> For the complete documentation index, see [llms.txt](https://docs.valohai.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.valohai.com/changelog/archive/updates-2026.md).

# Updates 2026

### November 18, 2026

#### Features & Enhancements

* **Catch missing required pipeline inputs up front**
  * Reject pipelines in which a required input has neither a value nor an incoming edge
  * Find out about a misconfigured node when you create the pipeline, not when the node fails
* **Audit environment variable changes**
  * Record changes to environment variables and environment variable groups in the audit log
  * See who changed a variable and when
* **Use same-record metadata labels in confusion matrices**
  * Build confusion matrices from predicted and actual labels logged in the same metadata record, and sort the matrix by metadata name
  * Plot classification results without restructuring how you log them
* **Improved API documentation**

#### UI/UX Improvements

* Each environment's description and GPU spec are now shown on its card, and environments are listed by name.
* URL values in the executions table are now rendered as compact external links.
* The list of enabled versions is now collapsed when a deployment has many of them.
* "Move to start" and "Move to end" buttons are now provided in the column popover.
* Only the visible part of long input lists is now rendered in execution details.
* Exact dataset and version name matches are now listed first in input search.
* Dates are now included in the time axis ticks when a chart spans several days.
* A parameter error is now shown when an input placeholder cannot be resolved.
* SSH keys can now be removed from a repository with a "Remove SSH key" button, and a repository can be switched from SSH to HTTPS, which clears the key.
* Each user's teams and each team's users are now shown in the organization settings.
* The final status of an execution is now included in finish notifications.

#### Bug Fixes

* Execution nodes that were already running are now allowed to finish when a pipeline is halted.
* Duplicate file names are now renamed when dataset versions are zipped.

### November 12, 2026

#### Features & Enhancements

* **Filter notifications by importance**
  * Mark important notifications, such as comment mentions, pipeline approval requests, failed executions, tasks, and pipelines, and trigger launch failures or emergency stops, with a badge, and add a filter that shows only those
  * Focus on the notifications that need attention

#### UI/UX Improvements

* **Improved dataset view**
  * Navigate between dataset versions and back to the dataset
  * Add a version picker to dataset version pages that lists the newest versions and searches older ones, and turn the line above dataset and version names into breadcrumbs back to the dataset and the datasets list
  * Move around datasets the same way you move around models
  * Fits better on small screens

#### Bug Fixes

* **Respect environment overrides on task executions**
  * Validate task child settings against the task's configuration, so an explicitly selected environment overrides the one in `valohai.yaml`
  * Run task executions on the environment you picked

### November 5, 2026

#### Features & Enhancements

* **Comment on model versions**
  * Post comments on a specific model version, see a version pill on version-specific comments in the model's Comments tab, and toggle model-wide comments on the version comments page
  * Keep feedback about a model version attached to that version instead of buried in the model's general thread
* **Open dataset pages inside the organization view**
  * Show dataset and dataset version pages within the organization's Datasets tab, keep the active tab in the URL, link back to the list you came from, and redirect old standalone dataset version links
  * Browse organization datasets without losing your place or navigation
* **Search datasets and versions with tags, URIs, and execution counters**
  * Add a `search` filter to the dataset and dataset version APIs that matches exact tags, accepts `dataset://` URIs, and finds versions by the producing execution's `#counter`
  * Find the right dataset version for an execution input by what you already know about it
* **Explain why an execution was not auto-restarted**
  * Add user-visible log entries when an execution errors on an interrupted machine and no restart reason is known, and describe `VH_AUTORESTART_COUNTER` in the environment variable editor
  * Understand why a spot execution stopped instead of restarting, and how the restart limit applies to copied executions
* **Notify on comment mentions and pipeline approvals by default**
  * Turn on comment mention and pipeline node approval notifications in default project notification settings, and reword labels to "execution stops" and "pipeline fails"
  * Hear about mentions and pending approvals without configuring them first
* **Clean up old in-app notifications automatically**
  * Delete in-app notifications older than 7 days on a schedule, while keeping each user's newest 50
  * Keep the notification list short without losing recent activity

#### UI/UX Improvements

* **Improve dataset selection in execution inputs**
  * Make the dataset and version selects clearable, clear both on Deselect, keep `latest` when only a dataset is picked, show the input's URL in its summary, and add spacing between the selects
  * Pick and change dataset inputs with fewer surprises
* **Improve the notification dropdown and user suggestions**
  * Make the notification dropdown easier to read and navigate by keyboard, list notifications newest first, suggest only active users, include the owner organization's members, and sort suggestions by username
  * Find the right notification and the right person to mention faster
* **Hide the "only mine" option where it does not apply**
  * Remove "only mine" from notification events that are not tied to a user's own work, such as dataset versions, deployments, models, and triggers
  * Avoid notification settings that silently do nothing
* **Truncate long tags in popovers and selects**
  * Bound execution tag popovers, truncate long tag chips and labels, and align title and tag layouts across creation views
  * Keep long tag names from breaking menus and forms
* **Explain the pipeline duration column**
  * Add a tooltip to the Duration column in pipeline tables explaining it is the sum of execution durations
  * Make it clear why pipeline duration can exceed wall-clock time
* **Make always-open panel headers selectable**
  * Render always-open panel headers as plain text instead of buttons
  * Let users select and copy header text

#### Bug Fixes

* **Fix inconsistent runtime config reset**
  * Restore execution runtime defaults and copied settings correctly, keep time-limit fields in sync, and keep the selected runtime visible when environment filters change
  * Stop execution creation from dropping or hiding runtime settings
* **Reset pagination after bulk deleting dataset versions**
  * Return to a valid page after deleting dataset versions in bulk
  * Stop the versions table from showing an empty page after deletion
* **Show user suggestions above disabled buttons**
  * Render the user autocomplete popup above disabled buttons
  * Keep mention suggestions visible and clickable

### October 28, 2026

#### Multi-repository projects

Full walkthrough in the [multi-repository announcement](https://docs.valohai.com/changelog/announcements.md#multi-repository-projects-and-fetching-for-large-repositories) and the [docs](https://docs.valohai.com/git-integration/multi-repository-projects).

