Changelog

v1.0.0 Aug 5, 2026

  • Final release — product line closed; no further feature development planned
  • Build expiration window is ~6 months (183 days from package date; cortex.beta.expirationDate)
  • Usage-limit UX, sticky active turn, and reload history visibility from the 0.2.2 line

v0.2.2 Aug 3, 2026

  • Provider usage limits no longer show “Retrying in 398047s”: clear reset time, Change model / billing, and stop the multi-day wait
  • Active turn stays sticky at the top while tools/thinking scroll underneath (until the turn goes idle)
  • Reload no longer hides the transcript behind empty-session opacity while history is re-fetching
  • Faster boot restore: only the active tab loads history at startup; skips image base64 and duplicate fetches that froze Cursor
  • Paginated session history: boot loads the newest ~40 messages; Load older / scroll-up fetches the previous OpenCode page
  • Poisoned-session guard: strips huge /.cortex/index DiffSummary patches so one bad tab cannot freeze Cursor
  • Rollback only on user messages: no fork/return on assistant or tool steps
  • No copy/resend bar under subagent or tool cards — actions stay on user prompts only
  • Permission toast no longer flickers on poll re-delivery of the same request
  • Chat scroll stays put when diffs/permissions appear; timeline paint no longer jumps to the top
  • Live subagent progress on the Parts→rows timeline: task/subagent card stays open with activity, steps, and elapsed time (not just the child session id)
  • Single timeline path: live activity, system/error/tool notices, and fork hints remount after paint (no dual legacy/timeline path)
  • Active tab restores session history after boot: message fetch times out instead of hanging; restore reclaims OpenCode and retries the focused tab
  • Boot recovers a hung OpenCode: if the port accepts TCP but health times out, Cortex kills and respawns instead of spinning past the boot ceiling
  • Boot splash no longer flashes English: ships configured locale and a short language roulette before quips
  • Restored session scrolls to the end after boot (jump waits until the chat panel is visible)
  • Enriched diffs + live progress on the Parts→rows timeline: Keep/Undo mini cards again; Thinking between tool steps; stuck running shimmer clears when idle
  • Timeline restore: fork/resend/copy on hover, scroll-to-end when opening a chat, and @ subagent picker in the composer
  • Permission auto-accept no longer hangs chat: replies hit real OpenCode routes and ignore SPA HTML 200s that left bash stuck on In progress
  • Tool hang watchdog no longer aborts while waiting on question cards or bash permission prompts
  • Hardware fit tip for Performance/Quality: detects RAM/CPU, shows if Quality (~150 MB MiniLM) suits this machine, caps embed concurrency on tighter PCs
  • Docs synced with v0.2.0: settings table, Command Palette commands, Servers/themes/notifications/composer; OpenCode CLI ≥ 1.18.0 in install requirements
  • Compare page: honest Cortex vs Cursor vs Devin vs Zoo/Roo table
  • Docs: migrate from Cursor / Copilot / Roo-Cline

v0.2.1 Aug 3, 2026

  • New session watermark: OpenCode Desktop Inter font + Cortex logo behind Cortex/opencode; empty composer lower, brand stays upper; boot and empty share the same layers (no resize jump)

