{"generatedAt":"2026-11-22T12:49:58.580Z","source":"ismscopilot-marketing","contract":"v1","product":"platform","canonicalHumanPage":"https://platform.ismscopilot.com/changelog","note":"Scrubbed mirror of the Platform product changelog (API, embed, console). GitHub PR URLs removed. Chat product ships are on /api/public/changelog/v1.","markdown":"# ISMS Copilot Platform Changelog\n\nAll notable user-facing changes to the **ISMS Copilot Platform** (API, embed, and platform.ismscopilot.com console), grouped by month.\n\n## How to add an entry (authors and agents)\n\n**Canon version:** `changelog-voice/2026-07-v1`  \nShared voice with chat: see `docs/CHANGELOG.md` How-to and `docs/multi-product-changelog-design.md`.  \nClassification: **VISIBLE-PLATFORM**. Weekly Mac `changelog-merge` owns gaps.\n\nChat-only ships stay in `docs/CHANGELOG.md`. heyGRC ships stay in `docs/heygrc/CHANGELOG.md`. Aevral console ships stay in `docs/aevral/CHANGELOG.md`.\n\n### Shape (July 2026 and later)\n\n```\n## Month YYYY\n\n### Features\n\n- **Short title.** One plain sentence.\n```\n\nCategories: `Features`, `Improvements`, `Fixes`, `Privacy & legal`, `Safety`.\n\nValidate: `node scripts/validate-changelog.mjs docs/platform/CHANGELOG.md`\n\nAfter editing this file, run `node scripts/sync-app-changelogs.mjs` so the platform Vercel app mirror updates.\n\nPre-split history: some earlier Platform ships also appeared on the chat product changelog before the multi-product cutover (2026-07-31).\n\n---\n\n## November 2026\n\n### Features\n\n- **Half price on cached input.** Prompt tokens the model host serves from its cache now cost half the input rate on `isms-fast`, `isms-thinking` and `isms-mini`, and the cached count shows in API responses, the usage view and the CSV export.\n- **Auto-reload for API credits.** Save your card on a top-up and your balance tops itself up when it falls below a threshold you choose, with a monthly limit, so your API traffic does not stop at zero.\n- **Invoices for API credits.** Top-ups and auto-reloads get a Stripe invoice with VAT computed on top of the amount, you still receive exactly the amount in credits, and your invoices are listed on the Credits page.\n- **Monthly spending limit and billing alerts for the API.** Set a monthly spending limit on the Credits page and requests stop there with `account_spend_limit_exceeded` until next month or until you raise it, and you can get low-balance and limit email alerts.\n- **API usage by day, model and key, with a CSV download.** The Usage page now charts your API usage per day stacked by model, filters it by key, and downloads your request log as a CSV (up to 50,000 rows, no prompts or outputs).\n\n### Improvements\n\n- **A front page that shows the price and what is inside.** The signed-out page now shows every framework module as a dot you can point at, a calculator for your monthly bill (cached-input rate included), your first request and the model prices.\n- **The rest of the console in your language.** Auto-reload, the monthly limit, email alerts, invoices and the Keys page are now translated in all 11 console languages.\n- **Billing and Get started in your language.** The balance, Add credits, the buy box, Get started, the Overview figures and the Billing rows are now translated in all 11 console languages.\n- **Your balance and Add credits on every page.** The console now shows your API balance and an Add credits button on every page, so topping up is always one click away.\n- **A simpler start.** The first page walks you through three steps: create a key, make your first call, add credits.\n- **Faster top-ups.** Pick $20, $50, $100 or any amount from $20 and pay in one click; you can save your card for auto-reload in the same step.\n- **Billing in one place.** The Credits page is now Billing: auto-reload, monthly limit, low-balance email and invoices sit in one short list you open when you need them.\n- **One name.** The console is now called ISMS Copilot API platform everywhere.\n- **Models as one table, API first on Pricing.** The Models page is now one table with each model's alias, what it runs on and its price, and the public Pricing page starts with the API rates and a Get an API key button.\n- **The API knows today's date.** The published system prompt (now `server-prompt-v4`) tells the model the current UTC date, so answers about deadlines and transition periods use today, as chat already does.\n- **Clearer console for API sign-ups.** After signup the console shows where the confirmation link went with a Resend button; Models lists each model once with its 128K context and a default rate limit of 120 requests per minute per key, enforced per API server.\n\n### Fixes\n\n- **Transaction amounts always visible on phones.** In languages with long dates, the amount column of Recent transactions no longer slides out of view.\n- **Copy buttons tell you when copying fails.** If your browser blocks the clipboard, the console now says so instead of doing nothing.\n- **The console works on a phone.** On small screens the sidebar becomes a top bar with a menu, so every console page fits the screen.\n- **Email sign-up from the home page.** Typing your email on the home page now opens account creation with your address filled in, instead of a sign-in form.