* **Connect multiple Git repositories to a project**
  * Add named repositories next to the default repository in project settings and the repository API, with settings, commit browsing, and downloads scoped to the selected repository
  * Run code from several repositories in one project without merging them into a single repository
* **Select commits from any project repository**
  * Add a repository selector to execution, task, notebook, pipeline, and deployment version creation, accept references such as `train:main`, and keep the repository when copying
  * Run and deploy each step from the repository its code lives in
* **Choose what to fetch**
  * Fetch all repositories together by default, or pick a repository and branch patterns with "Select what to fetch...", from a trigger, or through the fetch API, with results reported per repository in one notification
  * Fetch exactly what you need and see which repository a failing fetch is stuck on
* **Change the commit of a single pipeline node**
  * Pick a different commit or repository for one node in the pipeline editor, and see the differing commit on the graph node
  * Run one step from a newer or fixed version of its code without moving the whole pipeline

#### Features & Enhancements

* **Copy pipelines from their source commit**
  * Use the commit a pipeline was created from when copying it, and pick the matching ad-hoc commit when the source was an ad-hoc run
  * Make a copied pipeline run the same code as the original
* **Auto-disable triggers only after three failures within 12 hours**
  * Count only failed or errored runs from the last 12 hours toward a trigger's three-failure limit
  * Stop occasional failures spread over days from switching a trigger off

#### UI/UX Improvements

* **Redesign the notification channel modal**
  * Rebuild the channel form with the current form components, add selection of the triggers a channel applies to, and allow unnamed channels with a fallback label
  * Set up notification channels with clearer validation and fewer required fields
* **Tidy up the create project form**
  * Put owner and project name on one row, rename "Random name" to "Generate project name", reword template "Import" to "Use", and stack the form on small screens
  * Make creating a project quicker to read and fill in

#### Bug Fixes

* **Prevent read-only organization admins**
  * Reject making organization admins read-only, clear read-only status when promoting a member to admin, and allow removing read-only from existing admins that have it
  * Make sure admins always have the write access their role implies
* **Keep long data uploads working**
  * Refresh upload tokens every 10 minutes while uploading, and return clear errors for expired or invalid upload tokens instead of server errors
  * Upload or adopt data for longer than the 30-minute token lifetime
* **Copy pipelines created by triggers and the from-template API**
  * Copy YAML-form pipelines, including ones without edges, and report node problems in them as validation errors
  * Let the copy pipeline trigger action work on pipelines that triggers or the API created
* **Show errors when a pipeline template fails to load**
  * Display an error, naming the missing step, when creating a pipeline from a template fails
  * Stop the create button from doing nothing

### October 13, 2026

#### Features & Enhancements

* **Change the commit when copying an ad-hoc execution**
  * List the source ad-hoc commit in its own group in the commit dropdown, next to the project's Git commits
  * Rerun an ad-hoc experiment against a committed version of the code without rebuilding the execution from scratch
* **Filter executions to deleted only**
  * Add a "Deleted only" checkbox to the executions table filters in project and notebook execution lists
  * Find and inspect deleted executions without going through the API

#### UI/UX Improvements

* **Show repository fetch progress in a modal**
  * Move fetch progress into a modal with "Cancel fetch" and "Continue in background" buttons
  * Make it clear that closing the progress view leaves the fetch running
* **Keep project settings tab titles on one line**
  * Stop tab titles from wrapping, and switch to condensed titles on screens narrower than extra large
  * Keep the project settings tabs tidy on medium-sized screens

#### Bug Fixes

* **Update execution service buttons when they are re-emitted**
  * Replace a service button in place when an execution emits a new button with the same name, instead of keeping the first one
  * Show the current link when an execution updates a service button during a run
* **Hide deleted executions from the organization's latest executions**
  * Exclude deleted executions when loading the latest executions for an organization
  * Stop deleted runs from appearing among recent activity
* **Fix select field layout**
  * Truncate long selected values and keep the dropdown caret aligned to the right edge
  * Stop long values from pushing the caret out of the field

### October 7, 2026

#### Features & Enhancements

* **Rebuild deployment images**
  * Add a `rebuild-images` action to the API and a rebuild button under deployment version details, and allow rebuilding an image that is already built
  * Recover from a stale or broken image without recreating the deployment version
* **Let project environment variables override group variables**
  * Apply a project's own environment variable over the value coming from an attached environment variable group, in both the API and the UI, when the project inherits group variables
  * Override a shared group value on one project without detaching it from the group
* **Apply organization environment variables on trigger runs**
  * Resolve organization transient environment variables scoped to all members for runs started by a trigger, when the trigger project matches
  * Give triggered runs the same organization variables that manual runs get
* **Reduce background fetch polling**
  * Stop discovering project fetches while the browser tab is idle or hidden
  * Cut needless traffic from tabs nobody is looking at

#### UI/UX Improvements

* **Report deployment image build progress and failures**
  * Show how far an image build has progressed, highlight the build failure reason, and combine the three endpoint columns into a single "Status" column so longer build details fit
  * Replace a row of narrow columns with one readable account of what the deployment is doing
* **Wrap long scaling status text in admin**
  * Break long scaling status lines anywhere instead of overflowing
  * Keep long scaling logs in view

#### Bug Fixes

* **Count a dataset version purge once**
  * Increment the purge count a single time per purge instead of once per pass
  * Report accurate purge counts on dataset versions
* **Stop duplicate image builds from concurrent packing**
  * Re-read the packaging state under a row lock before starting a build, so two concurrent requests cannot both queue one
  * Stop the same image from being built twice at once
* **Stop showing raw job status as deployment status**
  * Return a deployment status instead of passing the underlying job status through, and catch the error the job queue actually raises when status data is missing
  * Remove jargon states such as `NONE` and `ACQUIRED` from the deployment view
* **Preserve task settings when changing the commit**
  * Reset the task definition only when the step actually changes, instead of on every commit change
  * Keep a configured task intact when you point it at a newer commit
* **Hide the empty commit warning on ad-hoc execution copies**
  * Stop warning about a missing commit when copying an ad-hoc execution, which has none by design
  * Remove a warning that never applied

### October 1, 2026

#### Features & Enhancements

* **Improved machine lifecycle handling**
  * You'll now have a more accurate view of machines as they move through their lifecycle.
* **Repository fetches run in the background**
  * Fetching larger repositories than before is now possible. If you are using the API to fetch repositories, instead of a request timing out, you will now get a 202 Accepted status code for your fetch if it's taking longer than was allowed before. You may also pass in `wait=0` just kick off a fetch and immediately return a 202 Accepted status code. You can use the new repository fetches API to then see the status of your fetch.
* **Speed up output uploads**
  * We took a critical look at what happens on our end for each output uploaded, and then fixed the silly bits that were taking too long. However, you'll still find it's more efficient to use a small amount of larger files rather than a whole bunch of tiny file confetti.
* **Improved pipeline editor sidebar**
  * We improved the pipeline editor in August, but now that leaves are falling outside, we had the time to cozy up with a tea and make the pipeline editor even better.

