eLLMsha v2026.10.10 Patch 3 — GPT-6 & managed runtimes

Highlights

• Codex CLI 0.162.1 and live model discovery, including GPT-6. Actual GPT-6 Astra inference and project-memory access verified in the installed macOS app.
• Managed / System / Custom CLI selection with resolved path/version, independent managed updates, active-session pinning and rollback when a previous managed version exists.

Fixes

• Fixed stale model catalogs rejecting available models before live validation; added bounded catalog pagination.
• Moved harness-managed Codex skills outside user repositories while preserving discovery and isolation.
• Fixed permission-profile setup and resumed Tencent memory-tool compatibility.
• Verified network access in Auto and Read Only modes while retaining filesystem restrictions.
• Improved runtime button/loading consistency and Light/Dark contrast.

Download & install

Download eLLMsha-2026.10.10-patch.3-mac-arm64.dmg below and replace eLLMsha in Applications. Existing chats and memory use the persistent eLLMsha Desktop profile.

This build targets macOS Apple Silicon. It is ad-hoc signed, not Developer ID signed or notarized. Windows/Linux binaries are not included.

Verification boundaries

Live GPT-6 Astra, memory reads, resumed network access, denied writes in Read Only mode, runtime source selection and native Light/Dark UI were verified. Cross-version managed rollback has focused test/review coverage but was not exercised live. Claude inference was not tested.

Version policy

Release versions now use the actual release date in Europe/Kiev: 2026.10.10, Patch 3. Historical tags remain unchanged.

Full changelog

DMG SHA-256

4490e139d4c829d39a755d8a8fb6079564ce7add15d7bbf53157c6d2814c9197