\n- **The new key window fits.** The window that shows a new API key no longer cuts off the key and its buttons on the right.\n\n## October 2026\n\n### Features\n\n- **Pick a framework in the model name.** Write `isms-fast:NIS_2_DE` to pin a framework from tools that only let you set a model, such as gateways and no-code connectors, at the same price as the model alone.\n\n### Improvements\n\n- **Clearer API key errors for agents.** A request with no API key or a wrong one now says which, with a link to get a key, and a brief outage on our side no longer looks like a bad key.\n\n- **Your first request, filled in.** After you create an API key, the console shows ready-to-run curl, Python and TypeScript examples with your key, and eligible new API accounts get $2 of free credit while the offer runs.\n- **A clear answer when your API balance is $0.** A call with no credit now says your balance is $0, names the $20 minimum top-up and where to add credit, and tells the OpenAI SDKs not to retry, with the same `insufficient_quota` code.\n\n- **Two-step verification at sign-in.** If you turned on two-step verification in ISMS Copilot, the console now asks for your authenticator code too, or a recovery code.\n\n- **The API persona names China's frameworks.** The published server prompt, now version 2, lists PIPL, the revised Cybersecurity Law, the Data Security Law and MLPS 2.0 alongside the frameworks it already named.\n\n- **Shorter API persona.** The published server prompt, now version 3, returns to the shorter framework list; framework coverage is unchanged.\n\n### Fixes\n\n- **Signing out of the console keeps other sessions connected.** Sign out now ends the session on that browser only, so other devices and connected AI tools keep working until they sign out themselves.\n\n- **Ask the API in Japanese, Chinese or Korean, get the answer in that language.** `isms-mini` and the GLM-backed `isms-fast` and `isms-thinking` no longer tell the model to avoid those scripts when you write in them.\n\n### Privacy & legal\n\n- **An optional question when you create an API key.** The Create API key dialog asks how you heard about us, from a fixed list; skip it and nothing is stored.\n- **Support chat loads only when you open it.** The console no longer loads the Chatbase support chat, or its cookie, when a page opens; it loads when you click the chat button.\n- **isms-mini runs outside the EU; the legal pages now say so.** Since 1 September `isms-mini` has run on GLM-5.3-Flash through OpenRouter (US), and the API legal pages now say so; use the `-eu` aliases for EU processing.\n- **The EU API aliases now run GLM 5.3 on Mistral in the EU.** `isms-fast-eu` and `isms-thinking-eu` move to GLM 5.3 before Mistral retires GLM 5.2 on 31 October: same EU endpoint, same $2.80 / $8.80 rate. `isms-fast-eu` now reasons lightly, like `isms-fast`, billed as output tokens.\n\n## September 2026\n\n### Improvements\n\n- **The Models page names the model behind each alias.** Every alias card now shows the serving model family: GLM 5.3 on the global fast and thinking aliases, Mistral (EU) on the -eu aliases, and GLM-5.3-Flash on `isms-mini`.\n- **Global API aliases move to GLM 5.3, mini to Flash.** `isms-fast` and `isms-thinking` stay at $2.80 / $8.80 with light/deeper reasoning, and `isms-mini` stays at $0.50 / $2.00 on GLM-5.3-Flash.\n\n## August 2026\n\n### Improvements\n\n- **The Models page groups Global, EU, and bulk lanes.** Aliases sit under section headers, and isms-mini is a full-width bulk-lane banner instead of a fifth card.\n- **API always-on.** The API no longer sleeps between requests: routine first calls after a quiet period no longer wait for a cold start while the server wakes up.\n\n### Features\n\n- **NSW CSP and Victorian VPDSS in the API catalog.** Completions can inject those two public-law modules, and `GET /v1/frameworks` lists them.\n\n- **isms-mini: a low-cost lane for bulk work.** A new model alias at $0.50 per million input tokens and $2.00 per million output tokens, processed in the EU with zero retention, built for formatting, extraction, and classification rather than analysis.\n- **Framework verification dates, live.** The public framework catalog (`GET /v1/frameworks`) now shows when each knowledge module was last checked against its official publisher, and a new public changelog feed (`GET /v1/frameworks/changelog`) records registry additions, edition updates, and verification refreshes.\n- **New console languages: Japanese and Chinese.** The Platform console now offers Japanese and Chinese alongside the existing languages, picked up from your browser or chosen in Settings.\n- **Continue with GitHub on the Platform console.** Create or sign in at platform.ismscopilot.com with GitHub, the same way as Google or Microsoft (chat login is unchanged).\n- **One-time USD 10 API welcome credit for paid plans.** Every paying ISMS Copilot subscriber gets a one-time USD 10 API credit so you can try the API without buying a top-up first.\n- **Growth+ widgets hide Powered by.** Active Growth, Scale, and Enterprise embeds no longer show the Powered by footer, while the on-widget AI disclosure stays.\n- **The API now speaks as a compliance assistant, with its full system prompt published.