#### Bug Fixes

* **Fix nested** `.vhignore` **ignore and override rules**
  * Ignore and override patterns from `.vhignore` files are now applied more consistently.
  * Dotfile ignore rules used to exclude the entire repository root; that's fixed too.
* **Preserve cleared inputs when copying executions**
  * Keep inputs that were deliberately emptied cleared in the copied execution instead of restoring the original values
* ... and some assorted bits and pieces!

### September 16, 2026

#### Features & Enhancements

* **Add workspace volumes to Valohai deployment endpoints**
  * Add a workspace volume field to endpoints with a mount path, a size in GiB, and an optional storage class, include the volume in the generated endpoint manifests, and expose the field through the endpoint API
  * Give an endpoint its own writable scratch space instead of only the read-only container filesystem or a shared volume
* **Fill endpoint workspace volumes from** `valohai.yaml`
  * Read the workspace volume definition from the endpoint's YAML when the request leaves the field unset, and pre-fill the same values in the endpoint UI
  * Keep workspace volume configuration in version control instead of retyping it per endpoint
* **Store parameter overrides on triggers**
  * Let a trigger action carry execution and pipeline parameter values that apply to every run it creates, with request-time `param.*` values winning name by name
  * Close the gap for scheduled triggers, which have no incoming request to carry parameters
* **Validate trigger parameter overrides when the trigger is saved**
  * Check override names against the resolved step or pipeline, resolving pipeline node names to their step, and skip nodes whose parameters cannot be known such as deployment nodes
  * Turn a misspelled parameter name into an error on the trigger form instead of a run of failures and an emergency stop
* **Resolve Azure Entra ID group overage**
  * Detect when the identity provider omits group claims because the user belongs to too many groups, and fetch the group list from the Microsoft Graph API instead
  * Keep SSO group mapping working for users in large numbers of Entra ID groups

#### UI/UX Improvements

* **Edit trigger parameter values in the trigger form**
  * Add a parameter editor to the run execution and run pipeline action editors for every trigger type, supporting several values for one parameter name and a "Show as JSON" modal for pasting or replacing a whole set at once
  * Set the parameters a scheduled trigger runs with directly in the form
* **Show the endpoint workspace volume on the endpoint details view**
  * Add a workspace volume editor next to the shared volume editor and display the configured volume on the details page
  * Make endpoint storage configuration visible without reading the manifest

#### Bug Fixes

* **Return a partial trace graph when a round times out**
  * Catch the query timeout and finish the round as incomplete so the client can continue with its token, and return a 408 rather than a 500 when the round produced nothing at all
  * Let large trace views load in pieces instead of failing outright
* **Include datum consumers matched through expanded URIs**
  * Match consuming executions by URI overlap rather than requiring a single input to carry every URI in the batch
  * Stop dropping executions that reference a datum through wildcard or query inputs from the trace graph

### September 3, 2026

#### Features & Enhancements

* **Add shared volumes to Valohai deployment endpoints**
  * Add a shared volume field to endpoints, include the volumes in the generated endpoint manifests, expose them through the endpoint API, and add an API that lists the persistent volume claims available in the deployments namespace
  * Let endpoints read and write shared storage, for example a model cache or a common data directory
* **Fill endpoint shared volumes from** `valohai.yaml`
  * Read the `shared-volumes` definition from the endpoint's YAML whenever the request leaves the field null or missing, and pre-fill the same values in the endpoint creation UI
  * Keep shared volume configuration in version control instead of retyping it per endpoint
* **Pass parameters to webhook-triggered executions and pipelines**
  * Accept `param.*` URL query parameters on webhook trigger requests, support repeated parameters for multi-value inputs, and reject unknown parameter names
  * Let one webhook trigger drive different runs depending on what the caller sends
* **Respect** `autorestart` **from** `valohai.yaml`
  * Preserve the `autorestart` setting through runtime configuration validation, keep an explicit `autorestart: false` in effect, and drop `autorestart: true` on environments that are not spot instances
  * Make spot instance restart behavior configurable from the step definition rather than only per execution
* **Support runtime configuration in the pipeline and task node UI**
  * Include runtime configuration in the node config value merge and read the initial `autorestart` value from the step
  * Let pipeline and task nodes carry the same runtime settings as single executions
* **Report which webhook trigger condition declined a request**
  * Give each trigger condition its own declined detail and error code, and return the condition that actually rejected the call instead of a generic validation error
  * Tell webhook callers whether the secret, the rate limit, or a payload filter blocked them, while keeping configuration details in the trigger log
* **Improve trigger logging**
  * Write a log entry on successful fetches, include more detail about partial fetch failures, and reset the error counter when a trigger is auto-disabled
  * Make it easier to see what a trigger did and why it stopped
* **Allow environment slugs in the machine inform view**
  * Accept an environment slug where the view previously required an identifier
  * Simplify worker configuration against a named environment

#### UI/UX Improvements

* **Show an execution alert when autorestart does not match the environment**
  * Add an autorestart unavailable alert to the execution view, write the same event into the execution log stream instead of an admin-only entry, and sort execution alerts
  * Show the reason an execution will not restart to everyone who can see the execution
* **Handle** `autorestart: true` **set in the configuration file**
  * Show the autorestart checkbox when the step config sets it, keep the checkbox visible once shown, warn when autorestart is checked on an unsupported environment, and ask for confirmation before dropping it on a non-spot instance
  * Prevent silent loss of an autorestart setting when switching environments
* **Do not lock environment selection for every runtime config property**
  * Mark only the runtime config keys that carry custom resources, so other properties no longer switch the resource config into custom mode
  * Keep environment switching available when the runtime config holds unrelated settings
* **Show shared volumes on the endpoint detail page**
  * Add a shared volume editor to endpoint creation and list the configured volumes on the endpoint detail page
  * Let users check an endpoint's storage setup without reading the manifest
* **Add a comment count column to the models list**
  * Add a `# Comments` column to the models listing table
  * Show which models have discussion attached before opening them
* **Explain verified domain purposes**
  * Add help text for the verified domain purposes and fix the verified domain layout in admin
  * Clarify what each purpose does when setting up a verified domain

