The Windows app updates through Microsoft Store; the container images and Helm chart are versioned separately. Breaking changes and required actions are also listed under compatibility.
Containers and Helm chart
Helm chart 0.2.2 — 2026-10-06
- Deploys the 2.5.2 images by default.
Container images 2.5.2 — 2026-10-06
- The server changes of the October 2026 review, the same as the Windows app 2.2.6 below: security hardening, better OpenAI API compatibility, rate-limit and log-level settings that take effect, and AI Gateway on Pro Commercial with only workers counting as licences.
- The images include
aisuite-server-cli, so API keys can be issued straight into a volume. - Each image reports its real version in the
X-AI-Server-Versionheader, and its base images are pinned by digest.
Helm chart 0.2.1 — 2026-10-05
- The licence server value now configures the setting the server reads (
license.registrationUrl; the oldlicense.serverstill works). An empty value falls back to the default. - Gateway API keys are mounted as a single file, keeping the rest of the server's credentials folder writable; the gateway gets a start-up probe.
- NetworkPolicy admits Prometheus's namespace when the ServiceMonitor is on, and the ingress controller's namespace is configurable.
- Gateway nodes no longer use a licence seat; only workers count.
Container images 2.5.1
softwaretailor/aiserver,softwaretailor/aigatewayandsoftwaretailor/aiserver-pro(Pro AI Engine A, NVIDIA GPUs).
Windows app
Windows app 2.2.6 — 2026-10-06
Safer, more compatible and fully documented
- Security: the Windows service now runs as its own low-privilege account, new API keys can no longer read server administration data unless you allow it, and keys are never written to logs.
- Your own code and tools work better: tool calling for agents, stop sequences, max_completion_tokens, named tool choice, embeddings dimensions and base64, and SRT/VTT subtitles now behave like the OpenAI API.
- Settings you choose now take effect: per-key and per-IP rate limits, log level, and a new 'Start the server when AI Server opens' option. HTTPS works everywhere in the app.
- AI Gateway is part of Pro Commercial, and only the servers behind it count as licences. The app, the service and the command line now always share one data folder.
- The server keeps its own log file, and the new documentation at softwaretailor.com/docs/ai-server covers installing, the API, security and operations.
Windows app 2.2.5 — 2026-09-26
Buy directly, fully translated Settings
- Buy Pro directly from Software Tailor: your licence is linked to your email, so it can be verified, resent and moved to a new PC, with customer-only offers, upgrade paths and self-service billing.
- Settings, subscription and upgrade screens are now fully translated, including plan names, periods, Free Pro status and every confirmation.
- More reliable subscription status, organisation sign-in and purchase links; the "why did you stop Pro?" question now only appears after a real cancellation.
Windows app 2.2.4 — 2026-09-01
Windows Service setup works again
- Fixed Install as service doing nothing after administrator approval on some PCs.
- Automatic Windows Service updates no longer fail with a misleading PowerShell parameter error.
Windows app 2.2.3 — 2026-08-31
Windows Service recovery that works
- AI Server can now replace very old Windows Service versions that did not respond to a normal stop request.
- Uninstall now waits for Windows to finish removing the service and keeps everything needed to install it again.
- If an earlier update left the service missing, Install as service remains available so you can repair it from the Server page.
Windows app 2.2.2 — 2026-08-31
Reliable shared model discovery
- AI Server now recovers automatically if its shared model list was interrupted during an earlier update.
- Model discovery is now saved safely when AI Server and several AI Suite apps are running at the same time.
Windows app 2.2.1 — 2026-08-31
Reliable Windows service updates
- Fixed an update problem that could remove the Windows Service and leave both service buttons unavailable. Affected installations can now repair the service directly from the Server page.
- AI Server now prepares replacement service files before stopping the current service and keeps its Windows registration intact if an update cannot finish.
- Before an app update asks for administrator permission, AI Server now explains that it needs to update the Windows Service and waits for your confirmation.
Windows app 2.2.0 — 2026-08-31
Faster local AI and a clearer live server view
- Use hardware-accelerated local models through AI Server on supported Windows devices. Your other AI Suite apps connect to the server instead of installing and running another copy of the local engine.
- Set up and manage the optional high-performance local engine from AI Server, with guided system checks and private on-device model storage.
- Active requests now update as work starts and finishes, show the effective client IP while the request is live, and keep recent completions for 15 minutes without storing prompts or responses.
- Local AI Suite apps can connect securely without managing an API key, service repairs prompt only when they are really needed, and the desktop title no longer labels AI Server as a preview.
Windows app 2.1.15 — 2026-08-17
The background service now stays up to date
- Fixed AI Server starting with Windows but leaving connected apps unable to use local models until you manually stopped and restarted the service.
