What changed
in Sluis.

Every release, in plain language: what is new, what improved, and what we removed.

11 October 2026

GatewayBilling

Batches at half price, sharper prompt-cache billing

Send thousands of chat requests as one batch with your data protection applied to every line, and pay half the live price.

New
Batch API
Upload a JSONL file with purpose=batch, then create, check and cancel it on /v1/batches, exactly as with the OpenAI SDK. Every line clears your data protection and residency policy before it leaves, results arrive within 24 hours, and each answered line is billed at half the model's live price. Works on Mistral, Azure, OpenAI and Qwen, on your own key or the managed one.
Fixed
Prompt-cache billing and opt-out
One-hour cache writes on Claude are billed at their own rate, Moonshot cache hits get their discount, xAI cached tokens use xAI's own price, and OpenAI's prompt_cache_retention now respects your organisation's provider prompt caching switch, including on Workspace turns.

9 October 2026

Workspace

Swipe back and a steadier layout

The Workspace apps now follow your device's back gestures, and every Workspace page uses the same width.

New
Back gestures in the apps
On Android the back gesture and button, on iPhone and iPad a swipe from the left edge, and on a Mac a two-finger swipe go back one step. An open menu or sheet closes first, sheets can be dragged down to close, and a recording keeps running.
Improved
One page width everywhere
Chat, Documents and Recordings now keep the same distance from the menu as Agents, Skills and Knowledge, and grow with larger screens in the same way.

8 October 2026

WorkspaceConsoleGatewaySecurityBilling

Speaker names and longer recordings

Recordings now fill in speaker names from the conversation and can run for up to 10 hours, and documents show tables. The updated privacy policy (version 18) covers our customer records and getting-started emails.

Improved
Recordings up to 10 hours
A Workspace recording can now be up to 10 hours long. An upload that did not arrive in full is refused instead of being transcribed in part, an interrupted upload continues where it stopped, and long meetings keep their ending in the minutes.
New
Speaker names come from the conversation
When someone introduces themselves or is addressed by name, Sluis fills in that speaker's name after transcribing. Names you typed yourself are never changed.
Changed
Minutes wait for the speakers' names
A document written from a recording now waits while a speaker has no name, so it never says "Speaker 2". Name the speakers and it is written; or choose Write anyway without names.
New
Recordings in the menu, a template or agent with AI
Recordings has its own place in the menu again. With no template of your own yet, the template list offers to create one with AI, or an agent that acts on every new recording: a chat opens and Sluis starts by asking what you need.
Changed
Privacy policy: our customer records
For every organisation account whose owner verified their email address, we now keep a customer record in our CRM, Attio: contact details, plan, a usage summary and the campaign that brought you. The privacy policy (version 18) describes what is stored, why, who receives it and for how long, so you are asked to accept it once.
Changed
Getting-started emails are covered
Section 2 lists account setup and getting-started emails next to verification, security and invoice emails, section 5 says each one has a link to stop them, and section 4 notes that we keep the date each was sent. Section 3 and section 4 also describe more exactly how our CRM enriches and deletes customer records.
Removed
The sluis/typed model is no longer offered
The plain sluis/typed model, which answered your own questions on Sluis infrastructure, is no longer offered. Nothing else changes: sluis/typed/router keeps working. For typed decisions with your own questions, connect your own TypeSafe key and use a typesafe/jev-* model.
New
Rename your organisation
Owners and admins can now change the organisation's name under Organisations in the console. The new name shows straight away in the organisation switcher and in invitations.
Fixed
Long prompts billed at the long-context rate
Models that publish a higher price for long prompts, such as Claude Haiku 5.5, the GPT-5 family, Gemini Pro and Grok, are now billed at that long-context rate for the whole call as soon as the prompt, cached tokens included, passes the model’s threshold. Until now these calls were billed at the standard rate, which was too low. Past calls are not billed again.
Improved
Long-context prices in the Models table
The Models table now shows each model’s long-context price on a second line under its normal input and output price, with a tooltip that explains when it applies.
Fixed
Claude and Gemini on Vertex at the EU regional rate
Claude models and the newer Gemini models on Google Vertex are now billed at Google’s EU regional rate, which is 10% above the global price, instead of the global price. Past calls are not billed again.
Changed
Anthropic long-context beta on Sluis-managed keys
On a Sluis-managed key, the Anthropic long-context beta header is now accepted for models Sluis prices at a long-context rate, and the call is billed at that rate. For a model without a long-context price it is still refused with a 400 error.
Changed
Status page on status.sluis.ai
The status of our services now lives on status.sluis.ai, hosted outside our own platform so it stays reachable even during an outage. It shows the API, the Console and Workspace, and you can subscribe to updates by email. Links to sluis.ai/status take you there.

7 October 2026

AppsWorkspaceSecurity

Recordings in the apps

Version 0.3.3 of the Sluis Workspace apps records meetings on your phone and computer, with a live waveform while you record. On sluis.ai, page views and loading speed are now measured with Google Analytics, only if you accept.

New
Record on iPhone, iPad and Android
In Documents, the + button starts or uploads a recording. Recording keeps going when you lock the screen or switch apps: iPhone shows it on the Lock Screen and in the Dynamic Island, Android in a notification where you can pause or stop it.
Fixed
The waveform moves while the apps record
In the Mac, Windows, iPhone and Android apps the waveform stayed flat during a recording. It now follows the sound being recorded, also for a call.
Fixed
Tables in documents, dates on chats
Tables in a document, such as the action items in minutes, now show as a real table on screen and in Word and PDF downloads, in your house style. The chat list shows when each chat was last active, like recordings do.
Changed
Page and speed measurement on sluis.ai
With your consent, the website measures which pages are visited and how fast they load, using Google Analytics without Google signals, with data stored in the EU for 14 months. Nothing is measured before you accept, and the Console, Sluis Workspace and the API are not measured. The Privacy Policy (version 16) describes it, so the cookie banner asks again once.

6 October 2026

SecurityWorkspaceGatewayConsoleApps

Meeting recordings, stronger security

The Workspace now records and transcribes meetings, the Mac app records your calls by itself, and you can share work with specific colleagues. Sluis is also harder to attack and easier to audit. Sessions are safer, uploads are checked for malware, connectors ask before they change data, API keys can expire and workloads can be limited to the endpoints they need. Name detection no longer re-reads the whole conversation on every turn, so long chats start answering sooner.

