• iFLYTEK Spark now runs on iFLYTEK Xingchen MaaS: one provider, one API key, one status row. • New Spark models: Spark X2.6 (Thinking), Spark X2.6, Spark X2 Flash and Spark X2.6 4B. • Action needed: older iFLYTEK keys no longer work. Enter your Xingchen MaaS key in Settings. • Conversations with retired Spark models still open. • Clearer messages during storage migration, in all 13 languages.

SageGate 1.2.0 — the big one. In August I wrote on our website that a major release was on the way. Thank you for waiting. Since 1.1.4 I've been building and testing what's here today. The headline is Cubicles (Beta): a private workspace for each project, with its own chats, documents, instructions and model, so work for one client or topic never mixes with another. Alongside it, memory is safer, costs are clearer and storage is more robust. Feedback shapes every release; write to contact@sagegate.ai. • Cubicles (Beta) — give each project its own workspace with its own chats, documents, instructions, and model choice. Chats in a Cubicle see only that Cubicle's material. Cubicles run with models that have passed SageGate's tool-use test. • Cubicle requests run on their own — each one sends only your text and text files, with no earlier messages, photos or PDFs. • More models proven for Cubicles — Claude Fable 5.1, GPT 6 Astra, GPT 6 Luna, GPT 6.1 Sol, GLM 5.3, GLM 5.3 Flash and Grok 4.7 have now passed the tool-use test. • Cubicle runs check the model's tested tool limits before they start, and tell you plainly if a run is over them so you can turn off some connectors. • Grok 4.7 replaces Grok 4.5 and Grok 4.3. Chats on the retired models move to Grok 4.7, a one-time "Model retired" notice shows the price change, and past usage of retired models keeps its price in the Cost Dashboard, exports and sorting. • Math in answers — formulas now appear typeset in chat and in PDF exports, with VoiceOver descriptions. • Safer memory — new memories have personal details such as card numbers, account and licence numbers, and passwords masked before they are saved, and memories you already have are cleaned once. Upgrading your memory data is protected: if something goes wrong, SageGate rolls back instead of losing it. • Clearer costs — when a model bills a long request at a higher rate, the Cost Dashboard now says so next to the usage. • Safer storage moves — if copying your data to a new location fails, the destination is put back as it was, and SageGate tells you plainly when a data folder can't be written. • Safer first launch after an iCloud upgrade — if the upgrade is interrupted while iCloud is unavailable, SageGate waits for iCloud and continues by itself, instead of offering a device-only switch that could lock you out. • Model Library is now grouped by tier, so the right model is easier to find. • Your saved API keys are no longer cleared when you reinstall the app, and subscription status refreshes when you return to SageGate.

Version 1.1.4: Thank you to everyone who sends us feedback. If you have suggestions, feature requests, or bug reports, please write to contact@sagegate.ai. • Web search expands to 6 more providers — Qwen, ERNIE, GLM, Kimi, Spark, and OpenRouter now support live web search (opt-in under Settings → Advanced Settings), with transparent per-search fee tracking in the Cost Dashboard. • Newest models on tap — GPT-5.6 (Sol, Terra, Luna), Grok 4.5, MiniMax M3, and Kimi K2.7 Code, with up-to-date pricing so your cost tracking stays accurate. • Modernized provider connections — GPT-5.6 (Sol, Terra, Luna), Grok 4.5, Qwen, and Doubao now run on each provider's latest Responses API endpoint, and DeepSeek connects through its new Anthropic-compatible endpoint — keeping SageGate up to date and bringing built-in web search to these models. • Listen to any answer — a new Play button on every assistant message reads the response aloud in your language, so you can catch up hands-free. • New Privacy Settings hub — Settings → Privacy Settings is now organized into Policy, Cloud Sync, Personal Knowledge, and Activity, so everything about your data lives in one place. • Model Router controls — see which routes are available at a glance, keep Auto choosing only from your Model Library, and manage AWS chat inference, all from one screen. • Memory Controls — turn auto-capture and semantic recall on or off, review what's been remembered, and export your memories to a file whenever you like. • Cloud & Privacy Activity — a clear, exportable log of every upload, deletion, erase, setting change, and export SageGate makes with your data, so nothing happens behind your back. • Plus a friendlier, editable Living Profile and safer cloud deletion and erase under the hood.

Version 1.1.3: Thank you to everyone who sends us feedback. If you have suggestions, feature requests, or bug reports, please write to contact@sagegate.ai. • More reliable messages — we rebuilt how every request is handled behind the scenes, so sending is steadier and far less likely to stall or fail silently. • Long conversations just work — when a chat gets long, SageGate now automatically trims the oldest context to fit the model's window, and retries once if a request is still too large, instead of failing. Old attachments are set aside before their text, so the conversation keeps its words. • Clearer error messages — when something does go wrong, you get a plain, translated explanation of what happened and what to try, instead of a raw provider error. • Fixed a usage double-count — reopening a chat while a reply was still streaming and then tapping Stop no longer counts the same response twice in your totals. • New Diagnostics screen in Settings — see at a glance what's connected and working, and export a status report to attach to a bug report. The report is fully redacted: it never includes your chat content, API keys, or account details.

