AhkLLM v1.4.0

AhkLLM v1.4.0

AhkLLM v1.4.0 adds an optional local Codex CLI backend and hardens chat state across concurrent threads, branches, retries, and cancellations.

Added

• Optional Codex CLI backend using ChatGPT subscription authentication.
• Codex reasoning summaries, native web search, image generation, usage tracking, and automatic thread titles.
• Per-thread Codex request isolation for concurrent chats, thread switches, and branch navigation.
• Mermaid diagram rendering and bracket-delimited LaTeX math support.
• Completion sounds and attention indicators for finished responses.

Improved

• New-chat defaults consistently control the selected assistant, model, and provider icon.
• Markdown rendering and WebView startup were refactored into focused modules.
• Chat links now open in the system's default browser.
• Streaming, cancellation, retry, branch, and thread-switch state handling was substantially hardened.

Fixed

• Duplicate messages and stale UI state during cancellation races.
• Cross-thread response, usage, error, and retry-state leakage.
• Model and settings preservation across edits, forks, retries, and reloads.
• Codex image, reasoning, title-generation, and search state persistence.
• Sidebar ordering, provider badges, trash disclosure state, and new-chat initialization edge cases.

Downloads

• AhkLLM-v1.4.0.zip is the versioned release archive.
• AhkLLM.zip is an identical stable-name alias used by the README download link.

AhkLLM v1.3.1

Backup patch release. Fixed backup settings validation so automatic backups require a destination, normalized selected folders consistently, recovered persisted configuration for manual backups, cleared stale backup errors after configuration changes, and improved pending/running/folder-required status messages.

AhkLLM v1.3.0

Added first-class user-added OpenAI-compatible Chat Completions providers with stable IDs and optional models.dev catalog mappings, plus rich cached-input pricing and compatibility metadata. Improved the resizable stacked Edit Models workspace, readable provider labels, complete pricing visibility, and provider protocol guidance. Fixed same-session custom-provider model creation, legacy ID migration, nested model references, nullable SSE fields, synchronous settings acknowledgements, runtime-safe blank pricing, user-specific refresh rewriting, and OpenRouter bulk-import behavior.

AhkLLM v1.2.1

Removed the default \n stop sequence from FIM Continue so continuation output is not cut off at the first newline.

AhkLLM v1.2.0

Changelog

[1.2.0] - 2026-09-01

Added

• Support for adding and resolving user-supplied OpenRouter models, including model metadata, reasoning levels, pricing, and provider icons.
• Expanded regression coverage for settings, streaming, branching, attachments, locking, usage tracking, and concurrent requests.
• Isolated parallel WebView2 E2E workers for faster and safer full-application verification.

Improved

• Refined settings labels, shortcut capture and feedback, usage-dashboard navigation, and input-window presentation.
• Clarified context-limit behavior and password-protection limitations in the documentation.

Fixed

• Made Escape cancellation reliable and corrected several settings, FIM, and UI lifecycle edge cases.
• Refreshed the documentation screenshots and custom-paste demonstration assets.

AhkLLM v1.1.0

AhkLLM v1.1.0

What's new

• Interactive screen-region selection for screenshots
• Screenshot preview before sending
• Attach Screenshot support for custom chat commands
• Vision-model, paste-mode, and FIM validation for screenshot commands
• Faster prompt-only commands
• Improved loading and reasoning-state handling for command-created chats

Fixes and reliability

• Screenshot capture, cleanup, cancellation, and IPC fixes
• Improved command/chat state and thinking-setting persistence
• Concurrency and temporary-file cleanup fixes
• More reliable E2E diagnostics and pinned AutoHotkey CI validation

AhkLLM v1.0.0

First public release of AhkLLM.

• System-wide configurable LLM hotkeys, including in-place Refine and FIM commands.
• Full chat GUI with branching/conversation tree.
• Attachments, web search, assistants, usage tracking, and API logs.
• OpenRouter openrouter/free support lets you try normal chat and commands without paid model usage. FIM Fill / Continue still require a provider with an explicit FIM endpoint; DeepSeek is the default setup.
• Requires AutoHotkey v2.0.18+ and the WebView2 Runtime. Windows 11 normally already includes WebView2.
• This is an early public release. Back up data you care about.