New
Record meetings and get a transcript
Under Documents, New, you can record a conversation in the Workspace or upload an audio file. Sluis transcribes it on its own servers in the EU, tells the speakers apart and gives the recording a title. Name the speakers, ask about the recording in Chat, download the transcript, and get minutes or any other document written from a template you choose.
New
The Mac app records your calls
When a Teams, Slack, Zoom or Google Meet call starts, the Sluis Workspace app for Mac records your microphone and the sound of the call, also on a headset. It stops shortly after the call ends and shows a recording icon in the menu bar, where Don't record throws the recording away. You can turn automatic recording off in Settings, and your organisation can switch it off for everyone. The Mac App Store version asks first, until you turn automatic recording on. Recording the sound of the call needs macOS 14.4 or later.
New
Agents start on new recordings
An agent can run by itself on every new transcript, for example to turn a call about a job into a quote. Switch it on in the agent, or describe the agent to the Workspace assistant and it sets this up when it creates the agent.
New
Share with specific colleagues
Chats, documents, recordings, agents, Knowledge and Skills can now be shared with the colleagues you choose, each with their own access level: View, Use or, where it applies, Edit. Private and Organisation work as before.
Fixed
Long answers from Claude on Amazon Bedrock no longer stop early
A request to a model on Amazon Bedrock without a maximum length stopped after about 1,000 tokens. In the Workspace this made building a template or an agent from a chat fail. Such a request now gets the model's own maximum.
Improved
Searchable dropdowns and a tidier small window
Every dropdown in the Workspace can now be searched by typing, as in the console. In a narrow window the model name in the message box shortens instead of pushing the other controls out, and a colleague you share with fits on one line with their access level.
Changed
Copilot comparison and calculator use current euro prices
The Copilot comparison, the European ChatGPT alternatives overview and the cost calculator now use Microsoft's euro prices checked on 5 October 2026: Copilot Business at €18.20 per user per month on an annual subscription (€21.84 billed monthly) and Microsoft 365 Copilot at €26. The break-even is recalculated: against the annual Business seat, Sluis costs less below about €16.50 of monthly usage per person.
New
Book a call from every guide and comparison
Guides and comparison pages now show a button to book a call with the founders, next to the text and at the top, together with our ISO 27001 certification and the customers that use Sluis.
Changed
Android app downloads directly
Until the Google Play listing is public, the Android app downloads straight from sluis.ai/download.
Improved
Long conversations with name detection are faster
When a conversation is sent again with a new message, name detection only inspects what is new. Messages inspected in the last few minutes are not inspected again, and the same goes for a document you send twice. Every message is still checked, and only the positions of what was found are remembered, for a few minutes, never the text.
Fixed
The acceptance screen shows the Privacy Policy version
When we published a new Privacy Policy, the screen asking you to accept it showed only the version and date of the Terms of Service. It now shows the version and effective date of both documents.
Fixed
Citizen service numbers written with spaces or dots are now recognised
A Dutch citizen service number (BSN) written in groups, such as 123 456 782, 123.456.782 or 1234.56.782, was not recognised and could reach the model unprotected. It is now found and protected like a number written without separators. Phone numbers and amounts that are not valid service numbers are left alone.
Fixed
Asking for a model that does not exist now answers "model not found"
Asking for a model that does not exist, or whose provider your organisation has not connected, used to fail with a generic server error (503). It now answers 404 "model not found" and names the model you asked for, so it is clear the request has to change and not be retried.
Changed
Inspection waits instead of refusing
When inspection is busy or slow, such as large embeddings batches or long chats that require Deep name detection, the request now waits its turn instead of being refused with "name recognition unavailable". It can take longer, and it only ends if you disconnect. An inspection that genuinely fails still refuses with the same error.
Improved
Faster Deep name detection
Deep name detection scans text faster, and capacity now scales up with demand, so busy periods cause shorter waits.
Changed
Safer sessions: sign in once more
Your browser session is now kept in a cookie that page scripts cannot read, so you may need to sign in once after this update. Signing out now also ends the session on our side, so a copied session stops working immediately.
New
Sign out everywhere
Your profile page has a new Sign out everywhere button. It ends every browser and app session of your account at once, this one included. Use it if you lost a device or think someone else has your session.
New
Require Microsoft sign-in for your organisation
The owner can switch on the new Single sign-on card in the settings and list the Microsoft Entra directory IDs that may sign in. Members must then use a work account from a listed directory. Password, Google and Apple sign-ins are refused for that organisation, and members who came in another way are signed out on their next request. To prevent a lockout, the owner must have signed in with Microsoft from a listed directory first. Switching it on or off is recorded in the audit log. People who set up a personal second factor cannot use Microsoft sign-in until they remove it on their profile.
Improved
Open sessions close the moment access ends
The live feed in the console and live Workspace chats now close right away when a member is demoted or removed, or when their password or sessions are revoked. Before, they could stay open for up to 30 seconds.
Changed
Creating an account never reveals who is registered
Signing up now always answers with a prompt to check your email, whether or not the address is already registered. An address that already has an account gets an email saying so instead of a new account. New accounts sign in with the password they chose. When you invite someone whose account is not yet verified, they are asked to verify the address before joining, and you no longer see whether an address is verified.
Changed
Common passwords are refused
A new password, whether at signup, reset, change or invitation, needs at least 8 characters, counted as characters and not bytes, and the most common passwords are refused.
New
Uploads are checked for malware
Files that members upload are checked for malware before they are stored or read: Workspace attachments, Knowledge documents, Skill archives and canvas style files. An infected file is refused and never stored. If the check cannot run, the upload is refused too, so try again later.
Changed
Connectors ask before they change data
When a connector tool is about to change data, the Workspace shows the connector, the tool and the exact values, and the member chooses Run or Don't run. Tools that only read never ask. Every connector starts at Ask before every change, so your existing connectors begin asking after this update. Scheduled agent runs have nobody to ask, so a connector set to ask never runs a changing tool there. Set it to Never ask if you want that.
New
Admin controls for connector changes
Admins can set a connector to Never ask, which is recorded in the audit log. They can turn on Read-only tools only, so the AI model is never offered a tool that changes data in that connector, and they can list the tools that only read so those never ask.
Improved
Tool results are treated as data, not instructions
The Workspace no longer treats instructions inside tool results, such as web pages, connector output and document text, as authoritative. They are handled as untrusted data.
Fixed
Conversations can only be shared inside your organisation
Sharing a conversation with someone outside your organisation, or with a user that does not exist, is now refused as if the conversation was not found.
Improved
Safer pages and downloads
Conversation attachment downloads and canvas exports are served as downloads that browsers never sniff or cache. Console and Workspace pages are no longer cached and do not pass the page address on to other sites. Workspaces on your own domain never load marketing trackers.
New
API keys can expire
An API key can now carry an expiry date, set in the console, with the Admin API or with the command line. After that date the key is refused like an unknown key. The key list shows an Expires column with an Expired badge. When you rotate a key you can set a new expiry; by default the new key keeps the old one's.
New
Limit a workload to the endpoints it needs
A workload can be limited to specific API endpoint families, such as chat, embeddings, files, images and audio, on the new Allowed operations card. A call outside the list is refused with 403 operation_not_allowed before anything reaches a provider, and the refusal is recorded in the audit log. An empty list allows everything, so existing workloads do not change.
Improved
Revoking a key works everywhere at once
Revoking, rotating or pausing a key now takes effect on every gateway instance at once. Before, other instances could keep accepting the key for up to 30 seconds.
Changed
Model lists now also limit automatic routes
A model allow-list that names concrete models now also limits sluis/auto, managed routes and organisation aliases. They only land on listed models, also after a failover, and if none is available the call is refused with 403 and recorded in the audit log. Lists with only aliases behave as before. Refusals are now audited too when a managed alias has no model that fits the residency policy, or an alias cannot be resolved.
New
Choose whether coding-agent routes with an unverified region are allowed
A new Residency setting, Coding-agent routes with an unverified region, is on by default. Turn it off to refuse Claude Code, Codex and Cursor routes with a 403 that is recorded in the audit log.
Changed
Word documents always lose their personal metadata
When Sluis anonymises a Word document, the author, editor, company and other document properties, the authors and dates of comments and tracked changes, and revision ids are now always removed, whether or not a detector matched them.
New
Strict mode for Word documents
The Documents API has a new docx_mode of strict. It accepts tracked changes, deletes comments, removes hidden text and drops all images, charts and embedded objects.
Changed
Getting the original values back needs permission
Asking the Documents API for the mapping from original values to placeholders (include_mapping) now needs the documents_reidentify operation on the workload, or the owner or admin role in the console playground. Every release is recorded in the audit log. Without permission the call is refused with 403 reidentification_not_allowed.
Improved
Old Excel and PowerPoint files are refused by name
The Documents API now says that legacy .xls spreadsheets and .ppt presentations are not supported, instead of failing in the Word reader.
Improved
Console changes are audited in the same step
Changes to members, MCP connectors, policy, keys, workloads and admin tokens are now recorded in the audit log together with the change itself. A change that cannot be audited is refused, so the audit log has no gaps.
Changed
Your own domain needs a DNS proof
Adding a custom domain for a Workspace now requires a DNS TXT record that proves you own the domain. The Brands screen shows the record to create, and saving is refused until it is found.
Improved
Stronger platform security
Isolation between organisations is now also enforced inside the database for all organisation data. Our servers run with restricted permissions, no Kubernetes API tokens in pods and a default-deny network policy. Our API hosts and website send strict security headers, and container images are deployed by immutable digest and scanned on every build and every week.
New
Service availability
The status page shows the latest availability checks for the gateway, Workspace and inspection service. Missing measurements are shown as unknown.
Changed
A shorter cookie banner on our website
The cookie banner on our website is now one short sentence with Accept, Reject and Settings. The full list of tools, events and retention stays in the Privacy Policy (version 15). Nothing changes in what we measure: no tag loads before you accept, and Reject loads none. Everyone is asked again.

