ≤30min remain before ENTERING peak; -Uninstall; -RestartExplorer. Installed and validated on the user's machine (NumClocks=2, registry read-back OK, full uninstall/reinstall cycle OK).
scripts/tests_windows_clocks_deepseek.ps1 (new): 18/18 PASS (window edges, Chinese lunch gap, day rollover).
Data source (checked 2026-08-13): DeepSeek pricing switches on 2026-08-16 16:00 UTC to peak/off-peak billing (off-peak = half of peak); PEAK windows = 01:00-04:00 and 06:00-10:00 UTC (= 09:00-12:00 and 14:00-18:00 Beijing; 22:00-01:00 and 03:00-07:00 Brasília). Current models: deepseek-v4-flash/deepseek-v4-pro.
2 bugs fixed during validation: DateTime.Parse('...Z') becoming Kind=Local (showed 13:00 instead of 16:00 UTC) and UTF-8 without BOM (PS 5.1 read it as ANSI — corrupted accents).

v2.230.0.0