#### Bug Fixes

* **Keep webhook trigger condition errors out of the rolled back transaction**
  * Wrap only the payload storage and run enqueueing in a transaction, so the log entry explaining an unmet condition survives the error response
  * Stop a webhook that has stopped authenticating from leaving an empty trigger log
* **Prevent duplicate restarts from a cached status value**
  * Refresh the execution status from the database before restarting
  * Avoid restarting the same execution twice
* **Fail cleanly on an invalid parent filter value**
  * Return a validation error instead of a server error when the parent filter receives a value it cannot parse
  * Give a usable message for a malformed filter
* **Require a datum when creating a datum alias**
  * Validate that the alias create request provides a datum, and group and clarify the error codes
  * Reject an incomplete alias request up front instead of failing later
* **Block shared volumes for public deployment targets**
  * Reject shared volume configuration on public deployment targets
  * Prevent a configuration that cannot work on shared infrastructure

### August 12, 2026

#### Features & Enhancements

* **Configure peon container resources for Kubernetes executions**
  * Accept an optional `peon` runtime config block that sets the resources for the peon container, with validation and defaults moved into shared constants
  * Let teams tune peon resource requests to fit their cluster
* **Return known feature flags from the user endpoint**
  * Add a `flags` array to `/users/me` listing the flags enabled for the user in the database and globally through settings
  * Make it easier to see which feature setup a user actually has when debugging

#### UI/UX Improvements

* **Keep Model Hub only users inside Model Hub**
  * Redirect model-hub-only users from project dataset views back to the model hub or front page, and hide the execution trace link unless the user can access executions
  * Stop external users from landing on views they cannot use

#### Bug Fixes

* **Fix the model version selector**
  * Repair a regression in `ModelVersionSelect`
  * Restore model version selection where it had stopped working

### August 4, 2026

#### Features & Enhancements

* **Show and edit model and model version properties in the UI**
  * Add property cards to the model and model version overviews, gated behind the `EditModelProperties`capability
  * Let users read and update properties without going through the API
* **Create, update, and delete model version properties through the API**
  * Support writing linked dataset version properties via the model version API
  * Give automation a way to keep model version metadata current
* **Filter models by property query**
  * Support a `property_query` parameter in model filtering, with user-friendly errors for invalid queries and insufficient permissions, and document it in the API spec
  * Let users narrow large model listings down to the versions they care about
* **Add the query builder to the Models listing table**
  * Wire property query filtering into the models table and surface query errors inline
  * Bring the same property filtering to the UI that the API now offers
* **Emphasize selected metadata properties**
  * Sort emphasized properties to the top of property lists, mark them with a bold key and an aligned band, and show them as chips above the model version title
  * Put the properties that matter most where users see them first
* **Search the metadata properties list**
  * Add as-you-type filtering to the properties list
  * Make long property lists usable
* **Add a pipeline-wide High Priority toggle to the Create Pipeline view**
  * Apply queue priority to every execution and task node at once, skipping nodes whose environment does not support priority and reused executions that ignore it
  * Remove the need to set priority node by node on large pipelines
* **Support dependency edges in pipelines**
  * Add backend and frontend support for edges that express a dependency without passing data
  * Let pipelines order nodes that have no input or output relationship
* **Resolve environment variable groups by name**
  * Accept a group name where only an ID was accepted before
  * Make environment variable groups easier to reference from the CLI and API
* **Expose active and running machine states**
  * Return `active` and `running` on the Machines API and use `active` as the signal in the frontend
  * Give a clearer read on what each machine is doing
* **Extend the productivity dashboard time range to three years**
  * Raise the query limit to three years, render today-relative ranges better, and bound the frontend range to what the backend allows
  * Let organizations look further back at their usage
* **Speed up the productivity dashboard**
  * De-correlate the node-reuse aggregation, batch the N+1 queries in time-to-first-value, and add `(target, identifier)` indexes to the log entry tables
  * Cut load time for organizations with long histories
* **Add a safe unpurge path**
  * Allow verified datums to be unpurged in the rare cases where the data still exists
  * Give admins a recovery option after an unintended purge

#### UI/UX Improvements

* **Rework the pipeline creation view**
  * Merge the title, tags, priority, and reuse rows, drop the extra card around the graph, lay the view out in two columns where space allows, and zoom the graph to the selected node
  * Make pipeline creation easier to scan and faster to work through
* **Replace the pipeline node editor with a collapsible sidebar**
  * Move node editing into a sidebar that collapses, zooms to the selected node, and fits the graph again on deselect
  * Give the graph more room while editing nodes
* **Move pipeline title and tags into the view header**
  * Show title and tags in a full-width view header, unframe the pipeline graph, and keep the fields editable only in the pipeline editor
  * Make the header consistent with other views and prevent accidental edits outside the editor
* **Improve the paginator and trigger logs**
  * Restyle the paginator component and apply it to trigger logs
  * Make paging through long logs clearer
* **Streamline the machine event log**
  * Derive human-readable titles for machine status events, render leaner rows, and strengthen the error cue
  * Make it easier to follow what happened to a machine
* **Use a consistent time format for machine age**
  * Align machine age formatting with the rest of the app
  * Remove the mismatch between machine views and everything else
* **Allow copying machine names from the machines table**
  * Add a copy action to the machine name cell
  * Save users from retyping machine names
* **Adjust the table row count label position**
  * Move the "Total of" row count label
  * Keep the count where users expect to find it

#### Bug Fixes

* **Fix the token creation modal**
  * Restore the submit button type that was dropped
  * Make token creation work again
* **Fix hardware stats graphs not showing side by side**
  * Make the layout width optional and render each chart as a flex item
  * Restore the side-by-side comparison of hardware usage graphs
* **Fix machine lifecycle init and interrupted states**
  * Correct the initial lifecycle fallback and mark interrupted machine terminations as failed
  * Show accurate lifecycle chains for machines that never started cleanly or were interrupted
* **Validate only added or updated metadata keys**
  * Skip validation for keys being removed
  * Allow deleting old keys that no longer pass current validation rules

### July 7, 2026

#### Features & Enhancements

* **Attribute team execution quota to the owning team**
  * Charge an execution's quota to the team(s) that own its project instead of every team the creator belongs to, falling back to the creator's teams for teamless projects
  * Stop a user in multiple teams from exhausting an unrelated team's quota by launching in one team's project
* **Include metadata in model version API responses**
  * Return model version metadata in the details response and, on request with `include=metadata`, in the list response
  * Let API consumers read model version properties without a separate lookup

#### UI/UX Improvements

* **Add a local/UTC time toggle to pipeline logs**
  * Bring the local-versus-UTC toggle from execution logs to pipeline logs and persist the choice app-wide
  * Keep timestamp display consistent across execution and pipeline views