5 October 2026

GatewayWorkspaceSecurity

Wider data protection, real values in more answers

Data protection now reads the shell output, patches, searches and code that coding agents send back to the model, and audio in a chat message is either refused or clearly marked as not scanned. With pseudonymisation on, more answers show your real values: custom term labels, agents built in the Workspace, tasks handed to sub-agents and provider error messages.

Fixed
More of a coding agent's history is protected
Data protection now scans and rewrites shell commands and their output, patches, code interpreter code and logs, search queries and results, and typed computer input that coding agents send back to the model, as well as custom tool calls in chat history and the text after the cursor in code completions. Before, personal data in these parts could reach the model unchanged.
Changed
Audio in a chat message is disclosed or refused
Sluis cannot check recorded audio for personal data. With the block data protection mode, a request that carries an audio part is now refused. In every other mode it is still sent, and the response and the audit log now say that the audio was not scanned.
Fixed
Real values in more answers
A streamed answer no longer shows a placeholder for a custom term with a long label. Building an agent in the Workspace no longer fails when a name contains quotation marks. Sub-agents now see and show the real values in the task they are given. Error messages from the model provider no longer end in half a placeholder.
Changed
Custom term labels are at most 32 characters
Saving a data protection policy with a longer custom term label now returns an error. Existing longer labels keep working and are shortened in the placeholder.
Fixed
Long streamed answers finish
With pseudonymisation on, a long answer streamed from the Responses API could stop with an error at its very end. It now arrives complete, with your real values restored.
Fixed
Real values in JSON answers, Cursor and tool cards
A streamed answer you asked to return as JSON stays valid JSON when a restored value contains quotation marks. Answers through the Cursor agent no longer show a placeholder that arrived in two pieces. In the Workspace, a tool card shows your real values instead of placeholders.
Changed
Cookie banner and Privacy Policy name X among the advertising tags
On our website, the advertising tags you can accept now include the X Pixel, but only if we switch it on. It loads only after you accept, and it sees the pages you view and whether you sent a contact request, clicked to sign up, booked a call or downloaded the app. It never receives what you type in a form or your email address. You can withdraw at any time via Cookie settings in the footer, and everyone is asked again.
Changed
Cookie banner and Privacy Policy cover sign-up measurement
When you create a Sluis account, you now land on a confirmation page on our website. With your cookie consent, the advertising tags we have switched on learn that an account was created, never who you are. We also store with the new account which campaign brought you: the campaign values of the sign-up link and, only with consent, the ad click identifiers and campaign values of your first visit. You can withdraw at any time via Cookie settings in the footer, and everyone is asked again.

