GnamiSI
GNAMISIgnamiai.com
Download
HomeChangelog
What changed

Changelog

Every release, in plain English.

WebsiteOctober 3, 2026
  • ·Removed phone notifications for scheduled cloud runs. Phones that had them turned on are unsubscribed automatically the next time they open the Account page. Scheduled-run notices still reach the desktop app, and MCP clients can still read run status.
v2.0.59October 3, 2026
  • ·Added scheduled cloud runs: save a prompt with a daily, weekly or custom schedule in your time zone, and GnamiSI starts it as a normal cloud run, even when the desktop app is closed.
  • ·Added scheduled run history (status, output, credits used and times) in the desktop app and on the Account page, with a desktop notice when a run finishes, fails or a schedule pauses.
  • ·Scheduled runs use the same credits and limits as other cloud runs. A run is skipped when your balance or a daily limit does not allow it, and the schedule pauses when credits run out until you resume it.
v2.0.58September 29, 2026
  • ·GnamiAI is now GnamiSI (SI stands for Super Intelligence). The desktop app, installer and website use the new name. Existing installs, sign-ins and updates carry over.
v2.0.57September 28, 2026
  • ·Removed the empty Settings screen from the desktop app.
v2.0.56September 28, 2026
  • ·Settings no longer shows provider API key fields.
v2.0.55September 26, 2026
  • ·Added installable VS Code and JetBrains IDE integrations for opening a local project and editor selection in GnamiSI desktop.
  • ·Added desktop confirmation for IDE links before switching the workspace; the prompt stays ready for review before running.
v2.0.54September 26, 2026
  • ·Cloud runs use the selected model and failed model requests cost zero credits.
  • ·Removed unsupported Cloud models and stopped Local runs from switching to an unrelated provider CLI.
v2.0.53September 25, 2026
  • ·Added Gnami 32B to the desktop app cloud model picker with separate input and output token prices.
  • ·Gnami can read selected workspace files and apply validated local edits with the configured permission setting.
  • ·Added prepaid Gnami credit checkout and balance display inside the desktop app.
v2.0.52September 24, 2026
  • ·Added macOS ZIP downloads for Intel and Apple Silicon.
  • ·Prepared separate macOS and Linux build workflows.
  • ·Terminal and agent commands now use the host operating system shell behavior.
  • ·macOS packages are unsigned and require approval in Privacy & Security.
v2.0.51September 22, 2026
  • ·Added GPT-6 Sol and GPT-6 Luna to the local model picker.
  • ·Replaced Claude Opus 5 with Claude Opus 5.5 in the local model picker.
v2.0.50September 20, 2026
  • ·Local and cloud runs receive prior chat messages, with the first message preserved when history is shortened.
  • ·The MCP app_read tool returns chat messages, rendered Markdown elements, and a screenshot of the GnamiSI window.
  • ·Captures are requested explicitly, scoped to your account, and retained for ten minutes.
v2.0.49September 20, 2026
  • ·Remote commands use an ordered, account-scoped queue with delivery retries and desktop acknowledgements.
  • ·app_status reports the installed version, last command poll, pending count, and recent command results.
  • ·Commands wait for the workspace to finish loading. Invalid models, unavailable workspaces, and declined runs return errors.
v2.0.48September 20, 2026
  • ·Desktop command polling now reports expired sessions instead of failing silently.
  • ·Queued MCP commands survive KV propagation races and are delivered to the current app state.
  • ·MCP prompts, model changes, mode switches, agents, and new sessions now use live state.
v2.0.47September 20, 2026
  • ·MCP app control now executes prompts, changes models, switches modes, opens agents, and starts sessions reliably.
v2.0.46September 20, 2026
  • ·Desktop MCP bridge is included in the release build: customer-scoped heartbeats and remote app commands now work from the hosted connector.
v2.0.45September 20, 2026
  • ·Agents can now import multiple skills.md files and edit or remove each file independently.
v2.0.44September 20, 2026
  • ·Agent skills now use importable skills.md files with editable Markdown instructions.
  • ·Desktop agent profiles and active-agent selection persist across restarts.
v2.0.43Sep 19, 2026
  • ·Natural agent responses: Removed artificial system prompt constraints that forced models into rigid numbered plans. Model intelligence runs cleanly without prompt interference.