- After an app update, AI Server now refreshes its background service automatically and keeps the same server address whenever possible.
- Future package updates are prepared automatically at Windows startup. An older service installation may ask for administrator approval once when you first open this update.
Windows app 2.1.13 — 2026-08-02
Speech-to-text now works on Arm PCs
- AI Server now comes in a version built for Arm PCs — Copilot+ PCs and other Snapdragon or Arm devices. Until now those machines installed the Intel version and ran it in compatibility mode.
- That is what broke voice features. Every request to turn speech into text failed with a server error, so voice input and dictation in any connected app quietly did nothing. Chat, embeddings and model downloads were unaffected, which made the real cause hard to see.
- There is nothing to change on your side — Windows now installs the version that matches your PC.
Windows app 2.1.12 — 2026-07-29
Better compatibility with standard AI tools and code
- Your installed embedding models now appear in the model list. Previously a model like nomic-embed-text worked but was listed nowhere, so apps and tools had no way to discover it — and AI Server looked as though it had no embedding support at all.
- Fixed embedding requests from the official OpenAI libraries. Asking for a single embedding — the most common way to call it — was rejected; it now works exactly as those libraries send it.
- Web pages can now use AI Server directly, if you let them. Browsers previously blocked all page scripts from calling the server. You can now list the web addresses you trust, and only those are allowed.
- Together these mean more third-party tools, sample code and scripts work against your server without changes.
Windows app 2.1.11 — 2026-07-28
Serving your network now works — including on "Public" Wi-Fi
- Fixed the most common reason other computers still couldn't reach your server: Windows treats any network you haven't marked as trusted as "Public", and the firewall rule AI Server created only covered Private and Domain networks — so it allowed nothing. The rule now covers the network type you're actually on, and Network diagnostics tells you when your network is Public and how to change it.
- Network diagnostics also detects firewall rules that BLOCK AI Server (Windows creates these silently if its "Allow access?" prompt is dismissed) — these override any allow rule, and the fix button now removes them. The fix button is no longer hidden when a rule only appears to be present.
- "Auto-pick a free port" works properly again for network serving. It keeps the same port between restarts instead of choosing a new one, so your firewall rule and the address other computers use stay valid.
- Verified end to end between two PCs: connections without an API key are refused, and with a valid key model listing and AI generation work across the network.
Windows app 2.1.10 — 2026-07-28
Trustworthy network diagnostics, a port that stays put, and a security hardening pass
- Network diagnostics now tells you the truth. It could report "an inbound allow rule for this port is present" when the port was actually blocked, and it described your saved Access setting rather than how the server is really listening. Both checks now read the live firewall rules and the actual network socket, so a server that can't be reached from other computers is reported as a problem instead of a pass.
- If the one-click "Add firewall rule" can't work — on a work computer managed by IT, or with a different firewall — the exact command to run as administrator is now shown and can be copied, and the result is re-checked rather than assumed.
- Serving your network now keeps a fixed port. AI Server's default port clashed with the local AI engine, so it silently fell back to a random port that changed on every restart, breaking firewall rules and other computers' connections. The default is now 11436, and if the port you chose is ever taken the Server page says so instead of moving quietly.
- Fixed "the installed service is still using its previous settings" showing forever for anyone whose model folder or user folder contains a space in its name.
- Security and stability hardening: error responses no longer include local file paths or internal addresses, the TLS certificate and audit-signing key files are now readable only by their owner, request identifiers supplied by callers can no longer inject text into the audit log, and the server stays bounded in memory under heavy or malformed traffic.
- The API keys page now applies the same key limit as your plan, the Connections page no longer stops updating if a log file is busy, and newly created keys show their proper name straight away instead of "Unknown client".
Windows app 2.1.9 — 2026-07-28
The Windows service now reports its status correctly
- Fixed the long-standing problem where a Windows service that was running perfectly still showed as "registered as a Windows Service but not currently running", with port 0 in Network diagnostics and addresses ending in :0. AI Server was unable to check the status of its own background service and wrongly assumed it had stopped.
- Fixed "the installed service is still using its previous settings or an older version" appearing immediately after a clean install. That notice now appears only when the service really does need updating — after you update AI Server.
Windows app 2.1.8 — 2026-07-28
Other AI apps can find your server again — and network serving keeps a fixed port
- Your other AI Suite apps can discover this server again. AI Server now publishes its connection details to a shared location every app can read, so "This computer" reappears under each app's AI Server connections.
- When you serve your network, AI Server now always uses your chosen port (11435 by default) instead of picking a new random port on every start — so firewall rules keep working and other computers stay connected.