4 October 2026

WorkspaceGatewaySecurity

Clearer comparisons with other AI tools

Every comparison page now opens with a verdict on where Sluis fits EU organisations best. The cookie banner and Privacy Policy now also name the click to book a call and a booked call, so the banner asks again.

Improved
A verdict on every comparison
The comparisons with Copilot, ChatGPT Enterprise, Claude, Gemini, Mistral, Langdock and the AI gateways now lead with a verdict and explain where Sluis stands out: ease of use, a clear overview for admins, data security and what you can do in the workspace. Features that are not part of the standard product are listed as available on request.
Changed
Cookie banner and Privacy Policy name the click to book a call and a booked call
On our landing pages, the advertising tags you accepted (Google, Meta, Reddit and LinkedIn, only the ones in use) now also learn that the button to book a call was clicked and that a call was booked, never what you type or your email address. The cookie banner asks for your choice again and the Privacy Policy lists the event. Declining or withdrawing consent still stops the tags.

3 October 2026

WorkspaceGatewayConsoleBilling

Apps sign in again, consent-gated advertising measurement

Version 0.3.2 of the Sluis Workspace apps fixes sign-in on every platform, and new accounts now state the organisation they are for. On sluis.ai you can compare Sluis with twelve other AI workspaces and gateways, read all five guides in your own language, follow web addresses in your language, and load pages faster. Advertising measurement with Google, Meta, Reddit and LinkedIn loads only after you accept cookies, so the cookie banner asks again, and everyone accepts an updated Privacy Policy once.

Fixed
The apps sign in again
Since 30 September the desktop and mobile apps could not sign in or load your conversations. Apps you already installed work again, and version 0.3.2 for iOS, iPadOS, Android, macOS, Windows and Linux connects to the right address. On iPhone and iPad, Sign in with Apple always sits next to Google.
Fixed
sluis/typed answers again
After a few long requests, sluis/typed could keep answering "decision engine busy" even when nothing else was running. It now runs every request whenever the engine is free, and turns away far fewer requests that arrive while another one is running.
Fixed
Streamed answers no longer show placeholders
With pseudonymisation on, a streamed answer could reach your application with a placeholder such as «PERSON_NAME_3» instead of the real value. This happened in tool call arguments, in Responses API events and in Anthropic Messages events. Every streamed field now gets its real values back, and tool call arguments stay valid JSON. Placeholders that already reached a conversation stay as they are.
Changed
New accounts are for organisations
Sluis is sold to businesses and organisations. Creating an account now asks for your organisation’s name and a confirmation that you sign up on its behalf, also when you continue with Google, Apple or Microsoft. Existing accounts and invited members are not affected.
New
Compare Sluis with other AI tools
New comparison pages set Sluis next to Microsoft 365 Copilot, ChatGPT Enterprise, Claude, Gemini, Mistral, Langdock, LiteLLM, Portkey, Kong, Cloudflare, OpenRouter and Azure API Management, plus two overviews for European teams. Each page lists its sources, the date we checked them, and when the other product is the better choice.
Improved
All guides in seven languages
The five guides on the AI Act, personal data in prompts, CLOUD Act exposure, provider due diligence and choosing an EU provider now read in Dutch, French, German, Spanish, Italian and Polish too, including the Markdown download.
Changed
Web addresses in your language
Pages in other languages now use translated addresses, such as /nl/prijzen or /de/preise. Old links keep working and forward to the new address.
Improved
A lighter website and console
sluis.ai and the console download about a quarter less code before they respond, so pages become interactive sooner, especially on mobile connections.
Changed
Cookie banner and Privacy Policy name each advertising vendor
Google, Meta, Reddit and LinkedIn tags on the marketing website load only if you accept, and the banner names only the ones in use. They see the pages you view and whether you sent a contact request or clicked sign up or a download, never what you type in a form. Declining or withdrawing stops them and removes what they stored on sluis.ai. The Console, Workspace and API stay free of advertising trackers. The Privacy Policy now lists each vendor, its purpose and the basis for the transfer to the United States.
Improved
Smoother video player in the docs
The play button on the docs videos now sits in the centre, and while a video loads, buffers or switches quality you see a loading animation instead of a frozen frame.
New
Say hi, get €50 of free tokens
Until your organisation buys its starting credit, the console now tells you about the offer and links straight to hello@sluis.ai.