  • ·Streamlined thinking UI: Eliminated fake simulated step lists, [ok]/[..] indicators, and strikethroughs in favor of a clean, real-time "Thinking..." status.
  • ·Cleaner execution pipeline: Stripped mock delay timers and artificial step progression across standard and routine runs.
v2.0.40Sep 19, 2026
  • ·Model catalogue refresh: added Grok 4.7, Gemini 3.1 Pro, Llama 4 Scout, Devstral 2, DeepSeek V4 Pro, DeepSeek V4 Flash and DeepSeek R1; replaced DeepSeek V3.
  • ·Cloud catalogue expanded to 16 Workers AI models (DeepSeek V4 Pro/Flash, GPT-OSS 120B/20B, Qwen3.8 27B, Kimi K2.7 Code, QwQ 32B, Qwen2.5 Coder 32B, Gemma 3 12B) — all IDs confirmed against the official Cloudflare catalogue; removed delisted entries (Llama 3.1 8B, TinyLlama, Qwen2.5 14B).
  • ·Fixed cloud runs sending the wrong model: the request now carries the selected cloud model instead of the local one.
  • ·Local runs now actually pass your model pick to the agent CLI via its real --model flag (Codex, cursor-agent, Copilot CLI, Grok CLI) — previously the selection was silently ignored.
  • ·Removed the silent model fallback in the cloud worker: an unknown model ID now returns a clear error instead of quietly switching to another model.
v2.0.39Sep 19, 2026
  • ·Real system prompts per mode: Autopilot plans before executing, Forge makes minimal targeted changes, Stream stays conversational. Every run now includes workspace context, long-term memory and recent history.
  • ·AI replies render as real Markdown (code blocks with copy button, lists, tables), with contextual follow-up suggestions and a "Memory applied" badge.
  • ·Workspace de-mocked: real empty state with clickable examples, framed composer, working Cmd/Ctrl+K command palette (10 actions), and a real 3-step first-run onboarding.
  • ·Removed all mocks and stubs: dead Autopilot/Stream views with fake blueprints, the fake autopilot plan, the fake welcome message and fake terminal boot lines. Native confirm() replaced with the in-app permission dialogue.
  • ·Model picker now shows "recommended for" tags on each model.
v2.0.38Sep 18, 2026
  • ·Cloud execution no longer requires containers — the AI generates code changes server-side and applies them directly to your local workspace. No more "containers have not been enabled" errors.
  • ·Fixed Grok CLI agent using incorrect flags (--budget, -p) — now uses --single as the Grok CLI expects.
  • ·Download routes updated to serve the new NSIS installer (GnamiAI-Setup-win-x64.exe) alongside the legacy installer name.
v2.0.37Sep 17, 2026
  • ·Model picker now shows cloud models (Qwen3 30B, Llama 3.1 8B, Mistral 7B, etc.) when Cloud is selected, and your local models (GPT-6, Claude, Grok, etc.) when Local is selected.
  • ·Cloud runs now auto-apply changed files directly into your local workspace — no more manual "Apply" button. The agent reply shows how many files were written.
  • ·Local and cloud model selections are remembered independently and persist across restarts.
v2.0.36Sep 17, 2026
  • ·Fixed the auto-updater crashing the app with "EBUSY: resource busy or locked" — the download now guards against concurrent writes, downloads to a .tmp file first, and reuses already-completed downloads.
  • ·Stale installer files from older versions are automatically cleaned from the temp folder.
  • ·All file-stream errors are now caught gracefully instead of crashing the main process.
v2.0.35Sep 17, 2026
  • ·Fixed cloud runs crashing with "sandbox id must be 1-63 characters long" — the ID was too long due to long session IDs combined with a full UUID; now hashed to 24 chars.
  • ·Fixed local models ignoring the effort setting (Low/Med/High/Max) — the effort flag was never forwarded to the agent CLI, so every run used maximum effort and took much longer than needed.
  • ·Fixed cloud runs starting empty — workspace files were collected but never sent in the POST body, so the run had nothing to work on.
  • ·Agent CLI timeout now scales with effort: low-effort runs cap at 60s, medium at 120s, instead of always waiting the full 5 minutes.
v2.0.34Sep 17, 2026
  • ·Cloud agent catalogue expanded to ten models across Qwen, Llama, Mistral, Gemma, Phi, TinyLlama and DeepSeek.
  • ·Cloud runs now pull in files from your open workspace (up to 40 files, 50 KB each) so the cloud run works on your actual code instead of starting empty.
