Loc.ai:Control Changelog
Track updates, improvements, and breaking changes to Loc.ai:Control.
August 2026
New
Set up a server or a Raspberry Pi with just a key. Registering a headless device no longer asks for a device name, your email address, or a password typed on the device. Install Loc.ai:Link with one command, generate a one-time registration key in the dashboard, and run the register command on the device. The key is the only credential involved. The device names itself after the account and machine it is running on, so the name in your dashboard always matches the machine in front of you.
A rebuilt Register Device dialog. The dialog now opens on two clear choices: Desktop app for a Mac you sit in front of, and Headless for a server, a Raspberry Pi, or any machine without a screen. The headless side has its own macOS, Linux and Windows tabs with the exact command for each, and generates your registration key in the same place. On Windows, headless installs now use PowerShell; the Command Prompt option has been retired.
Improved
Guided setup is shorter. The setup guide for new accounts is now three steps instead of four. The naming step has gone, because every way of installing now names the device for you.
Clearer device status. A device shows Pending while it is starting up and Online once its agent has reported in, so a device that is registered but not yet connected is easy to tell apart from one that is up and running.
Collapse the sections on the Models page. "Your Models" and "Globally Shared" can each be collapsed independently, which makes a long library much easier to scan.
Shared models appear in your dashboard's recent models. The models card on the dashboard now shows your most recent models whether they are your own or globally shared, with shared ones marked. Previously an account working mainly with shared models saw an empty card.
Fixed
Devices no longer get stuck showing "Pending". A device whose start-up message was lost on the way to us could stay marked Pending in the dashboard indefinitely, even while it was working normally and reporting data. Any device that is sending data is now marked Online automatically.
A more accurate serving count on the dashboard. The dashboard now tells you how many of your devices are serving a model, alongside how many models that adds up to. The figure can no longer briefly show more devices than you own while the page is catching up after a change.
The Download for Mac button. The button pointed at the previous name of the installer file, so the download could fail. It now points at the current one.
Serving commands show your real model name. The ready-to-use API commands shown when a model is serving now include the model's actual name, instead of a placeholder you had to replace by hand.
July 2026
New
Download Loc.ai:Link for Mac. Setting up a Mac is now a one-click download instead of a terminal command. You can grab the Loc.ai:Link app from the guided setup, from the Register Device dialog, or from the new "Download for Mac" button in the top bar. The app walks you through installation and registers your Mac for you, so there is no key to copy. This early release is for Apple Silicon (M-series) Macs; Intel Macs and other systems still use the terminal setup.
Cancel a model deployment. You can now stop a model deployment while it is still downloading, instead of waiting for it to finish. A Cancel option appears on the deploy dialog and in the new Deployments tracker, so a large model started on the wrong device or by mistake no longer ties up your device.
Deployments tracker. The notifications bell in the top bar has a Deployments tab that shows all your model deployments in one place, across every device, with live download progress. You can watch what is in flight, see what just finished, and cancel anything still running without leaving the page.
Redesigned Models page. The Models page has a cleaner layout with summary cards (total models, registry storage used, and how many are deployed), plus search, filtering, and sorting by type, framework, deployment state, and file size. Model types are now grouped so large libraries are easier to browse.
Import from Hugging Face and Kaggle (coming soon). The Models page now shows Hugging Face and Kaggle import options. Direct import is not available just yet: for now the option points you to download the files and use Upload Model, and you can opt in to be notified as soon as one-click import goes live.
Fleet management dashboard (early access). Organizations we onboard now get a dedicated dashboard for their whole fleet. You can see every device with its live health, connection status, agent version, and configuration in one place, roll out agent and model updates to your fleet in stages, and manage a shared device configuration from a single form. This builds on the fleet enrollment we introduced last month and is still onboarded by the Locai team rather than self-serve; see the Fleet Management guide. Contact us at hello@locai.co.uk if you would like to enrol a fleet.
Redesigned dashboard. Your dashboard has a new layout: a greeting with quick actions, stat cards showing devices online, models serving, and average confidence over the last 24 hours with a trend sparkline and median latency, a recent inferences table tinted by confidence, plan usage bars that turn amber at 80% and red at 100% of your limit, summary cards for your models and devices, and a live activity feed that updates in real time without refreshing the page.
Control your usage analytics. Settings has a new Privacy card with a Usage analytics switch that applies to your whole account, wherever you sign in. Analytics are anonymous product events only: never your name, email, browsing data, or anything about your devices and models. Session recording is disabled for all accounts and cannot be turned on. See Usage analytics and privacy.
Improved
A tidier top bar. Your deployments, uploads, and notifications now live together under a single bell, each on its own tab, so the top bar is cleaner and everything is in one place. The light and dark theme switch has moved into your account menu.