- Fixed the Windows service still showing as "not currently running" (with port 0 in Network diagnostics) after updating. Starting the service now also brings its data location up to date.
- You can now enter an API key when connecting an AI app to a server on another computer — required, because servers that serve a network only accept authenticated connections. Find it under Settings → AI Server connections.
Windows app 2.1.7 — 2026-07-24
The app and the Windows service now share one place — status is finally accurate
- Fixed the big one: the Windows service could be running perfectly while the app still showed "registered but not running" (and Refresh didn't help). The app kept its data in a private, per-app location the background service couldn't see, so they never agreed. AI Server now keeps its data in one shared location, so the app always shows the service's true status, port and network address.
- This also means the API keys and license you set up in the app are now visible to the background service and the command-line tool — so network serving and licensing work consistently however AI Server is run.
- Your existing settings, API keys and license are carried over automatically the first time you open this version.
Windows app 2.1.4 — 2026-07-24
The server always starts — and tells you why, in plain language
- The server (including the Windows service) no longer fails to start when you've chosen network serving but haven't created an API key yet. It now starts serving this PC, and shows exactly what to do to open it to your network — instead of dying silently with the reason buried in Event Viewer.
- API keys are now part of Pro (Personal). Because serving your network needs at least one key to keep the server authenticated, you can create keys on the Pro plan — no Commercial plan required. (Commercial still adds unlimited keys and the governance controls.)
- Every start problem now shows its reason on screen: a failed service start reports the exact cause, a service that failed while the app was closed shows why when you open the app, and a server running local-only when you asked for network shows a persistent explanation with the fix.
- Starting the Windows service after an app update now also updates the service to the new version automatically.
- Fixed a settings hint on the Server page that could run off the edge of the card.
Windows app 2.1.3 — 2026-07-23
Settings now apply to the Windows service, and accurate running status
- Changing the network binding (This PC / This Network), port, HTTPS, rate limits or gateway mode and then starting the Windows service now actually applies your change — the service is re-registered with your latest settings instead of quietly keeping the old ones.
- Fixed AI Server sometimes showing "running" right after you reopened the app when the server wasn't actually running — which also left the Stop button unresponsive. Running status is now accurate, the session server is stopped cleanly when you close the app, and Start / Stop always work.
Windows app 2.1.2 — 2026-07-22
Secure connections, network diagnostics, and a more reliable service
- Serve over an encrypted connection (HTTPS): turn on a secure channel for network serving using an automatic self-signed certificate, or import your own certificate (PFX or PEM). A "Download certificate" button gives you the file to trust on your client computers.
- New Network diagnostics on the Server page: one click checks whether the server is running, whether other computers on your network can reach it, and whether the Windows Firewall is blocking the port — with a one-click "Add firewall rule" fix for non-technical admins.
- When you serve your network, AI Server now shows the address other computers should use, instead of just "localhost".
- Fixed "Start as a Windows service" failing with "The system cannot find the file specified" — the service now installs to a stable location and starts reliably, even after an app update.
- Added the Report an issue and Inbox buttons to the title bar, and a Documentation link to the menu.
Windows app 2.1.1 — 2026-07-22
"Run as a Windows service" now works reliably
- Fixed installing and starting AI Server as a background Windows service: it now stays running instead of stopping after a few seconds, and keeps serving your apps in the background.
- When you install or start the service, AI Server now automatically hands over from the "This session" server it's already running — so you no longer have to stop the session yourself, and there's no more port clash.
- The Windows service now uses your existing downloaded models, instead of acting as if none were installed.
Windows app 2.1 — 2026-07-21
A rebuilt AI Server — one shared AI runtime for your whole PC
- AI Server is rebuilt from the ground up: turn this PC into one shared, governed place to host, download, and run the AI models every AI Suite app uses — one model download, one model library, one set of API keys, used everywhere.
- Run it just for this session, or install it as a background Windows service so it keeps serving after you sign out or reboot.
- Manage everything from one place: a Models page to download and organize models (and move your whole model library to another drive without re-downloading), a Providers page to connect cloud AI services with your own keys, and live Connections, Active requests, Usage and Logs pages so you can see exactly what's happening.
- Serving stays free for this PC. Upgrade to Pro to share your local models with other devices on your network, and to unlock governance — per-key usage limits, content filtering, signed audit export, model-lifecycle controls and regional-compliance settings. Commercial plans add organization enrollment for centralized, multi-device management.
- Your privacy is built in: API keys are stored only as secure hashes, usage reporting is content-free, and your data stays on your device.
- Upgrading from the previous AI Server? Your existing settings carry over automatically on first launch, and older apps that connected to it keep working.
- A guided Help you can step through, plus a full translation into 16 languages.