  • ·Expanded cloud model availability surfaced across the site, and the model picker remembers your preferred cloud model.
v2.0.32Sep 16, 2026
  • ·Removed hardcoded AI responses - the agent now responds naturally to all inputs including "what's my name" without canned replies.
  • ·GnamiSI does not collect, store, or use provider API keys.
  • ·Site chat now requires GnamiSI sign-in to post (anti-impersonation), 7TV emotes inline, 1 msg/min/account rate limit, profanity filter.
  • ·Local runs use the Codex CLI or Grok CLI sign-in on your computer. GnamiSI does not receive those credentials.
v2.0.31Sep 16, 2026
  • ·New Settings button in the desktop app. GnamiSI does not collect, store, or use provider API keys.
  • ·Local runs use the Codex CLI or Grok CLI sign-in on your computer. GnamiSI does not receive those credentials.
v2.0.30Sep 16, 2026
  • ·Live indicator cache shortened to 2 minutes so the on-air badge flips shortly after going live.
  • ·Removed the offline notice text from the Live tab.
v2.0.29Sep 16, 2026
  • ·Chat anti-spam: 1 message per minute per account on top of the existing per-IP and global flood valves.
  • ·Whole-word profanity filter (EN + FR) with leet-speak normalization — normal words containing flagged letters are never touched.
v2.0.28Sep 16, 2026
  • ·Live tab chat is fully self-contained: no Twitch login, no bot, no credentials anywhere — pick nothing, just sign in with GnamiSI and chat.
  • ·Posting requires a GnamiSI sign-in with server-verified identity (impersonation impossible); reading stays open to everyone.
  • ·7TV emotes (global + channel sets) render inline, cached daily with graceful fallback.
  • ·Messages auto-delete after 1 hour, history caps at 100, and per-IP plus global flood valves keep traffic at fractions of a cent.
v2.0.27Sep 16, 2026
  • ·Live tab player rebuilt on the direct Twitch player embed.
v2.0.26Sep 16, 2026
  • ·Live tab player rebuilt on the direct Twitch player embed (no script dependency) with fullscreen, autoplay and picture-in-picture support.
v2.0.25Sep 16, 2026
  • ·New Live tab: embedded Twitch player and auto-connected chat for twitch.tv/gabfreakypie with live/offline indication, in site styling.
v2.0.24Sep 16, 2026
  • ·Fixed a startup crash in the desktop app (React #310): the routine scheduler hook now runs unconditionally instead of appearing only after sign-in.
v2.0.23Sep 16, 2026
  • ·Routines: schedule agents to run on autopilot (every 15 min to daily) with their own model, effort and permissions — local-only, so scheduled runs never spend cloud credits.
  • ·Memory: persistent facts injected into every agent run, editable in the sidebar, plus a Remember button on any reply to save it.
  • ·Slash commands in the chat input: /new /stop /clear /models /files /remember /routine /effort /perm /help with keyboard navigation.
v2.0.22Sep 16, 2026
  • ·Geo-routing completed: Canadians on gnamiai.com now land on gnamiai.ca, and only Canadians can browse .ca — everyone else routes to .com.
  • ·www hosts normalize to their resolved apex domain.
v2.0.21Sep 16, 2026
  • ·Mobile header overhaul: compact non-wrapping sign-in and download buttons, tighter spacing, short label on very small screens.
  • ·Hero call-to-action buttons slimmed on phones; desktop sizing untouched.
v2.0.20Sep 16, 2026
  • ·Added Google Analytics (gtag.js) site-wide, loaded after interactive so it never blocks rendering.
v2.0.19Sep 16, 2026
  • ·Added llms.txt at the site root for AI assistants and crawlers.
v2.0.18Sep 16, 2026
  • ·Removed the outdated homepage line that said cloud execution was unavailable — cloud execution is live.
v2.0.17Sep 16, 2026
  • ·Cloud brain upgraded to Qwen3 30B: notably better reasoning, instruction-following and multilingual replies than the previous model, at roughly half the inference cost.
  • ·Automatic fallback to Llama 3.1 8B if the primary model ever fails, so cloud runs keep working.
v2.0.16Sep 16, 2026
  • ·Fixed chit-chat reaching the full billed pipeline: only messages with real task intent (task verbs, code, file refs, commands like "do it") now run the metered agent.