1 October 2026

AppsConsoleWorkspaceGatewaySecurity

Apps sign in again, consent-gated advertising measurement

Fixed
The apps sign in again
Since 30 September the desktop and mobile apps could not sign in or load your conversations. Apps you already installed work again, and version 0.3.2 for iOS, iPadOS, Android, macOS, Windows and Linux connects to the right address. On iPhone and iPad, Sign in with Apple always sits next to Google.
Fixed
sluis/typed answers again
After a few long requests, sluis/typed could keep answering "decision engine busy" even when nothing else was running. It now runs every request whenever the engine is free, and turns away far fewer requests that arrive while another one is running.
Fixed
Streamed answers no longer show placeholders
With pseudonymisation on, a streamed answer could reach your application with a placeholder such as «PERSON_NAME_3» instead of the real value. This happened in tool call arguments, in Responses API events and in Anthropic Messages events. Every streamed field now gets its real values back, and tool call arguments stay valid JSON. Placeholders that already reached a conversation stay as they are.
Changed
New accounts are for organisations
Sluis is sold to businesses and organisations. Creating an account now asks for your organisation’s name and a confirmation that you sign up on its behalf, also when you continue with Google, Apple or Microsoft. Existing accounts and invited members are not affected.
New
Compare Sluis with other AI tools
New comparison pages set Sluis next to Microsoft 365 Copilot, ChatGPT Enterprise, Claude, Gemini, Mistral, Langdock, LiteLLM, Portkey, Kong, Cloudflare, OpenRouter and Azure API Management, plus two overviews for European teams. Each page lists its sources, the date we checked them, and when the other product is the better choice.
Improved
All guides in seven languages
The five guides on the AI Act, personal data in prompts, CLOUD Act exposure, provider due diligence and choosing an EU provider now read in Dutch, French, German, Spanish, Italian and Polish too, including the Markdown download.
Changed
Web addresses in your language
Pages in other languages now use translated addresses, such as /nl/prijzen or /de/preise. Old links keep working and forward to the new address.
Improved
A lighter website and console
sluis.ai and the console download about a quarter less code before they respond, so pages become interactive sooner, especially on mobile connections.
Changed
Cookie banner and Privacy Policy name each advertising vendor
Google, Meta, Reddit and LinkedIn tags on the marketing website load only if you accept, and the banner names only the ones in use. They see the pages you view and whether you sent a contact request or clicked sign up or a download, never what you type in a form. Declining or withdrawing stops them and removes what they stored on sluis.ai. The Console, Workspace and API stay free of advertising trackers. The Privacy Policy now lists each vendor, its purpose and the basis for the transfer to the United States.
Improved
Smoother video player in the docs
The play button on the docs videos now sits in the centre, and while a video loads, buffers or switches quality you see a loading animation instead of a frozen frame.

30 September 2026

WorkspaceConsoleGatewaySecurityBilling

A cleaner audit log, tighter checks

Choose how deeply the model thinks in each Workspace chat, filter the audit log to Workspace activity, read small costs in full, pay list price for image generation, watch product films in the docs, and rely on images, cached answers and agency two-factor rules behaving as set. Owners and admins also accept a clearer Data Processing Agreement once. Everyone also accepts an updated Privacy Policy once, which now says which Sluis apps you use.

New
Choose a reasoning level per chat
Next to the model in the composer you can now set how much the model thinks before it answers: Off, Low, Medium or High. Off is fastest, High is most thorough. Each chat remembers its choice. Leave it on Automatic and Sluis decides per question, as before. The control is greyed out for models that cannot reason.
Fixed
Image generation billed at list price
OpenAI image models such as gpt-image-2 and gpt-image-2.5 were billed far above their list price. They now bill text and image tokens at OpenAI's own rates and appear in the model list. Where a request was overcharged, the audit log shows a price correction under it and the difference is credited to your spend and budgets.
Changed
Fair pricing for models without a known price
A model whose price we cannot yet confirm is billed at the typical price of the same provider's comparable models, never at the most expensive model we know. We are alerted at once and add its real price. We now also check prices against two independent sources after every update.
New
Videos in the docs
A new Videos page in the docs shows Sluis in real use: how it works, a day with Workspace, an agent going live on Gateway, and full tours of the console and of Workspace. Every film has a voice-over and captions in your language, plays in 1080p or 1440p, and can be downloaded.
Changed
Privacy Policy updated: which Sluis apps you use
Our Privacy Policy now says that Sluis records which app you open (web Console, web Workspace, macOS, Windows, Linux, iOS or Android), on which day, and its version. 7Lab uses this to support customers and improve Sluis, keeps it for 12 months, and deletes it with your organisation. It is not shared with anyone else. You are asked to accept the new policy once.
Improved
Workspace filter in the audit log
A new Workspace filter shows only the requests made from Workspace chats. The cost column is wider, so amounts below a cent are no longer cut off, and the page drops two banners that repeated information shown elsewhere.
Fixed
Images follow your document inspection
Images sent in a request now get the same checks as documents. With document inspection on, PNG and JPEG images are scanned, and personal details in them are masked or replaced according to your settings. With the block setting, an image that cannot be checked is refused; otherwise it is sent on and marked as unscanned in the audit log and in the response.
Fixed
Cached answers show your own details
With pseudonymisation and the answer cache on, an answer served from the cache could show a placeholder such as «EMAIL_1» instead of the detail you sent. Cached answers now show your own details, just like fresh ones.
Fixed
Agency two-factor rules reach client organisations
When an agency requires two-factor sign-in, its owners and admins now need it in the client organisations they manage as well, not only in the agency itself.
Changed
A clearer Data Processing Agreement
Version 8 of the Data Processing Agreement spells out who is responsible for account data, how long cached answers, Workspace chats and the names in them are kept, and that the mapping behind pseudonymized names is never stored outside Workspace. Owners and admins are asked to accept it once.

29 September 2026

WorkspaceConsoleGatewayBilling

Leaner Workspace, clearer costs

Workspace features are on by default, requests carry only what they need, and the audit log shows cost per token type.

Changed
Workspace features on by default
Documents, Skills, Agents, Knowledge, Chat and the Agent Harness are now on for every organisation, including ones that had switched them off. Owners and admins can still switch each one off under Settings › Features. Compliance as code stays off until an owner turns it on.
Improved
Faster, leaner Workspace answers
Workspace now sends a model only what your request needs. A plain question or a poem carries the everyday tools; document tools come along when you ask for a document or the chat already has one, and your connected apps are loaded only when the assistant needs one. A simple request is now about a tenth of its former size, so answers start sooner and cost fewer tokens. Every permission check still applies.
New
Cost per token type in the audit log
Open a request in the audit log to see input, cached input, cache writes and output separately: tokens, price per million and cost for each, plus any other usage and the total. Requests from before this release show tokens and the total only.
Fixed
Emails readable in dark mode
Emails from Sluis no longer show white text on white in mail apps that use a dark theme, such as Apple Mail, Outlook and the Gmail app.
Changed
System One hosted by Sluis
System One now appears in the model overview as a Sluis-hosted model in the EU. Its decisions are billed at EUR 0.01 per million input tokens, with output free, instead of a flat fee per call.
New
Call the router behind sluis/auto
You can now call the same router Sluis uses for sluis/auto. Send a request's text to sluis/typed/router and you get back whether it needs step by step reasoning, how demanding it is, what kind of work it is, and the route sluis/auto would pick for that text. It runs on Sluis infrastructure with your data protection settings, and existing callers keep the answer they already read.