#### Bug Fixes

* **Apply checksum settings to the default S3 storage handler**
  * Extend the request checksum calculation and response checksum validation options to the default S3 storage handler
  * Restore reliable S3 access for affected stores

### June 30, 2026

#### Features & Enhancements

* **Capture Kubernetes cluster status snapshots**
  * Save a best-effort cluster status snapshot when an execution ends and show it on the execution details, flagged as the last known state when no live data is available
  * Let users inspect the cluster state of finished Kubernetes executions
* **Select a dataset as execution input**
  * Add a dataset selector for execution inputs, along with an option to ignore purged data and a purge icon for purged datums
  * Make it easier to run executions against datasets and see which data is no longer available
* **Add a model metadata endpoint and expose model properties in the API**
  * Add an `add_metadata` endpoint for models, mirroring datasets, and include model metadata properties in model details and list responses
  * Give API consumers a way to read and write model properties
* **Open the data multi-detail endpoint to SME users**
  * Allow the `data/multi_detail` endpoint for SME access and restore non-model datum handling
  * Let SME users fetch datum details that previously failed
* **Speed up the library step API**
  * Prefetch the commit instead of joining per row when listing library steps
  * Cut response time for projects with many steps

#### UI/UX Improvements

* **Enrich the environment card status view**
  * Build the Status tab summary cards on 30-day machine lifecycle aggregates with a lifecycle breakdown
  * Give a clearer picture of recent machine activity when reviewing an environment
* **Show machine lifecycle and state in the Machines tab**
  * Add a per-row lifecycle chain derived from status reports
  * Make it easier to follow how each machine moved through its states
* **Link to a machine's current or last execution**
  * Expose the execution ID on machine status report entries and link the active and last execution in the Machines tab
  * Let users jump straight from a machine to the work it is running or last ran

### June 25, 2026

#### Bug Fixes

* **Stop new environments from being starved during scaling**
  * Sort never-scaled environments first in the scale order and track a dedicated last scale attempt time for round-robin scaling
  * Ensure brand-new environments get scaled instead of being processed last when the scale loop hits its time limit

### June 24, 2026

#### Features & Enhancements

* **Auto-restart spot-interrupted executions in pipelines and tasks**
  * Restart and replace spot- or preempt-interrupted autorestart executions inside pipelines and task children, carrying previous outputs forward as restart inputs and using a separate autorestart budget
  * Stop pipelines and tasks from concluding with half-finished work after a spot interruption
* **Reserve workspace volume PVCs on create**
  * Reserve the workspace volume PVC when an execution is created rather than only when its manifest is sent
  * Reduce the chance of a reserved PVC being reclaimed before the execution attaches it

#### UI/UX Improvements

* **Pre-fill the create execution runtime config from the environment default**
  * Use the environment's default runtime config as the base when creating an execution, without flipping the UI into custom mode, and apply the selected preset's values on creation
  * Save users from re-entering cluster-specific defaults on every new execution

#### Bug Fixes

* **Ignore deleted environments in situation report matching**
  * Stop matching situation reports against environments that have been deleted
  * Avoid stale environments appearing in machine status data
* **Mark a PVC as used at decision time**
  * Mark a PVC as used as soon as the decision to use it is made, not only when the manifest is sent
  * Narrow the window where a PVC could be reclaimed between decision and attachment
* **Hide disabled deployment versions from aliases**
  * Stop offering disabled deployment versions when choosing an alias target, while keeping a disabled target selectable when editing an existing alias
  * Prevent aliases from pointing at versions that cannot serve traffic

### June 17, 2026

#### Features & Enhancements

* **Workspace volume controls for Kubernetes workers**
  * Add per-execution and runtime config preset control of the worker workspace volume to the UI
  * Let teams tune scratch storage for Kubernetes jobs instead of relying on a single default
* **Manage deployment endpoint ephemeral storage**
  * Add backend, admin, and frontend controls for setting ephemeral storage on deployment endpoints
  * Give users a way to size scratch space for endpoints that need it
* **Speed up execution input signing**
  * Cache the SigV4 signing key in the experimental S3 signer and trim per-datum overhead in worker input resolution
  * Cut input resolution time for executions that read many datums

#### UI/UX Improvements

* **Rework the environment machines tab**
  * Rebuild the machines table with status summaries, situation report codes, a default Status tab, and an Age column in place of Uptime
  * Make it easier to see machine health at a glance when monitoring an environment
  * Further improvements to follow
* **Show workspace volume settings in the preset summary**
  * Display workspace volume configuration in the runtime config preset listing summary
  * Let users confirm preset storage settings without opening each preset
* **Hide numeric query operators for non-numeric values**
  * Stop offering numeric comparison operators in the query builder when the entered value is not a number, and show validation errors before logic errors
  * Prevent users from building queries that can never match

#### Bug Fixes

* **Handle executions without cumulative metadata in expression evaluation**
  * Stop expression evaluation from failing when an execution has no cumulative metadata yet
  * Avoid errors on executions that have not reported metadata
* **Fix the experimental S3 signer for AWS China and gateway endpoints**
  * Correct signed URL generation for China partitions and S3 gateway endpoints
  * Restore reliable input access for stores in those configurations
* **Remove a warning on the endpoint detail page**
  * Fix a stray warning shown on deployment endpoint details
  * Clean up the endpoint view

### June 3, 2026

#### Features & Enhancements

* **Editable datum properties with a unified properties list**
  * Reuse the dataset properties editor for datum properties and consolidate the layout into a single properties list with row-level validation errors
  * Give users a consistent editing experience across datasets, dataset versions, and datums
* **Environment scale divisor**
  * Add a scale divisor setting on environments that lets multiple executions share one machine when computing desired scale, both in the backend and as a configurable field in the UI
  * Allow administrators to match autoscaling behavior to workloads that pack many executions onto a single machine
* **Include result metadata in dataset and dataset version API responses**
  * Return the resulting metadata payload after creating or merging properties on dataset and dataset version endpoints
  * Save clients an extra fetch to confirm the stored metadata state
* **Improve API error response documentation**
  * Document additional error response shapes in the OpenAPI spec, including field-specific validation errors and list-of-errors responses, and add an `invalid_metadata` error code for dataset and dataset version metadata endpoints
  * Make API integrations easier to build against by exposing the full set of error formats