** By default, each completion carries server prompt v1: a compact compliance persona (\"a compliance AI assistant powered by ISMS Copilot\") plus the Reference Integrity intellectual-property policy, published word for word at [docs.ismscopilot.com/docs/api/system-prompt](https://docs.ismscopilot.com/docs/api/system-prompt).\n- **Per-request control over the prompt and knowledge.** `ismscopilot: { \"system\": \"none\" }` removes the persona; `\"frameworks\": \"auto\" | \"none\" | [\"ISO_27001\", ...]` selects exactly which knowledge modules are injected, or none.\n- **Responses disclose injected knowledge.** The `x-isms-frameworks` header and an `ismscopilot` response object list injected modules with a labelled token estimate; `x-isms-policy-version` and `x-isms-policy-applied` report the prompt contract that applied.\n- **Public framework catalog.** `GET /v1/frameworks` enumerates every knowledge module the API can inject.\n- **COBIT 2019 over the API.** Completions can inject COBIT 2019 knowledge when a request matches that framework.\n\n### Improvements\n\n- **One price for every GLM 5.2 alias.** Fast and thinking, global and EU, now share a single rate of $2.80 per million input tokens and $8.80 per million output tokens, down from $4.20/$13.20 on the global tier.\n- **Oversized requests fail fast.** Prompts that would exceed the model input budget are rejected with `context_length_exceeded` before you are charged, instead of an opaque upstream error.\n\n### Fixes\n\n- **API credit top-ups now land on your balance.** A paid Stripe top-up credits the API balance as soon as checkout completes.\n- **API top-up returns to the Platform Credits page.** Stripe no longer sends you to chat.ismscopilot.com after paying.\n\n### Privacy & legal\n\n- **Partner terms cover Growth+ branding and hosted white-label.** The embed terms now describe the hidden Powered by footer on Growth+, the always-on AI disclosure, and the support split when a partner hosts the assistant under their own brand.\n- **Retention stays minimal.** Request content is still never stored; we additionally record only aggregate injection mode and size metadata for billing transparency, never which frameworks a request touched.\n\n## July 2026\n\n### Features\n\n- **API Zero Data Retention (ZDR) on the console.** Platform API pages show that prompts and outputs are not stored, with only usage metadata kept for billing.\n- **Partner embed speaks as your product.** Visitors hear your bot name and company, not Better ISMS, while Powered by ISMS Copilot stays on the widget.\n- **EU processing for embed.** Growth partners can enable EU processing so the embed uses the Advanced Data Protection (Mistral EU) path.\n- **Set your embed assistant name.** Partners set the public assistant brand in the embed console; the widget uses the server name and cannot be spoofed by the snippet title.\n- **Embed layout modes and a cleaner install snippet.** Partners can choose floating or inline layout, destroy the widget cleanly, and build the install snippet from the Assistants console.\n- **Platform console: Dashboard, product lanes, and Models page.** The developer Platform opens on a Dashboard with Assistants, API keys and models, and product lanes after sign-in.\n- **API developer console: clearer keys, models, and first request.** Keys leads with create and a first-request sample; Models lists the four API aliases with honest notes.\n- **A developer front door at platform.ismscopilot.com.** A pre-login landing page with API and embed quickstarts and published pricing; general API signup is not yet open.\n- **Assistants console: try chat first, then install the embed.** Platform Assistants opens on the playground; embed install expands under it instead of burying the snippet.\n- **Dark mode on the Platform console.** platform.ismscopilot.com follows system appearance with a sidebar System/Light/Dark toggle.\n\n\n\n### Improvements\n\n- **Models page explains why Fast and Thinking share a price.** Platform Models notes that each path uses one unit price; Thinking usually costs more because it uses more tokens.\n- **Models page uses denser cards.** Platform Models shows aliases in a compact card grid with honest price and capability rows.\n- **Allowed domains prefill https://.** The embed allowed-domain field starts with the scheme so partners paste a host without inventing a protocol.\n\n### Privacy & legal\n\n- **API Zero Data Retention for request content.** Completions do not store prompts or outputs in Supabase (usage metadata only; OpenRouter and Mistral stay ZDR); see [docs](https://docs.ismscopilot.com/docs/api/zero-data-retention) and the [API DPA](https://trust.ismscopilot.com/api/dpa).\n- **AI transparency link on the embed disclosure.** The partner embed footer links to the Trust Center AI transparency page next to privacy.\n- **Optional marketing email consent at Platform signup.** Sign-up can record marketing consent separately from product attribution for Platform users.\n\n### Fixes\n\n- **API thinking models keep a minimum output budget.** Very low max_tokens on isms-thinking no longer returns empty replies after reasoning.\n- **Free EU processing opens the upgrade path.** Choosing EU processing on Free now routes partners to upgrade instead of a dead control.\n","byteLength":15363}