  • ·Greetings, reactions (ok, cool, slt, cc, parfait…) and short chit-chat in both languages take the cheap 2-credit reply lane — locally they stay free and never touch your files.
v2.0.15Sep 16, 2026
  • ·Cloud means cloud now: switching to Cloud syncs your workspace files into the cloud run, runs there, and offers changed files back with one-click Apply to workspace.
  • ·Billing is metered per action instead of flat per run — planning, each execution, repair passes and summaries bill separately, capped at 50 credits per run.
  • ·Failed executions get one automatic repair retry; brief chit-chat bills a 2-credit reply instead of a full run (corrected in v2.0.16).
v2.0.14Sep 16, 2026
  • ·Fixed the updater offering the same version repeatedly: downloads are now cache-busted per release and SHA-512-verified against the manifest before ever prompting to install.
  • ·Stale temp installers from older versions are cleaned automatically, and edge caching on the installer binary was shortened.
v2.0.13Sep 16, 2026
  • ·Cross-domain accounts: signing in on gnamiai.com, gnamiai.ca or gnamiai.live now signs you in on the others automatically via a silent one-time-token handshake.
  • ·Single-use 120-second tokens, strict host allowlist, and anti-fixation checks; explicit logout is respected and never auto-undone.
v2.0.12Sep 16, 2026
  • ·Geo-split domains: gnamiai.live routes Canadians to .ca and everyone else to .com; non-Canadian visitors on .ca route to .com.
  • ·Redirects apply to pages only — API, downloads, updater and OAuth stay pinned and untouched on every domain.
  • ·SEO follows the split: .com canonical base with per-host canonical correction and .com sitemap.
v2.0.11Sep 16, 2026
  • ·Multi-domain launch: gnamiai.live, gnamiai.ca and gnamiai.com all serve the full site, API and downloads. SEO stays consolidated via canonical tags on gnamiai.live.
v2.0.10Sep 16, 2026
  • ·Custom domains live: gnamiai.ca and gnamiai.com now permanently redirect (308) to gnamiai.live, consolidating all SEO value on the canonical domain with zero duplicate content.
v2.0.9Sep 16, 2026
  • ·Cloud runs are now real: switching to Cloud executes tasks remotely (AI-planned script, executed in the cloud, summarized back).
  • ·Bill-safe by design: fixed 50 credits per run deducted upfront, 20 runs per user per day, 120-second max runtime, fresh cloud run that scales to zero.
  • ·Greetings in cloud mode are answered free without burning credits.
  • ·Credit balance now refreshes automatically after each cloud run.
v2.0.8Sep 16, 2026
  • ·New Local/Cloud execution switch in the workspace toolbar. Local (default) runs agents on your machine at 0 credits; Cloud targets a cloud run.
  • ·Cloud execution is hard-gated server-side and currently disabled, so zero cloud spend can occur — switching to Cloud simply explains that until you enable it.
  • ·SEO overhaul: full Open Graph/Twitter metadata, canonical URLs, sitemap.xml, robots.txt, and SoftwareApplication structured data.
v2.0.7Sep 16, 2026
  • ·Single plan: GnamiSI Pro is now the only subscription at $29/month with 12,500 cloud credits per month (Team retired).
  • ·Local runs through your own subscriptions (Codex, Grok, etc.) explicitly cost 0 credits — enforced in the desktop client and shown on every agent reply.
  • ·New Community tab: Twitch and Instagram cards with live auto-updating follower counts, an on-air badge when streaming, plus a Discord coming-soon card.
v2.0.6Sep 16, 2026
  • ·Fixed the agent modifying workspace files for casual messages: greetings and small talk (hi, allo, comment sa va, thanks, bye) are now answered instantly without touching any files.
  • ·Works in both English and French, including messages without a workspace folder open.
  • ·Real tasks are unaffected — anything mentioning files, fixes, or code still goes to the agent.
v2.0.5Sep 16, 2026
  • ·Pressing Enter in the chat input now sends the message (Shift+Enter for a new line).
  • ·Fixed sessions disappearing after restart: chat history is now saved to disk instead of localStorage, which was wiped because the local server uses a new port every launch.
  • ·Added a red Stop button while the agent runs — kills the agent process and keeps any partial output.
  • ·The agent now closes with its own plain-language summary of what it did, which files changed, and how to verify.