#### UI/UX Improvements

* **Inline validation errors in the query builder**
  * Show property query validation errors directly in the query builder and dataset browser instead of as a separate alert
  * Help users spot and fix invalid query rules without leaving the builder
* **Reset to step or copy source on revert**
  * Rename Revert to → Reset to across pipeline and step revert buttons, reset parameter variant state and deployment node configs on revert, and refresh the pipeline blueprint by node match when the commit changes
  * Make revert behavior predictable when switching commits or recovering from copy-source edits
* **Show absolute and relative time in commit info**
  * Replace the standalone relative time component with a combined absolute-and-relative time display in commit info
  * Give users both a quick "X days ago" read and the exact timestamp on hover

#### Bug Fixes

* **Unstick UI on pipeline copy initialization error**
  * Catch initialization errors when opening the Copy Pipeline card, surface them as a permanent inline error, disable the continue button, and report context to Sentry
  * Stop the UI from hanging at "Just a moment..." when a commit mismatch or missing node breaks initialization
* **Fix ChangeCommitModal discarding validation results**
  * Preserve validation results when interacting with the Change Commit modal instead of throwing them away
  * Keep validation state intact while users switch commits
* **Treat task and child creation lock timeouts as retryable**
  * Mark "unable to start children" and task create timeout errors as transient so the pipeline retries them on the next stepping pass instead of permanently failing the node
  * Avoid spurious pipeline failures when advisory locks are briefly contended
* **Persist commit refs as each commit is fetched**
  * Record a CommitRef immediately after each successful fetch instead of waiting until all refs are fetched
  * Preserve already-fetched commit references even if the overall fetch times out partway through

### May 12, 2026

#### Features & Enhancements

* **Validate property names on dataset, version, and datum endpoints**
  * Validate metadata property names server-side when creating or merging dataset, version, and datum properties, including the multi-datum endpoint and execution-provided metadata
  * Reject malformed names early so downstream queries and filters stay reliable
* **Skip invalid metadata when uploading datums**
  * Drop datum metadata entries with invalid property names during upload rather than failing the whole request
  * Keep bulk uploads moving even when a subset of entries has bad property names

#### UI/UX Improvements

* **Edit property names, types, and values in the property editor**
  * Make property names editable, allow changing a value's type (string, number, boolean) with a boolean value selector, show a read-only type indicator, and trim names on entry
  * Add per-row value validation, an invalid property name indicator, and a uniform row layout in the structured editor
  * Let users adjust existing dataset and model properties without dropping into JSON
* **Validate property names inline in the property editor and query builder**
  * Surface invalid property name errors in the editor and reuse the same validator in the query builder, with reserved names exposed to the frontend
  * Catch naming mistakes before they hit the API
* **Tighter query builder layout and filter behavior**
  * Flatten the rule layout, move error display inline, and suppress filter triggers while a rule is invalid or blank
  * Reduce visual noise and avoid running incomplete filters

### May 6, 2026

#### Features & Enhancements

* **Dynamic dataset property columns in dataset tables**
  * Add a property column popover and dynamic columns to dataset and dataset version tables, sourcing keys from a new project property-keys endpoint with per-key counts; reserved `valohai.*` keys are filtered out
  * Let users surface custom dataset properties directly in list views
* **Persist dataset property column visibility**
  * Remember which dynamic property columns are shown or hidden in dataset tables across sessions
  * Avoid reconfiguring columns on every visit
* **Store model version in deployment files**
  * Record the model version when a model URL is used to define a deployment endpoint's input, and surface it in the frontend
  * Make deployments traceable back to the specific model version that produced them

#### UI/UX Improvements

* **Structured fields mode for the property editor**
  * Add a structured field-based editor alongside the JSON editor with row add and delete controls and improved focus handling; de-emphasize the JSON editor mode
  * Make routine property edits faster without needing to write JSON

#### Bug Fixes

* **Fix image compare blending for plain images**
  * Use direct CSS `mix-blend-mode` on `<img>` elements for plain images, reserving SVG overlays for tint colors
  * Restore correct visual blending in the image comparison view

### April 28, 2026

#### Features & Enhancements

* **Include metadata in dataset and dataset version list APIs**
  * Add an `?include=metadata` query parameter to dataset and dataset version list endpoints to return metadata properties inline, with a join to avoid N+1 queries
  * Skip extra requests when fetching property data for many datasets at once
* **Filter datasets and versions by property values**
  * Add a Pydantic-based property query parser, dataset and version filters, nested depth guard, and protection against unsafe property names
  * Let users search large dataset collections by structured property values
* **Cursor-style pagination for datums**
  * Add `id_before` and `id_after` filters for datum listings, using ULID ordering as a stable cursor
  * Allow paging through large datum lists until full cursor pagination ships
* **Create executions and tasks without a repository**
  * Let library-step executions run without requiring a project commit, and hide task-only UI until a commit or library step is selected
  * Reduce setup friction for users running library steps without a connected repository
* **Dataset QueryBuilder and property editing (feature flagged)**
  * Add an initial QueryBuilder UI behind the `QUERY_BUILDER` and `DATASET_PROPERTIES` flags, wire property editing for datasets, refresh version properties on save, and sort properties by insertion order
  * Begin rolling out structured property querying and editing to early-access teams

#### UI/UX Improvements

* **Clearer reauthentication prompt**
  * Add context to the "please reauthenticate" message so users understand why it appeared
  * Reduce confusion when sensitive actions trigger a reauth prompt
* **Help text for the environment scale group field**
  * Add inline help text explaining the purpose of the `scale_group` field on environments
  * Make environment configuration easier to understand
* **QueryBuilder polish**
  * Clear the final rule instead of deleting it, add a `not_contains` operator, allow slashes in property keys, and drop `allowedValohaiKeys` validation
  * Smooth out rough edges in early QueryBuilder usage
* **Fix sidebar utility bar icon alignment**
  * Correct icon misalignment in the SidebarUtilityBar
  * Restore consistent sidebar visuals

#### Bug Fixes

* **SSO users with password or MFA can pass reauthentication**
  * Record SSO logins through allauth's `audited_login` so users with a usable password or MFA can complete reauth-protected actions
  * Unblock SSO users who were previously locked out of sensitive operations

### April 21, 2026

#### Features & Enhancements

* **JSON editor for upcoming dataset property editing**
  * Add a JSON-based property editor with relaxed syntax support (single quotes, trailing commas, unquoted keys) and protection for reserved `valohai.*` properties
  * Lay the groundwork for property editing on dataset and dataset version detail pages
* **Rework trace view with server-side graph loading**
  * Replace client-side graph traversal with continuation-based server-side loading and progressive rendering
  * Reduce database queries from \~230 to \~30 for large traces, improving load times for complex lineage graphs
* **Enforce reauthentication for sensitive account operations**
  * Require users to reauthenticate before changing passwords, managing MFA, or generating API tokens
  * Strengthen security for account-critical actions