Data transfer usage at a glance. The plan card in the sidebar now shows your monthly data transfer against your plan's allowance, alongside your device usage.
Clearer device setup instructions. The guided setup now reminds you to paste the install command into your device's terminal, and points out where to find it: Terminal on macOS or Linux, and Command Prompt or PowerShell on Windows.
A recommended model to start with. New accounts are now pointed to a suggested starter model, so the first deployment is one click away.
Act without leaving the dashboard. Upload Model, Register Device, Deploy Model, and Configure Model now open as dialogs right on the dashboard, so you can finish the task without losing your place.
Model types at a glance. Models are colour-coded by type across the dashboard and Models page: purple for language models, green for audio transcription, teal for audio classification, amber for image classification, and grey for everything else.
Fixed
The dashboard no longer gets stuck loading. We fixed an issue where the dashboard could hang on a loading spinner in some browsers after being left open and idle for a while. Your session now refreshes quietly in the background, and signs you out cleanly once it has genuinely expired.
Deleting or uninstalling a model works reliably again. We fixed a bug where confirming a model delete or uninstall did nothing and left the page unresponsive.
Deployment status no longer gets stuck. Deployments now reliably reflect their final state. Downloads that finished a long time ago no longer linger in the tracker as if they were still in progress.
Served-model connectivity for Workspace. Fixed a misleading "document too large" error when running inference from Workspace, caused by the Workspace address missing from the approved origins baked into served models. Newly deployed models are fixed automatically. Models deployed before this update keep their previous settings until they are repaired or redeployed; there is no automatic fix for existing deployments. See Fixing "document too large" errors.
June 2026
New
Guided device setup. New accounts are now welcomed with a step-by-step setup guide that walks you through naming your first device, picking your platform (Linux, macOS, or Windows), installing the agent with a ready-to-copy command, and deploying your first model, all without leaving the page. You can minimise the guide at any time and pick it back up later from the Finish setup button, or hide it for good if you would rather set things up yourself. You can always add devices from the Devices page as before.
What's New, right inside the app. A new sparkles icon in the top bar keeps you up to date with the latest releases. A badge shows when there is something new, and opening it shows recent updates grouped by product and type, with links to learn more and to the full changelog. You can browse past releases or turn off the pop-up if you prefer.
Sign in with GitHub. You can now sign in or create your account using GitHub, in addition to email and password and Google. Choose Continue with GitHub on the login or sign-up screen and approve access. If you already have a Locai account with the same verified email, your GitHub login is linked to it automatically, so there is no duplicate account to manage.
Remove a model from a single device. From a device's detail panel you can now uninstall a model from just that device. The model stays in your library and on any other devices it is deployed to. If the device is offline, the removal is queued and runs automatically the next time it reconnects (within 7 days).
Delete a model from your whole account. From the Models page you can now permanently delete a model. This removes it from your library and queues its removal from every device it was installed on, and frees up the storage it was using. Both removal actions ask you to confirm first.
Fleet management (early access). Organizations can now enroll devices at scale, managed by Locai. This is currently onboarded by the Locai team rather than self-serve in the dashboard. Contact us at hello@locai.co.uk if you'd like to enroll a fleet.
Improved
Live device metrics. Device CPU, memory, temperature, and available storage now update in real time on the dashboard, with no manual refresh and smoother, more responsive readings.
More egress on the free tier. The free Developer plan's monthly data transfer allowance has increased from 2 GB to 10 GB.
Clearer deployment progress. The progress bar when deploying a model is now more accurate and consistent from start to finish.
More storage on the Developer tier. The Developer plan's model registry storage has increased from 1 GB to 5 GB.
More reliable model deployment and serving. Model deploy, serve, and update commands are now built and validated centrally before they reach your devices, and commands that cannot be delivered immediately are retried instead of being silently dropped. The result is fewer stuck commands and more accurate "deployed on" information across your fleet.
Stronger account protection. We added rate limiting across sign-in, sign-up, and single sign-on to protect accounts against automated abuse and brute-force attempts. Normal usage is unaffected.
Plans and pricing
Starter plan update. The Starter plan is now GBP 50 + VAT per month and includes 10 connected devices, with additional devices available as you grow. Existing Starter subscribers keep their current pricing.
Fixed
Windows installation on corporate networks. The Windows device install command now works reliably on networks where certificate revocation checks are blocked (common behind corporate proxies and firewalls), so onboarding no longer fails with a certificate error.
Activity history accuracy. Fixed an issue where some start and stop actions for inference and serving were not recorded in your activity history.
Served-model connectivity for Workspace. Corrected the allowed origins for served language models so the Workspace application (then named SafeChat) connects to them as expected.