Every version of the app
Grab the newest build below, or go back to an older one if you need to. The app also updates itself, so most of the time you can ignore this page entirely.
Version
Released
What changed
31 Jul 2026
Tier 2 (Advanced Core) now runs on a llama.cpp/GGUF backend (Qwen2.5-3B-Instruct, Q4_K_M) instead of full-precision transformers, cutting VRAM requirement from …
30 Jul 2026
Fix warning-leak bug: transformers/torch UserWarning text (do_sample/temperature/top_p/top_k no-op warnings) was leaking into the visible Private AI chat reply …
30 Jul 2026
Fix LLM daemon boot crash (STATUS_ACCESS_VIOLATION / rc=3221225477). Root cause: torch resolves msvcp140.dll via ctypes.CDLL name-based lookup which could resol…
30 Jul 2026
Fix STATUS_ACCESS_VIOLATION (0xC0000005) crash in LLM daemon boot caused by PyInstaller _MEIPASS PATH leak into portable python child processes. Rebuilds PATH f…
30 Jul 2026
Fix stale desktop UI model names/VRAM thresholds (TinyLlama/Phi-3 -> Qwen2.5-0.5B/1.5B) to match the v2.81.1 backend precision fix and updated tnsor_models.json…
30 Jul 2026
Fix bf16 precision bug on RTX 3050-class 6GB GPUs: fp32 for tier_1/tier_2 LLM inference, TinyLlama->Qwen2.5-0.5B (tier_1), Phi-3-mini->Qwen2.5-1.5B (tier_2, gat…
30 Jul 2026
Drop bitsandbytes quantization (fixes LLM gibberish/crash on Ampere GPUs); add bounded retry with backoff for transient image-gen failures.
30 Jul 2026
Fix: SD-Turbo pure-noise/NaN bug in image generation. Upcast UNet + VAE to fp32 at their forward-call boundaries (rest of pipeline stays fp16) - fixes intermitt…
29 Jul 2026
Fix: replace --ignore-installed (v2.80.7) with proper stale dist-info cleanup before pip install. v2.80.7 fixed the uninstall-no-record-file abort but caused du…
29 Jul 2026
Fix: pip install --ignore-installed to survive corrupted RECORD-less installs left by prior crashed launches (v2.80.6 DLL fix root-caused this).
29 Jul 2026
Fix PyInstaller DLL bundling bug that caused the host_engine.exe sidecar to crash silently before emitting any output (root cause of the persistent "Installing …
28 Jul 2026
Reapply the v2.80.1 SD-Turbo/compel embedding fix on top of the v2.80.4 clean baseline. Verified end-to-end: host_engine.exe launches correctly (v2.80.4 baselin…
28 Jul 2026
Rollback: full revert of host_engine.py to the last known-working v2.80.0 baseline. The install pipeline (host_engine.exe process) was silently failing to launc…
28 Jul 2026
Fix a destructive module-reinstall bug: a transient portable-Python health-check failure previously deleted the entire AI deps environment (including already-in…
28 Jul 2026
Fix: SD-Turbo image generation was routing every prompt through compel with the wrong embedding layer (SD1.x default vs SD-Turbo's SD2.1 architecture), producin…
20 Jun 2026
v2.79.0 - Blender pipeline fixes: relay handler, signed URL uploads, RLS bypass
19 Jun 2026
v2.78.1 - Blender Bootstrapper 403 Fix - **Spoof browser User-Agent** in `blender_bootstrapper.py` download request to bypass `download.blender.org` 403 Forbid…
19 Jun 2026
v2.78.0 - Headless Parameterized Blender Render Engine New Features - **Blender Bootstrapper**: Auto-downloads and caches portable Blender 4.4.3 when host acti…
19 Jun 2026
v2.77.0 - JWT Expiry Resilience - **Upload fallback**: When JWT returns 401/403, automatically retry with anon key so generated images are not lost - **Token r…
19 Jun 2026
Root Cause Module-level `_load_model_config()` printed config logs to stdout before `check_modules` could emit its JSON result. The Tauri Rust handler returned …
19 Jun 2026
Fixes - **VRAM Gatekeeper fallback**: When nvidia-smi is unavailable, engine now probes via portable python + torch before returning 0 GB - **Frontend race cond…
31 May 2026
v2.75.0 — Rollback: Restore Clean v2.73.0 LLM-Only Codebase Reverts v2.74.0 (Blender engine) completely. All Blender-related code, workers, API routes, and UI …
30 May 2026
TNSOR Agent v2.74.0 — Distributed Blender Render Engine New: Full-Stack Blender 3D Rendering Pipeline - **Automated Bootstrapper**: Portable Blender 4.1.1 down…
30 May 2026
v2.73.0 — Decoupled Hardware Registry (Dynamic Runtime Hydration) All hardcoded HuggingFace model strings removed from the engine and frontend. Phase 1 — `tns…
30 May 2026
v2.72.0 — Host UI: Private LLM Card Sync with Image Gen What changed The Private LLM compute module card on the Host Dashboard now mirrors the Image Gen card e…
30 May 2026
v2.71.0 — Dynamic LLM Routing, Chat Templates, Model Selector UI Phase 1 — Worker: Dynamic Model Loading + Chat Templates - Worker accepts model alias via `sys…
30 May 2026
v2.70.0 — Persistent LLM Daemon (Warm REPL) Eliminates the 8-second cold-start per message. Model loads once; subsequent prompts dispatched via stdin pipe. Wo…
30 May 2026
v2.69.0 — UTF-8 Subprocess Encoding Patch **Fixes:** UnicodeEncodeError: charmap codec crash when LLM generates non-ASCII tokens on Windows. Root cause Window…
30 May 2026
v2.68.0 — Private LLM Streaming Pipeline Phase 1 — OS-Level I/O Hardening - stdin=DEVNULL, stderr=STDOUT, CREATE_NO_WINDOW on all LLM subprocess calls - TOKENI…
Every build is cryptographically signed, and the app refuses to install an update that doesn't match the signature.