28 September 2026

GatewayWorkspaceAppsConsoleSecurityBilling

Faster, steadier, no dropped connections

Requests, data protection and the Workspace are faster, and updates to Sluis no longer cut off answers that are still streaming. The privacy policy now states how long your organisation's content is kept and which VAT details we check.

New
Canvas and Documents
Letters, emails, slides and code now open in a canvas in the chat that you and the model both edit. Every change is a version you can restore, and each canvas is saved in your Documents, where you open it with its chat beside it. Download it as PDF, Word, PowerPoint, Markdown, text or an email file, or send an email from your own mail program. Templates and house styles keep your documents consistent; owners and admins manage the organisation's in the console. It is on for every organisation; owners and admins can switch Documents off in the console under Settings › Features.
Fixed
Inspection no longer fails on an idle connection
Requests whose policy requires name detection or prompt injection inspection were now and then refused after a quiet moment, because an internal connection had closed. They are now inspected as normal. The gateway also no longer pauses when it briefly needs more processing power.
Improved
Faster data protection
Finding personal data in a request takes a fraction of the time it did, most of all for text in European languages and for follow-up messages. What is detected has not changed.
Improved
Long answers stay connected
While a model thinks before it answers, streamed API responses now carry a small keep-alive signal every 15 seconds, so proxies and load balancers in your network no longer close the connection early. Your SSE client ignores it.
Fixed
Updates no longer cut answers short
When we update Sluis, answers already streaming keep running for up to five minutes, and one that runs longer ends with a clear error you can retry instead of stopping silently. Workspace and the console reconnect on their own right away, and a chat that was busy when a server failed no longer stays busy.
Improved
Less waiting before each request
Rate limits and budgets are now checked in a single step, and account and billing details are refreshed in the background, so busy workloads see steadier response times. Console reports can no longer slow down live traffic.
Improved
Workspace updates at once
An open conversation now shows a new title, status or sharing change as soon as it happens, and Stop ends a running answer faster.
Fixed
A brief outage no longer rejects valid API keys
When the gateway briefly cannot check a key, it now answers 503 auth_unavailable with Retry-After instead of 401, so SDKs retry instead of treating your key as invalid.
Improved
Faster usage, audit and credit figures
The usage page, the audit log, API key activity, workload spend and your prepaid balance no longer read your whole history on every visit. The figures are unchanged.
New
Cursor paging for audit, traces and operator audit
The audit, traces and operator audit endpoints return a next_cursor. Pass it back as cursor to fetch the next page at the same speed however deep you go. Offset paging keeps working.
Improved
Knowledge uploads process in the background
A document you add to Knowledge appears at once as Processing and becomes searchable when it is ready. If it cannot be processed, it is marked as failed.
Improved
Quicker answers from agents with tools
Workspace now runs several web searches at the same time within one answer, and connects to your MCP servers in parallel at the start of a turn. Results still appear in the order the agent asked for them.
Changed
Retention and VAT checks in the privacy policy
Retained prompts and responses are kept for 30 days by default, and your organisation can change that. In Workspace, personal details in chats become placeholders after 7 days by default, and chats are kept as long as your organisation chooses. A daily cleanup enforces these periods. The policy also names the VAT details we use for invoices, including the check in the EU VIES system for VAT numbers outside the Netherlands. You accept it once more on your next console visit.
Changed
New chats start on Automatic
When your organisation has not set a default model, a new Workspace chat starts on Automatic, which picks a suitable model your policy allows. A default set by an administrator still comes first, and scheduled agent runs follow the same order. Automatic leads the model picker, which still hides models your policy refuses.
Fixed
Real regions for managed models
The model picker now shows where the model behind Automatic and the other managed routes actually processes data, instead of always showing EU.
New
Agents remember how you like to work
When you correct a Workspace agent, it now remembers the correction for you and brings it into your later chats with that agent. Other members keep their own notes for the same agent. If you correct the same point again, the agent offers to change its instructions, or tells you an administrator can. You can view and edit what an agent remembers about you on its page, and your memory settings apply to it.
Improved
Memory keeps itself tidy
When Workspace remembers something new, points that say the same thing are merged and a point the new one makes untrue is removed, so your memory never lists a wish twice or both ways. What you type yourself in your memory settings is kept as written.
Improved
Managed aliases rank every model you may use
Each sluis/* alias now orders every model that fits its job, using benchmark results, price and the speed and uptime we measure ourselves. When your residency policy rules out the first choice, the alias uses the best model your policy allows. Among providers that serve the same model, EU-owned providers come first, then the healthiest, fastest and cheapest. The rankings in our public model catalog (catalog.json) now follow the same order.
Improved
Better model picks for each task
Each managed sluis/* alias now compares every model with all other models on the market. A model without a benchmark for the task ranks below an equally capable model that was measured. Fast, cheap and support tasks only use models that clear a minimum quality level, and a model released more than 18 months before the same maker's newest available model is skipped.
Fixed
The checkout total includes VAT
Prices exclude VAT. When you buy credit or set up automatic top-up, the console now lists the credit excluding VAT, the processing fee, the VAT at your rate and the total you pay. Before, the Pay button showed the total without VAT while the payment included it. Your transactions show the credit you bought, with the VAT next to the fee.
Changed
Calls interrupted after sending are billed
When a connection closes, a Workspace turn ends early or a call fails unexpectedly after the model provider received the request, the audit log now bills an estimate: the prompt plus any output already streamed, or the images requested. It is marked as estimated and counts once against budgets, prepaid credit and member caps. Calls that end before they are sent still cost nothing.
Removed
Agent Word and PowerPoint templates
Agents no longer take an uploaded Word or PowerPoint template for the documents they create.
Improved
Version 0.3.1 on every platform
Sluis Workspace 0.3.1 is out for Mac, Windows, Linux, iPhone, iPad and Android. The store versions arrive as Apple and Google approve them.
Fixed
Workspace fits phones and narrow windows
On a phone, the agent cards on a new chat stay within the screen. In mid-sized windows, search, the composer and the side column are no longer cut off.
Changed
The desktop sign-in window only signs you in
The window the desktop app opens to sign in with Apple, Google, Microsoft or a passkey no longer offers to create a new account. Your administrator gives you access.
Improved
sluis/auto picks the right model faster
Sluis now reads each request itself, in all seven languages, and sends code, translations, extraction, support replies and hard reasoning to a model suited for the job. The choice takes a fraction of a second and no text leaves Sluis to make it.
Improved
A clear retry instead of lost work under heavy load
When Sluis is very busy, it now asks your client to try again, with a Retry-After header that says when, instead of taking on a request it cannot fully record. Every request Sluis answers lands in your audit log and usage, also at peak load. Structured decisions on sluis/typed answer the same way when the local engine is busy.