#### Bug Fixes

* **Fix organization and team name exposure in API responses**
  * Remove organization and team names from team and access-grant API responses where they shouldn't be visible
  * Close an information disclosure issue in multi-tenant environments
* **Fix session cookie security configuration**
  * Correct the session cookie security setting name so it takes effect properly
  * Ensure session cookies are secured as intended

### April 15, 2026

#### Features & Enhancements

* **Preview video files in the output browser**
  * Add in-browser preview support for video files (mp4, webm, mov, avi, mkv, mpeg, ogv, wmv) in the output browser
  * Let users inspect video outputs without downloading files
* **Add minimap and node search to pipeline graph**
  * Add a toggleable minimap (on by default for graphs with more than 8 nodes) and a search input to find and focus on nodes by name
  * Make it easier to navigate and locate nodes in large pipeline graphs
* **Add Inputs tab to Execution view**
  * Move execution inputs into a dedicated tab with inline input titles
  * Give users a focused view for inspecting execution input sources
* **Show properties list on dataset and dataset version details**
  * Display metadata properties as a read-only list on dataset and dataset version detail pages
  * Give users direct visibility into attached properties without needing the API
* **Expand datum directory listing to all views**
  * Enable folder-based directory listing for datums across all contexts, not just within a project scope
  * Allow browsing datum folder structures wherever datums appear
* **Add member email access to organization API**
  * Allow organization admins to request member email addresses via `?include=emails` on the organization detail endpoint
  * Enable admin tooling and integrations that need member contact information
* **Improve OpenAPI documentation coverage**
  * Add missing request body schemas, default field descriptions, and pagination parameter documentation across the API
  * Provide more complete and accurate API documentation for integration developers
* **JSON export for the upcoming query builder**
  * Allow exporting the current query as a JSON document from the query builder interface
  * Make it easier to save, share, and reuse structured queries once the query builder ships

#### UI/UX Improvements

* **Drag-and-drop reordering for the upcoming query builder**
  * Support reordering query conditions by dragging them into position
  * Make it faster to rearrange complex queries without deleting and recreating conditions
* **Upgrade syntax highlighter for code views**
  * Replace the previous syntax highlighting engine with Shiki for improved language support and rendering
  * Provide sharper, more accurate code highlighting across file previews
* **Truncate long property names in the middle**
  * Show the start and end of long property key names instead of cutting off at the end
  * Make it easier to distinguish properties that share a common prefix

#### Bug Fixes

* **Fix crash when using `?only` parameter on API endpoints with nested serializers**
  * Prevent `?only` field selections from propagating into nested serializers where the fields don't exist
  * Eliminate 500 errors when filtering API response fields on endpoints with nested data
* **Fix N+1 query in pipeline detail endpoint**
  * Precompute shared execution IDs in a single aggregation query instead of running per-node queries
  * Improve response times for pipeline detail pages with many nodes
* **Fix Azure uploads failing without storage key**
  * Correct upload info generation for Azure stores configured without a direct storage key
  * Restore reliable file uploads to keyless Azure storage configurations

### April 8, 2026

#### Features & Enhancements

* **Browse files by directory tree in execution outputs and dataset versions**
  * Add a collapsible directory tree sidebar for navigating files by folder path in the execution outputs tab and dataset version contents view
  * Make it easier to find specific files when executions or dataset versions contain deeply nested output paths
* **Dataset and dataset version metadata properties**
  * Add API endpoints for creating, updating, and deleting metadata properties on datasets and dataset versions
  * Allow teams to attach structured key-value properties to datasets and versions, visible in dataset details
* **Staged metadata apply on dataset creation**
  * Apply pending metadata property updates automatically when a dataset version finishes building
  * Prevent metadata from being silently lost when a version completes during property writes
* **Datum multi-purge API**
  * Add an API endpoint for purging multiple datums in a single request
  * Make bulk cleanup faster and reduce the number of API calls needed for large-scale purges
* **Track purge count per dataset version**
  * Record how many datums have been purged per dataset version and show a "partially purged" indicator in the UI
  * Give users visibility into storage reclamation at the version level
* **Execution input origin links and direct URI downloads**
  * Restore the source datum link on execution inputs and make all URI-type inputs trigger a direct download
  * Let users trace an input back to its originating datum and download URI inputs without extra steps
* **Runtime Config Presets visible in restricted environments**
  * Show and allow administering Runtime Config Presets when viewing restricted environments
  * Remove a gap where preset configuration was unavailable for restricted-access environments
* **Upcoming query builder component**
  * Add a new query builder UI component with structured query building and async suggestions
  * Lay the groundwork for upcoming filter and query interfaces across dataset and datum views

#### Bug Fixes

* **Fix terminate button shown on machines that don't support termination**
  * Hide the terminate button for machine types that do not support the termination action
  * Prevent users from triggering an unsupported action that would fail silently
* **Fix dataset project filter leaking across organizations**
  * Correct a bug where filtering datasets by project could return results from other organizations
  * Ensure project-scoped dataset queries are always properly isolated to the correct organization

### March 24, 2026

#### Features & Enhancements

* **Add datum "last used" tracking**
  * Show when and where each datum was last used, with a "Last used" column in the datum browser and usage details in the datum panel
  * Help users identify stale data and understand how files flow through their workflows
* **Delete dataset versions with optional file purge**
  * Add delete buttons for dataset versions with a toggle to also purge their underlying data files
  * Let users reclaim storage directly when removing dataset versions, without a separate cleanup step
* **Support `project:counter` retrieval for Pipeline API**
  * Allow fetching pipelines by project-scoped counter in addition to UUID
  * Bring the Pipeline API in line with other APIs that support shorthand lookups

#### UI/UX Improvements

* **Faster output preview grid**
  * Load preview images on demand as you scroll instead of rendering them all at once
  * Improve scrolling performance when previewing large numbers of output files

#### Bug Fixes

* **Fix large dataset version comparison crash**
  * Resolve a crash when comparing dataset versions with a large number of datums
  * Restore reliable comparison for heavy dataset versions
* **Fix copied pipeline commit change resetting too much state**
  * Preserve compatible parameter values when switching commits on a copied pipeline
  * Prevent losing manual configuration changes during commit switches
* **Fix back-navigation trap on deployment endpoints**
  * Fix the back button getting stuck in a loop when navigating away from a deployment endpoint
  * Pressing back now returns to the previous page as expected
* **Handle dataset loading errors gracefully**
  * Show proper error states when a dataset fails to load (e.g., after deletion) instead of crashing
  * Prevent broken UI when navigating to a removed dataset
* **Fix plural labels in admin log entries**
  * Correct verbose name plurals for several models displayed in admin and log views
  * Ensure consistent grammar in system messages and audit logs