  • ·In-app auto-updater now downloads the correct GnamiAI.Installer.exe file.
v2.0.4Sep 16, 2026
  • ·Workspace animations smoothed: mode switches now use a gentle slide-fade instead of a jarring scale pop.
  • ·Sidebar collapse eased to 300ms with a natural cubic-bezier curve instead of snapping at 200ms.
  • ·All interactive buttons gained tactile hover scale and active press-down for a responsive feel.
  • ·Tab bar transitions extended to 200ms with softer inactive hover states.
  • ·Model picker dropdown uses a new origin-anchored scale animation for fluid opening.
  • ·Thread list items now slide in horizontally on render instead of appearing instantly.
  • ·Live status dot uses a soft ping animation with scale growth instead of opacity blink.
  • ·Permission dialog buttons gained scale interactions matching the rest of the UI.
  • ·File context chips and permission mode toggles smoothed with all-direction transitions.
v2.0.3Sep 16, 2026
  • ·Website animations overhauled: staggered entrance reveals on homepage, changelog, FAQ, guides, docs, contact, and account pages.
  • ·Fixed broken modal and banner animations (animate-in classes were non-functional without tailwindcss-animate plugin).
  • ·Modals now animate in with a proper overlay fade and dialog scale-up.
  • ·FAQ and guide accordions use a smooth expand animation.
  • ·Reduced-motion support added for accessibility (prefers-reduced-motion).
  • ·Workspace inline keyframes consolidated into globals.css.
  • ·Provider model icons fixed: color now inherits from parent text instead of hardcoded gray, so icons match selected/highlighted states.
  • ·All emoji and unicode UI symbols removed from the workspace app, replaced with crisp lucide-react icons (folder, menu, close, logout, etc.).
  • ·Chat UX improved: agent now shows a real-time activity feed while running instead of a static bouncing-dots indicator.
  • ·Activity steps show a clean checkmark when done and a spinner on the active step; a prominent "Thinking..." label is visible throughout.
v2.0.1Sep 15, 2026
  • ·Replaced the stock Windows installer wizard with a custom GnamiAI installer.
  • ·Added a Choose folder button for selecting the install location.
  • ·Added GnamiAI branding, a desktop shortcut, gnamiai:// link handling, and an uninstaller.
  • ·Fixed Grok permission commands so Ask, Auto, and Full use valid CLI options.
  • ·Fixed provider icon sizing in the model picker.
  • ·Unsupported providers now show a clear message instead of starting a different CLI.
v1.3.0Sep 15, 2026
  • ·Auto-update: app now checks for new versions on launch and installs them on next restart
  • ·Added real provider logos (OpenAI, Anthropic, Google, xAI, Meta, Mistral, DeepSeek) in model picker
  • ·Hippo logo now shows in website header, browser tab, and app sidebar
  • ·Session state (threads, messages, model choice) now persists across app restarts
  • ·Fixed codex "not on PATH" error — agent detection now uses shell lookup (where/which)
  • ·Removed provider selector buttons — model picker alone drives which CLI runs
  • ·Homepage rewritten — removed placeholder copy and fake benchmarks
v1.2.0Sep 15, 2026
  • ·Workspace UI redesigned with sidebar, tab bar (Autopilot / Forge / Stream), and animated transitions
  • ·Model picker now grouped by provider with effort slider (Low / Med / High / Max)
  • ·Permission mode selector: Ask · Auto · Full
  • ·Thinking animation shown while agent is running
  • ·Forge mode: file tree + textarea editor with save to disk
  • ·Stream mode: terminal with colored output, runs commands in workspace folder
  • ·Grok output-format flag fixed (plain instead of text)
v1.1.0Sep 14, 2026
  • ·Google Sign-In via OAuth 2.0 — works in both browser and desktop deep-link flow
  • ·Desktop session persisted to disk (userData/session.json), restored on next launch
  • ·Local HTTP proxy in Electron forwards /api/* requests to gnamiai.live with session cookie
  • ·Workspace folder picker — agents and terminal run inside selected directory
  • ·Agent detection on launch via where/which lookup
v1.0.0Sep 10, 2026
  • ·First release — Windows 64-bit installer (.exe) built with Electron + Next.js static export
  • ·Cloudflare Worker backend: Google OAuth, KV sessions, PayPal billing, R2 installer hosting
  • ·Codex, Cursor Agent, Grok CLI integration in desktop app
  • ·Black-and-white UI, monospace font throughout