25 September 2026

WorkspaceGatewayConsoleAppsBillingSecurity

Sluis Workspace, redesigned

A redesigned Workspace, new ways to sign in securely, retention you choose, and more control over MCP tools, workloads and the audit log.

New
A redesigned Sluis Workspace
Build agents and Skills by describing them in chat, and change them later the same way. Workspace can remember what you share, saves your work as you go, starts a fresh chat when you open Chat and lets you stop an answer while it runs.
New
Spreadsheets and presentations
Upload spreadsheets and presentations in chat and Knowledge, and have agents create new ones for you.
New
Scheduled agents and sub-agents
An agent you create in chat can be scheduled straight away, as often as every 15 minutes. Agents can hand work to sub-agents, which administrators control for the organisation.
New
Continue with Microsoft
Sign in with your Microsoft work or school account, next to Google and Apple.
New
Two-factor with an authenticator app
Use an app such as Google Authenticator or 1Password as your second factor, and require it for the whole organisation if you like.
New
Your sign-in methods and password
Profile lists every way you sign in and when each was last used, and signing in with Google, Apple or Microsoft no longer replaces another one. You can change your password there too, which signs you out everywhere else.
New
Choose how long data is kept
Set how long request content is kept under Data protection, 30 days by default, and how long personal details in chats and whole conversations are kept in Workspace settings. New periods also apply to what is already stored.
New
MCP tools with your own account
Members connect MCP tools with their own sign-in, and can request a private server that an administrator approves.
New
Decide what MCP servers may see
Mark each MCP server as trusted, ask permission or not trusted. A trusted server receives the real values behind masked details in tool calls; with ask permission, members approve that in the chat first. The AI model still sees only placeholders, and servers you already had stay not trusted until you change them.
New
Connect MCP clients by signing in
MCP clients reach your Sluis agents once you sign in and approve, with no key to copy. Revoke a connected client at any time.
New
Scope guard and answer timeout per workload
Describe what a workload is for; a small model checks each request against it and blocks or logs what falls outside. You also set how long it waits for a provider’s answer.
New
Follow requests live in the audit log
Watch a request move through inspection, sending and the streamed answer, including any retries, until it is sealed. The Failed filter shows calls that went wrong and why.
New
Sluis Workspace 0.3.0 apps
The iPhone, iPad, Android, Mac, Windows and Linux apps are now called Sluis Workspace and bring the redesigned Workspace. On iPhone, iPad and Android you can sign in with Apple, Google or Microsoft right in the app.
New
A notification when your answer is ready
The apps for iPhone, iPad, Android and Mac tell you when an answer finishes while you are away. The notification shows the conversation name and nothing else, and signing out stops notifications for that account on the device. With notifications on, an answer keeps running after you close the app, so the notification still arrives.
New
Cookie choices on sluis.ai
Our website loads analytics only after you accept, and Cookie settings in the footer lets you change your mind at any time. The updated privacy policy explains this; you accept it once more on your next console visit.
New
Practical guides
Five guides for the people who sign off on AI, from the AI Act to personal data in prompts, at sluis.ai/guides.
New
Workspace storage overview
Owners and administrators see how much storage the organisation uses for documents, attachments and Skills, per member, and can remove a member’s private documents. Each organisation has a storage limit.
New
Typed decisions answered in Sluis
The sluis/typed model now answers your own yes/no, scale and choice questions on Sluis infrastructure, with no outside provider, so it works under the default EU-only policy. TypeSafe stays available as a provider with your own key.
New
A bell for what’s new
A bell at the top right of the console and Workspace shows a dot when there is a new release and opens these release notes in your language.
Improved
An audit log grouped per request
Every step of a request sits together, including each data protection check and every tool call next to its result.
Improved
Stronger redaction
Signed, handwritten and filled-in scans are redacted properly, and name recognition covers up to 100,000 names per request. Web search results reach the model as published.
Improved
Release notes and a friendlier website
These release notes are new, in seven languages. A broken link now suggests where to go next, an API address opened in a browser takes you to the developer docs, and the pricing calculator compares €5, €10 and €15 per person per month, our fee included.
Improved
Longer lists load in pages
Conversations and documents load in pages with Load more, so large workspaces open quickly, and search still finds older items.
Changed
Name recognition billed per request
AI name recognition costs €0.005 per request with Swift and €0.01 with Deep, shown under Data protection. Knowledge uploads are billed at the embedding model’s price.
Changed
Failed calls that reached the provider
When a provider times out or a streamed answer breaks off, the call is billed for the usage the provider reports, or for an estimate marked as such. Failures on our side and error responses from the provider stay free.
Changed
Provider-side features need your own key
With Sluis’s shared keys, requests that store data at the provider, use the provider’s hosted tools or ask for premium options we cannot meter are refused. Connect your own provider key to use them.
Changed
Safer sign-in and invitations
Signing in to the desktop app through your browser now asks you to confirm, so update older desktop apps. People who already use Sluis accept an emailed invitation before they join another organisation, and Microsoft sign-in explains what to fix when an account has no email address.
Changed
Sluis Chat is now Sluis Workspace
The app now lives at /workspace. Old links and bookmarks take you there automatically.
Changed
Safer agents and schedules
A schedule pauses when someone other than its creator changes the agent, until it is saved again. After a password reset your schedules pause and your personal MCP connections are removed. Memories saved from web or tool results are marked as unverified.
Changed
One address for questions
Billing, security and support questions all go to hello@sluis.ai.
Fixed
Redaction, checks and model routes
Phone numbers written in groups are redacted whole, the scope guard and security scan refuse fewer valid requests, name recognition holds up under bursts, and managed sluis/ routes are refreshed again and move to another allowed model when your key cannot use the first.
Fixed
Screening, limits and VAT
Text sent as form uploads and in other request formats is screened too, cached answers stay within their workload, spending limits count calls that are still running, VAT follows your confirmed VAT number, and a blocking scope guard refuses a request it cannot read.
Fixed
Placeholders in Workspace answers
With pseudonymisation on, an answer that used a tool or followed an earlier message could show a placeholder such as «LOCATION_4», or the wrong detail, instead of what you wrote. Answers now show your own details again.
Fixed
Workspace details
Conversation titles no longer show attachment markers, an agent no longer creates the same file over and over, answers know today’s date and no longer repeat themselves after a tool call, and Connections says a connection is not available instead of showing an error. On a phone, + now starts a new chat.
Fixed
Automatic model choice with tools
A Workspace answer on sluis/auto no longer fails when the turn can use tools: it goes to a model that accepts tools in that form.
Fixed
Storage limits, uploads and agreements
Uploads made at the same moment can no longer go past the storage limit, and publishing an unchanged Skill package again at the limit works. Tools that cost money stop once the organisation budget is used up. Agreement messages about verifying your email or choosing the signer now appear in your language.
Fixed
Sluis Edge and agency safeguards
Revoking a Sluis Edge gateway or licence now cuts its live connection within a minute, and Edge spending limits hold when many requests arrive at once. Accepting the data processing agreement for a client organisation needs a verified email, and an agency can no longer choose or change a client’s legal signer.