Version 1.1.2: Thank you to the users who pointed out these bugs. If anyone else has any recommendations, new feature requests, or bug reports, please send them to contact@sagegate.ai. • Auto mode now shows its work — when SageGate picks the fastest model for you within your privacy setting, the message header tells you exactly which one answered, no more guessing. • Fixed a cost-tracking bug — messages answered by Auto now count toward your usage and cost totals. • Fixed a scrolling bug users reported — chat now scrolls smoothly and naturally as replies stream in, and the AI's thinking process folds into a compact box you can tap to expand. • New Feedback & Support in Settings lets you send bug reports and feature requests directly; contact details are optional. • We rebuilt token counting for every provider — cached, thinking, and regular tokens are now tallied by each provider's official rules, and we keep monitoring these numbers over time, because a cost dashboard you can trust is what this app is built on. Every model's pricing, official release date, and knowledge cutoff are disclosed where you can see them (when we can't verify a figure, we leave it blank rather than guess) — and one honest note: some providers spend your free or promotional tokens first, so your balance can look untouched even though the request really happened.

Thank you to the users who suggested these features — your requests make SageGate better for everyone. Everything below is new since v1.0.7. The app is still free and still needs no account; all the cloud features are optional. SAGEGATE CLOUD — YOUR OWN KNOWLEDGE BASE, SEARCHED BY EVERY MODEL (OPTIONAL) Upload your documents once and every AI you chat with can search them by meaning — your knowledge stops being trapped inside a single model. Three optional paid tiers, still with no account (your subscription is your identity): • T4 Managed Cloud ($19.99/mo) — we host your encrypted vault; zero setup. • T3 Private Cloud ($5.99/mo) — the same, but your documents live in your own AWS. We handle only the transport (convert, chunk, embed, ingest, retrieve) and keep nothing on our side. • T2 Self-Host ($1.99/mo) — your AWS only, extracted on your device; we never touch a byte. Includes upload with live processing status, a searchable document library, a page balance with one-tap top-ups, and Erase Cloud Data whenever you want. Everything moves over Apple's secure transport layer (URLSession + TLS). Managed and Private Cloud index your documents with Cohere Embed v4 (multilingual), with Amazon Titan Text Embeddings v2 as a fallback. UPLOAD BY WEBSITE Add and manage your cloud documents from any web browser at portal.sagegate.ai, securely authorized by your phone — no separate login. AMAZON BEDROCK Connect your own AWS and run Claude and Amazon Nova models through Amazon Bedrock, signed with your AWS credentials. Set it up with AWS IAM Identity Center sign-in, a CloudFormation quick-create, or manual keys. AUTO MODEL ROUTER + PRIVACY CEILING Let SageGate pick the best model for each message by cost, speed and capability — while never crossing a privacy line you set (from "my keys only" up to managed cloud). SEARCH YOUR CONVERSATIONS Find past chats by meaning, not just keywords — powered by Apple's on-device text model, so nothing leaves your phone. CONTROL CENTER + QUICK ACTIONS Add a SageGate tile to Control Center to jump straight into a new chat; sagegate:// deep links open the app right where you want. CUSTOM API ENDPOINT (PER PROVIDER) Set your own API address for any provider. Only clean https hosts are accepted; anything else falls back to the official endpoint, so your key is never sent somewhere unexpected. LANGUAGE SETTING MOVED TO THE TOP The app-language picker now lives on the main Settings screen instead of under Advanced Settings. TENCENT HUNYUAN — NEW ENDPOINT Tencent moved HunYuan to its new TokenHub gateway. SageGate now connects there — if you use HunYuan, add your TokenHub API key. Everything you already use stays free, forever. Subscribe, redeem codes, and manage plans in Settings — billed through your Apple ID.

New model: • Claude Sonnet 5 joins the catalog as the new default — Anthropic's latest Sonnet, with a 1M-token context window

New connections and models: • Custom inference servers now work over Tailscale — connect to your own OpenAI-compatible server using its Tailscale (.ts.net) hostname • New models — GLM-5.2, now with a 1M-token context window, plus Doubao Seed 2.1 Pro and Turbo, join the catalog

Updated models and attachments: • New models — Claude Fable 5 and Opus 4.8, Grok 4.3, Gemini 3.5 Flash, Qwen 3.7, ERNIE 5.1, and MiniMax M2.7 join the catalog • PDF attachments — Claude and Gemini read PDFs natively, GPT, Doubao, and ERNIE receive rendered page images, and models without PDF support now say so clearly instead of failing silently • DOCX, XLSX, and PPTX attachments — document text is extracted when you attach and delivered with your message

Updated models and attachments: • Grok 4.20 — replaces Grok 4.1 Fast in reasoning and everyday tiers, xAI's newest flagship • MiniMax M2.7 — replaces M2.5 in best and everyday tiers, with stronger agentic and coding performance • Attachment fix: PDF, DOCX, XLSX, and PPTX content is now extracted at attach-time and delivered to every supported model — earlier builds only passed the filename, now models see the real document

Model catalog refresh: • GPT-5.5 Pro joins as OpenAI flagship; GPT-5.5 replaces GPT-5.1 in the everyday tier • DeepSeek V4 Pro and V4 Flash launch with native thinking-mode toggle • Kimi K2.6 added with dual reasoning / non-thinking modes; Kimi K2.5 retired

Model catalog refresh: Claude Opus 4.7 replaces Opus 4.6 as the Anthropic flagship GLM-5.1 introduced as Zhipu's new flagship GLM-5 Turbo reclassified to the everyday tier; GLM-5 removed ERNIE 5.0 moved to the reasoning tier with corrected 119K context window ERNIE 4.5 Turbo updated to the 20260402 snapshot (123K context)

What's New in Version 1.0.1 - Improved accessibility: VoiceOver labels applied across all 18 views - HunYuan: thinking/reasoning token usage now tracked correctly - Gemini: thinking stream routed properly for models with reasoning output - ERNIE: pricing and model tier corrections - Stability improvements and minor bug fixes