SpiritBox Ultimate v1.12 R7.7.1
The bundled spiritbox-llm engine is primary. LM Studio is optional. Start with Minimal for authenticated chat and saved conversations.
SpiritBox_Ultimate_v1.12_R7.7.1_Windows.zip · 82,666,332 bytes
a1f8a3c9ef539bc7743eb3f6b00a45825cf02370f7a189094bb1cc58e371d1c4.Windows launcher hotfix
R7.7.1 fixes Windows launcher path quoting with six small edits across three batch files. All 56 lightweight launcher and preflight tests passed against the extracted hotfix. Prior R7.6/R7.7 runtime evidence is unchanged; a fresh installation with real Docker and a model was not rerun for this hotfix.
Bounded qualification evidence
- Archive integrity, repeatability and all 565 manifest payload entries were independently verified. The R7.7.1 archive changes only three Windows batch files and their manifest entries from sold R7.7.
- The bundled Qwen3-14B route completed a three-bullet checklist on R7.6. Saving, reopening and app-restart persistence passed.
- The optional LM Studio GPT-OSS20B route completed a short factual summary on R7.7. The identical answer survived reload, reopening and app restart. The preceding optional attempt reached its output limit without a final answer.
- R7.7 fixes persistence of that incomplete-answer notice. An offline replay of the captured empty response verified the notice after reload and reopening. No automatic retry was made.
Evidence boundary
These are narrow checks on a configured host, not qualification of other hardware, every model or long conversations. No new voice generation was performed for this release. Full retrieval/graph, ComfyUI, VoiceLab training and live Telegram/n8n workflows require separate configuration and validation. n8n is user-managed and is not installed by changing the runtime profile.
Setup and rollback
Use Windows with Docker Desktop, WSL2/Linux containers and separately downloaded models suited to your hardware. Extract into a new empty folder and follow QUICK_START.md. For optional LM Studio, follow LMSTUDIO_FIRST.md. Back up configuration and user data before updating. Keep the previous archive with its matching data backup. Do not delete Docker volumes or overwrite model/user-data directories.
Get Pro