23 September 2026

Apps

Workspace on iPadOS 27

A focused fix for the iPad app.

Fixed
The iPad app starts on iPadOS 27
Sluis Workspace for iPad no longer closes at launch on iPadOS 27.

21 September 2026

WorkspaceConsoleAppsBilling

Sign in with Apple

Another way to sign in, sign-in screens that work first time, and fewer billing emails.

New
Sign in with Apple
Sign in or create your account with your Apple ID on the web. The desktop app offers Sign in with Apple too.
Fixed
Google sign-in and account creation
On the web, the Google button appears straight away again and creating an account works again.
Improved
Fewer credit reminders
Credit emails only go to organisations that used Sluis in the last 30 days and have paid before. The out-of-credit email stops after four in 30 days.

17 September 2026

GatewayBillingApps

A new website, Nebul and prepaid credit

A rebuilt sluis.ai, a new EU provider and a clearer way to pay.

New
Nebul provider
EU-owned, EU-hosted models from Nebul, with your own key or ours, priced in euros.
Changed
Prepaid credit
Top up from €25 and switch on automatic top-up if you like. Usage draws from your balance.
Improved
Smarter sluis/auto
Automatic routing only picks models your organisation can actually call.
New
Workspace 0.2.3 apps
The Mac, Linux and Android downloads carry the new Sluis symbol, with native fixes.
New
A new sluis.ai
The website is rebuilt, in all seven languages.

16 September 2026

ConsoleSecurity

Audit filters and a support panel

Find the requests you need faster, and reach us from any console page.

New
Filter the audit log by workload and API key
Combine them with decision, model and date. Each request keeps the workload it was admitted under.
New
Support from the console
Report a problem or request a feature from the top bar. The page you are on is attached for you.
Improved
Steadier security scanning
Two scanners share the load, and overlapping scans wait briefly for capacity instead of failing straight away.

What changedin Sluis.

Batches at half price, sharper prompt-cache billing

Batch API

Prompt-cache billing and opt-out

Swipe back and a steadier layout

Back gestures in the apps

One page width everywhere

Speaker names and longer recordings

Recordings up to 10 hours

Speaker names come from the conversation

Minutes wait for the speakers' names

Recordings in the menu, a template or agent with AI

Privacy policy: our customer records

Getting-started emails are covered

The sluis/typed model is no longer offered

Rename your organisation

Long prompts billed at the long-context rate

Long-context prices in the Models table

Claude and Gemini on Vertex at the EU regional rate

Anthropic long-context beta on Sluis-managed keys

Status page on status.sluis.ai

Recordings in the apps

Record on iPhone, iPad and Android

The waveform moves while the apps record

Tables in documents, dates on chats

Page and speed measurement on sluis.ai

Meeting recordings, stronger security

Record meetings and get a transcript

The Mac app records your calls

Agents start on new recordings

Share with specific colleagues

Long answers from Claude on Amazon Bedrock no longer stop early

Searchable dropdowns and a tidier small window

Copilot comparison and calculator use current euro prices

Book a call from every guide and comparison

Android app downloads directly

Long conversations with name detection are faster

The acceptance screen shows the Privacy Policy version

Citizen service numbers written with spaces or dots are now recognised

Asking for a model that does not exist now answers "model not found"

Inspection waits instead of refusing

Faster Deep name detection

Safer sessions: sign in once more

Sign out everywhere

Require Microsoft sign-in for your organisation

Open sessions close the moment access ends

Creating an account never reveals who is registered

Common passwords are refused

Uploads are checked for malware

Connectors ask before they change data

Admin controls for connector changes

Tool results are treated as data, not instructions

Conversations can only be shared inside your organisation

Safer pages and downloads

API keys can expire

Limit a workload to the endpoints it needs

Revoking a key works everywhere at once

Model lists now also limit automatic routes

Choose whether coding-agent routes with an unverified region are allowed

Word documents always lose their personal metadata

Strict mode for Word documents

Getting the original values back needs permission

Old Excel and PowerPoint files are refused by name

Console changes are audited in the same step

Your own domain needs a DNS proof

Stronger platform security

Service availability

A shorter cookie banner on our website

Wider data protection, real values in more answers

More of a coding agent's history is protected

Audio in a chat message is disclosed or refused

Real values in more answers

Custom term labels are at most 32 characters

Long streamed answers finish

Real values in JSON answers, Cursor and tool cards

Cookie banner and Privacy Policy name X among the advertising tags

Cookie banner and Privacy Policy cover sign-up measurement

Clearer comparisons with other AI tools

A verdict on every comparison

Cookie banner and Privacy Policy name the click to book a call and a booked call

What changed
in Sluis.