Gentleman-Programming/gentle-ai v1.45.0

on GitHub

Highlights

This release brings GPT-5.6 model profiles to Codex, strengthens the contract between memory and user-facing replies, and makes install/upgrade behavior safer across Codex, Homebrew, themes, and persona configuration.

GPT-5.6 Codex profiles (#1067)

• Added gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna to the Codex model catalog.
• Updated the Low Cost, Recommended, and Powerful presets with deliberate model and reasoning-effort defaults for strong, mid, and cheap SDD lanes.
• Added the same models to the custom model picker for per-phase assignment.
• Codex >=0.144.0 is now required before Gentle AI writes GPT-5.6 profiles. Install and sync fail before profile mutation when the runtime is missing, too old, or cannot be parsed, with an actionable upgrade command instead of silently writing unsupported configuration.

Memory must never replace the reply (#1075)

• Strengthened the Engram delivery contract across installed agent outputs: saving memory is bookkeeping, not a substitute for answering the user.
• Agents are explicitly required to compose and send the user-facing reply after memory capture, including degraded-memory and partial-failure paths.
• Added adversarial contract tests so future wording changes cannot accidentally invert this behavior.

Safer installs and upgrades

• Homebrew upgrades now detect whether the active gentle-ai executable is owned by a formula or cask and use the matching upgrade strategy.
• Dual installations, shadowed binaries, ownership uncertainty, stale results, and legacy cask migration now fail closed with actionable guidance instead of falling through to a competing installer.
• Full installs no longer apply the unsafe generic theme component while preserving supported Claude theme and OpenCode Gentle Logo behavior.
• Persona visual opt-out is now honored by the TUI (#1060).
• The generated OpenSpec configuration template is valid YAML again (#1008).

Codex requirement

GPT-5.6 profiles require Codex >=0.144.0.

npm install -g --ignore-scripts @openai/codex@0.144.0

There is no legacy-model fallback: upgrade Codex before running an install or sync that targets Codex.

Validation

• Full Go test suite passed locally.
• Unit Tests and E2E passed on Ubuntu, Arch, and Fedora in CI run 29056296790.
• GoReleaser completed successfully in release run 29056481022.
• Release archives and checksums were published for macOS, Linux, and Windows.
• Homebrew Formula and Scoop manifests were updated to 1.45.0.

Upgrade

brew upgrade gentle-ai