v0.2.0 Aug 3, 2026

  • SNAPSHOT Debug Mode: in-app debug console (events, health, turn trace, export JSONL) gated to SNAPSHOT builds
  • Session timeline is always Parts→rows (tools, DiffSummary, retry, MessageNav, Load older, jump-to-latest); removed Desktop timeline beta toggle
  • Tool hang watchdog: auto-abort when bash/tools stall with no progress so leaving mid-turn no longer leaves chat stuck busy
  • Restored session tabs show history again after reload (no empty chat with leftover turn ticks)
  • OpenCode Go model guide in the picker (tag, description, best-for) and sharper Auto routing from the same signals
  • Quality loading tip is a speech bubble on the Quality button (stays ~4s minimum)
  • Composer placeholders rotate witty hooks with / and @ hints (Desktop-style)
  • Session tabs: pencil on hover (or double-click) to rename inline
  • Boot shows Cortex/opencode watermark immediately (no blank flash)
  • Sidebar height fixed to stop the always-on side scrollbar
  • New session empty state matches OpenCode Desktop (centered composer + Cortex/opencode acid-green watermark); Idle status hidden; docked chat layout after the first message
  • Topbar action icons use soft lilac/white instead of theme green
  • Composer + menu matches OpenCode Desktop: Images and files (Ctrl/Cmd+U), Commands (/), Context (@), Shell command (!)
  • Model variants (Default/Low/Medium/High…) when available, next to the model picker
  • Attachments gated by model inputs; incompatible files removed with a warning on model switch
  • Model picker: search, hover details (inputs / reasoning / context), and Manage models shortcut
  • Sessions history is its own screen (Desktop-like); chats renamed to sessions in the UI and commands
  • Settings → Servers matches OpenCode Desktop: Local + remote HTTP connections (health, add/edit/delete, default/active) — not the old MCP list
  • Servers: ⋯ menu no longer clipped by the card
  • Providers/Servers forms: Desktop-style icon back (←); Providers API key + Submit stacked
  • Models: provider-level toggle to show/hide all models for a provider
  • Default model picker moved to Models; removed OpenCode plugins / instructions / skills from General
  • Shortcuts: VS Code keybindings (not OpenCode Desktop); Inline Edit Ctrl+Alt+I / Cmd+Alt+Shift+I; terminal selection Ctrl/Cmd+Alt+T — avoids DevTools and Reopen Closed Editor
  • Providers screen aligned with OpenCode Desktop: popular + full catalog, API key / OAuth connect, custom OpenAI-compatible providers
  • OpenCode Zen is one provider in the catalog (no longer a separate Connect-only flow)
  • Talks to OpenCode via the official SDK for providers, auth, and config; requires OpenCode CLI ≥ 1.18.0
  • Default model stored in OpenCode opencode.json; cortex.providerID/modelID deprecated
  • Models: show/hide + favorites in the picker; default model picker (opencode.json)
  • Providers / Models Desktop UX: official icons; Providers Connected/Popular + Connect picker; Models list with visibility toggles and in-row default radio (no separate dropdown)
  • Models favorite star is larger and gold when active
  • Models provider lists are collapsible (active provider open by default)
  • Settings shell aligned with OpenCode Desktop: Desktop (General, Shortcuts) + Server (Servers, Providers, Models); dedicated chat menu vs ⋯ for workspace/global; close with X
  • Settings: theme picker no longer clipped; Idle status hidden on settings screens
  • OpenCode default model picker grouped by provider
  • General settings Desktop parity: shell, auto-accept permissions, reasoning/tool expand, color scheme, fonts, notifications, sounds
  • OpenCode themes in chat: GitHub catalog + cache, custom ~/.config themes, follow tui.json; Cortex theme embedded until upstream
  • Chat spacing tuned (less top empty margin, more room under composer); timeline context/Shell cards follow the active theme
  • Localized status “waiting for reply” and Plan/Build labels in session meta (all languages)

v0.1.6 Jul 31, 2026

  • Windows: OpenCode CLI detection uses PATH (where.exe) and known install dirs — no longer fails when opencode is only on PATH
  • Windows: npm .cmd shims for OpenCode spawn correctly via shell
  • Chat no longer stays stuck on “Retrieving local context…” after a large Quality MiniLM reload (status, hydrate, and retrieve timeouts)
  • Explicit “make/traza un plan” switches Build→Plan even outside Auto; queued messages wait for question/permission answers; stale request-not-found errors suppressed

v0.1.5 Jul 31, 2026

  • Opening a chat from History no longer closes the Cortex panel in Cursor
  • Quality (MiniLM): native onnxruntime in ~/.cache (runtime-v2, not in VSIX); unlock target ≤~30s
  • Quality: send locked until ready (meter on send button); no silent Performance fallback
  • Quality file changes re-embed only touched files (no full reindex / % reset)
  • Docs: Quality MiniLM cache, send gate, and incremental index documented in config/architecture
  • Quality MiniLM starts after chat unlock — does not freeze onboarding/boot
  • Quality MiniLM progress shown as % in the status bar and chat toolbar
  • Onboarding wizard: OpenCode CLI + API key checklist with language and context mode
  • Paste API key in onboarding or Connect — saved to OpenCode auth.json even if the local server is not up yet
  • Chat without a project folder — uses your home directory for general / system questions
  • Windows: path case mismatch (C:\ vs c:\) no longer blocks OpenCode Connect or chat boot
  • Windows: OpenCode install check uses PATH (where.exe) and known install dirs, not cwd-relative opencode
  • Clearer error when the OpenCode binary is missing on startup
  • Chat timeline chronology and command-output defaults; affected-files bar pinned above the composer
  • Progress panel no longer leaks across chat tabs — each tab caches its own progress state

