Repository navigation
π GhostEdit v1.9.0 β Per-Provider API Keys, Grok, Simplified Hotkeys (+ everything since v1.5.1)
Note on versioning: the upstream repo (
nareshnavinash/GhostEdit) last published v1.5.1. This v1.9.0 release rolls up everything new since v1.5.1 β v1.6.0 + v1.7.0 + v1.8.0 + v1.9.0 β so nothing you missed in between is lost. If you are coming from v1.5.x, read all four sections below.
β¨ v1.9.0 β What's new in this release (headline)
π Per-provider API-key system (your API-key system)
Every API preset now keeps its own model + base URL + encrypted key (~/.ghostedit/api-key-<profile>.enc), so you can configure several providers at once instead of one global key:
- New Grok (xAI) preset:
https://api.x.ai/v1withgrok-3-mini/grok-3/grok-2-1212 - Full 11-preset matrix: OpenAI, OpenRouter, Groq, Together AI, Grok (xAI), Claude (Anthropic API), Codex (OpenAI API), Gemini (Google API), Ollama, LM Studio, Custom
- Each Providers-tab card shows Configured / Not configured β hosted providers = key saved; keyless (Ollama / LM Studio / Custom) = model confirmed
- Curated model dropdowns per provider + Refresh to list live models from the host (saved custom model always kept)
- Settings β General β Default correction model only offers providers you actually configured (Local always; CLI only when installed; API only when configured)
- Legacy single-key (
~/.ghostedit/api-key.enc) configs auto-migrate into the profile you had selected
β¨οΈ Simpler hotkeys + tray
- 2 correction shortcuts instead of 3:
Cmd/Ctrl+Shift+E(Local) andCmd/Ctrl+E(General-tab provider, resolved via the newprovider-shortcutmodule) - Tray now shows Correct (Local) + Correct (API) β the dedicated CLI item is removed (CLI status still shown as
CLI: <Name> / <model>line) - Old configs auto-migrate: former
Ctrl+E/Ctrl+Shift+Epair swapped, customAlt+Epreserved when customized
π Safer, smarter corrections
- New
replaceSelectedText(): re-copy β verify β Backspace β paste (fixes editors whose paste inserts at caret instead of replacing) - Each CLI (Claude / Codex / Gemini) remembers its own model (
cliModels) - Proofreading system prompt hardened β edits only, never answers embedded questions, dialogue, or sensitive content
- Custom endpoints fall back to the SSE streaming path (for browser-backed bridges) + friendly
conversation_deletedbridge error - Onboarding installs the local model before the Try-it step (new
onInstallLocalModelflow) - Tests: 530 passing across 28 files (new
provider-shortcutcoverage; fixed 8 stale expectations),tscclean
π¦ Included from v1.8.0 β OpenAI-compatible API provider (first appearance upstream)
- π OpenAI-compatible provider: streaming + non-streaming corrections via Chat Completions, with native support for Anthropic Messages, OpenAI Responses, and Google
generateContent/SSE endpoints - β‘ 10 presets at introduction (β 11 in v1.9.0 with Grok): OpenAI, OpenRouter, Groq, Together AI, Ollama, LM Studio, Custom + Claude/Codex/Gemini native APIs, each with default base URL + model
- π Encrypted API key storage (Electron
safeStorage,api-key-store.ts) +GET_API_KEY/SAVE_API_KEY/GET_API_MODELSIPC; keys never inconfig.json - β¨οΈ API hotkey + Correct (API) tray entry (was
Ctrl/Cmd+Alt+Ein v1.8.0 β simplified toCtrl/Cmd+EGeneral-provider hotkey in v1.9.0) - π§ Onboarding API setup step (preset/URL/model/key) + Bonsai size picker, scrollable layout; Settings API key + Load Models discovery (β per-profile Refresh cards in v1.9.0); Local Model section; T5 β Bonsai auto-migration
- π οΈ
CORRECT_INLINEnow honors the selected provider instead of forcing local - ποΈ Host-aware Forge makers (Squirrel on Windows, DMG on macOS, ZIP cross-builds for win32/darwin from Linux)
- π¨
ghost-*design tokens replace hardcoded blue/red/yellow across History, HUD, Suggestions, DiffView
π¦ Included from v1.7.0 β Bundled llama.cpp runtime + startup model setup + cross-platform installers
- π¦ Self-contained installers: pinned llama.cpp server
b11429, SHA-256 verified per platform/arch (scripts/download-llama-server.mjs), hooked into ForgeprePackage; newllama-runtime-manager.tsresolves the bundled path (packaged + dev) - β¬οΈ On-demand Bonsai models to
~/.ghostedit/models/bonsai/(1.7B 248 MB default / 4B 572 MB / 8B 1.1 GB) with streaming progress + latest-revision check; installers no longer ship weights - π Startup setup flow:
StartupSetupStatusIPC, progress UI in Settings with retry, tray Download Local Modelβ¦ action, Welcome triggers download when Local chosen;PREVIEW_READYhandshake replaces fixed 150 ms delay - ποΈ Debian makers +
make:deb,make:win,make:mac:arm64/x64(+ arm64 variants) for Linux/Windows/macOS x64+arm64
π¦ Included from v1.6.0 β Smarter correction engine (identifier + token safety)
- Skips kebab-case (
my-repo), snake_case (user_name), PascalCase (StringBuilder), env vars ($NODE_ENV), CLI flags (--no-cache), version strings (v1.2.3), and dot-qualified names - ~30 common tech abbreviations added to the dictionary (
src/main/data/tech-dictionary.txt) - Token-preservation hardening (URLs, @mentions, code, emoji, paths untouched) with 85 new token tests + 80 dictionary tests
β¬οΈ Downloads (6 files β 4 zips + 2 debs)
| π₯οΈ Platform | π Arch | π¦ File | Approx. size |
|---|---|---|---|
| π macOS | Apple Silicon (arm64) | GhostEdit-darwin-arm64-1.9.0.zip |
~480 MB |
| π macOS | Intel (x64) | GhostEdit-darwin-x64-1.9.0.zip |
~483 MB |
| πͺ Windows | x64 | GhostEdit-win32-x64-1.9.0.zip |
~492 MB |
| πͺ Windows | ARM64 | GhostEdit-win32-arm64-1.9.0.zip |
~490 MB |
| π§ Debian/Ubuntu | amd64 (x64) | ghostedit_1.9.0_amd64.deb |
~322 MB |
| π§ Debian/Ubuntu | arm64 | ghostedit_1.9.0_arm64.deb |
~318 MB |
Each bundle includes the pinned llama.cpp runtime; Bonsai models download on demand to ~/.ghostedit/models/bonsai/. Windows ZIPs are portable builds from Linux β for native Squirrel Setup.exe + auto-update, build on Windows with npm run make:win.