v0.1.4 Jul 30, 2026

  • Live tool/subagent labels follow the UI language (no more Spanish “Leyendo / En curso” when English is selected)
  • Landing/Marketplace demo refreshed — @explore subagent and clarifying question card
  • Auto/Turbo specialty routing — best available model per task (coding / reason / speed); Cost/Balance/Intelligence still tilt quality vs speed
  • Model picker groups by provider (collapsible sections with counts)
  • Chat opens as a right editor panel (not secondary sidebar) so Cursor tab icons stay intact
  • Pin Cortex chat tab (`cortex.chat.pinEditorTab`, on by default); brand logo as tab icon
  • Optional header version badge (`cortex.ui.showHeaderVersion`, on by default — turn off for demos)
  • Connect redesign: Get API key → paste → Save; Models for Go/paid Zen/BYOK
  • Auto/Turbo with OpenCode Go: prefer Go models first; free Zen only as fallback
  • Auto Plan/Build: clear tasks stay in Build; explicit “create a plan” enters Plan; ambiguous asks get a one-time card
  • Chat history grouped by recency; tool steps show real file path, grep, or command
  • Removed Terminals panel — use the IDE terminal list; keep @Terminal and Send selection to chat
  • Docs: full settings table (optimizeFor, Zen/BYOK unlocks, autocomplete, context include/exclude) + keyboard shortcuts

v0.1.3 Jul 29, 2026

  • Credentials: one OpenCode card (key mask + catalog); no Zen/Go switch or duplicate Back
  • Auto-detect OpenCode Go after unlock (non-blocking) and switch automatically
  • Enable OpenCode Go from Credentials (reuses Zen API key; no TUI) — Subscribe link for billing
  • Landing: Auto 2.0, Policy Brain, editor parity; OpenCode-first get-started (no Cortex account)
  • Auto 2.0 intelligent routing — intent + complexity picks model tier; Auto switches Plan/Build
  • Model availability filtering — unavailable models dimmed; free Zen unlocked; paid Zen/BYOK opt-in
  • Retrieval Performance/Quality toggle in chat (hash ↔ MiniLM)
  • Active editor file auto-attached each turn; context detail popup on the meter
  • Hint when the active model is free/light (even with paid access unlocked)
  • Cursor-like per-hunk diffs; Plan mode can save/refine plans under .cortex/plans/

v0.1.2 Jul 28, 2026

  • Logo next to the "Cortex" title in the chat sidebar header
  • Policy Brain: multi-tier rules/skills/memory/hooks (compatible with .cursor/rules and .cortex/rules)
  • Policy Inspector, global user rules, and built-in changelog gate
  • Parallel chat tabs no longer mix thinking spinners or stream progress across chats
  • Undo/Reject/Restore Checkpoint only reverse that chat's file delta — other chats' non-overlapping edits are kept
  • File diff cards stay above the final assistant reply (edits happen during tools) instead of under the summary text
  • Policy stats and retrieval stay out of the chat; Policy Inspector is per chat
  • Resend / fork no longer appends (or stacks) a language hint onto the user message

v0.1.1 Jul 28, 2026

  • Spanish landing copy: codear → codificar, refactorea → refactoriza
  • Second workspace no longer fails when opencodePort is busy — Cortex uses the next free port
  • Hero demo slot on the landing (GIF/MP4 when available; placeholder until then)
  • Portrait hero demo no longer cropped to 16:9
  • Landing header and favicon use the new Cortex logo
  • Removed legacy SVG brand assets; PNG only
  • Comparison table: footprint, native OpenCode, and IDE UX vs Cursor / Windsurf
  • Active product version shown in the nav and footer (links to changelog)
  • Open VSX install CTA on the landing and install docs
  • Marketplace / Open VSX listing points to the public cortex repo (not the private source)
  • End-user README only; proprietary license (source not open source)
  • Marketplace gallery icon and banner
  • Icons for all menu items in the "More" menu (compact, undo, reindex, MCP/skills, remember, rules, plans, background agent, language, OpenCode)
  • Flag emojis for each language in the language submenu and language button

v0.1.0 Jul 27, 2026

  • Sidebar chat with OpenCode integration
  • Local workspace indexing (auto/manual)
  • Lexical retrieval (TF·IDF + path boost) with hash embeddings (optional MiniLM)
  • Symbol-aware chunking (LSP DocumentSymbol)
  • Context preview in UI
  • Streaming replies via OpenCode SSE
  • Session history browser and undo turn
  • Keep / Undo / Review per changed file (diff before/after)
  • Permission / question approval UI
  • Ghost-text autocomplete via a fast no-tools OpenCode complete agent
  • Manual autocomplete trigger (Ctrl/Cmd+Alt+\) and Inline Edit (Ctrl/Cmd+Shift+I)
  • @-mention support for files, folders, terminals, and agents
  • /slash commands (agent switch, provider connect, help)
  • Beta expiration system with warning notifications