### March 19, 2026

#### Features & Enhancements

* **Search datums by ID and datum:// URI**
  * Add support for searching datums by their UUID or datum:// URI in the datum browser
  * Make it faster to locate specific data files when you already have a reference
* **Show dataset version inputs in deployment version view**
  * Display dataset version inputs alongside file inputs in the deployment endpoint details, with a type column and URI references
  * Give users full visibility into all input sources attached to a deployment version
* **Improve pipeline input error messages**
  * Show human-readable error messages when pipeline inputs are misconfigured
  * Help users diagnose and fix pipeline setup issues without reading raw error payloads

#### UI/UX Improvements

* **Truncate long input source lists in execution details**
  * Collapse large datum and URI input lists with an expand option
  * Prevent execution detail pages from becoming unwieldy when inputs have many sources
* **Improve execution input list layout**
  * Compact datum input rows and streamline the visual style of execution inputs
  * Make the execution input section easier to scan
* **Reword configuration snippet help text**
  * Clarify help text for configuration snippets in both admin and public views
  * Reduce confusion about how config snippets work

#### Bug Fixes

* **Fix session key exposure in API responses**
  * Remove the full HTTP session key from API output
  * Close a security issue where session identifiers could leak to clients
* **Handle advisory lock timeouts in task execution creation**
  * Catch lock timeout errors when tasks try to create child executions concurrently
  * Prevent hard failures when multiple task nodes compete for execution creation
* **Fix broken external links**
  * Correct external link URLs that pointed to wrong destinations
  * Ensure documentation and help links navigate correctly
* **Avoid redundant git commit fetching**
  * Check for existing commit records before performing a full fetch
  * Reduce unnecessary git operations when commit data is already available

### March 11, 2026

#### Features & Enhancements

* **Show execution input sources grouped by type**
  * Display execution inputs organized by source (dataset versions, model versions, URIs) with clickable links in the execution detail view
  * Make it easier to trace where each input came from and navigate to the source
* **Preview tabular data files (JSONL, NDJSON, Parquet)**
  * Add in-browser preview support for JSONL, NDJSON, and Parquet files in the output browser
  * Allow users to inspect tabular output data without downloading files
* **Restrict views for external model hub users**
  * Limit model-hub-only users to model views and hide execution-related links and actions
  * Provide a focused experience for external collaborators who only need model access
* **Make Snowflake stage cleanup optional**
  * Allow keeping Snowflake staging data after execution runs instead of always cleaning up
  * Give users control over stage lifecycle for debugging or reuse purposes
* **Optimize API list endpoint performance**
  * Apply serialization caching to 13 additional list serializers for shared related objects
  * Improve response times for list endpoints where items share owners, projects, or environments
* **Add OIDC identity provider support**
* Enable OpenID Connect as an authentication method for the platform
* Allows using Valohai credentials for single sign-on to Valohai LLM

#### UI/UX Improvements

* **Improve execution input detail layout**
  * Use smaller badges for input types, add collapsible input groups with file counts, and display formatted datum queries
  * Make the execution inputs section more compact and scannable
* **Show deployment target details in deployment detail view**
  * Display target base URL and configuration directly in the deployment detail page
  * Reduce navigation needed to find deployment endpoint information
* **Collapse long tag lists in dataset table cells**
  * Truncate long tag lists in dataset tables with an expand option
  * Prevent tag-heavy datasets from stretching table rows
* **Support dark mode for ClearFeed chat widget**
  * Apply dark mode styling to the in-app support chat
  * Match the chat widget appearance with the platform's dark theme
* **Improve CORS handling for Gateway API deployments**
  * Implement manual CORS headers for Kubernetes Gateway API endpoints
  * Ensure cross-origin requests work with gateway controllers that lack native CORS filter support
* **Don't block UI while stores are loading**
  * Allow data views and organization settings to render before store data finishes loading
  * Remove unnecessary loading delays when navigating to data or settings views
* **Improve store error messages**
  * Show clearer error messages when a store endpoint URL is misconfigured
  * Help users diagnose and fix store connection issues faster

#### Bug Fixes

* **Fix datum alias display in execution inputs**
  * Resolve incorrect URI handling for datum aliases in the input source view
  * Ensure aliased data inputs display and link correctly
* **Fix metadata chart selection leaking to single execution views**
  * Scope chart selection state to prevent it from carrying over between multi-execution and single-execution views
  * Eliminate unexpected chart configurations when switching views
* **Fix empty string in deployment target routing**
  * Handle empty target routing values that prevented the default target from being selected
  * Restore correct default target assignment for deployments
* **Handle missing datum in deployment version UI**
  * Account for null datum IDs when displaying deployment version endpoint details
  * Prevent UI errors when a deployment version references a deleted or missing datum
* **Filter out missing datums from endpoint API responses**
  * Exclude datums that no longer exist from deployment endpoint detail responses
  * Prevent API errors when endpoint files reference removed data
* **Handle access control for org-owned deployments without a project**
  * Fix a 500 error when accessing dataset or model version inputs on org-owned deployments that lack a project
  * Ensure deployment version input access checks work regardless of project association

### February 26, 2026

#### Features & Enhancements

* **Add model commenting system**
  * Introduce commenting functionality for models with user tagging support
  * Enables collaboration and discussion around model versions
* **Support Gateway API for Kubernetes deployments**
  * Add support for Kubernetes Gateway API routing alongside Ingress
  * Provides modern, flexible routing options for deployment endpoints
* **Add SME access grant support**
  * Enable Microsoft Entra ID integration for SME workflow access
  * Simplifies authentication for subject matter experts in enterprise environments
* **Soft-delete model comments**
  * Allow deletion of comments while preserving metadata and conversation context
  * Maintains audit trail while removing inappropriate or outdated content
* **Optimize Datum API performance**
  * Implement serialization caching and optimize URL generation
  * Significantly improves response times for datum listing endpoints

#### UI/UX Improvements

* **Rename Notes to Comments throughout UI**
  * Update terminology from "Notes" to "Comments" for consistency
  * Applies to models, executions, and tasks
* **Add axis labels to confusion matrix**
  * Display predicted and actual labels on confusion matrix visualizations
  * Improves readability and interpretation of classification results
* **Hide "Show all outputs" for SME users**
  * Remove output filtering option from SME interface
  * Simplifies workflow result viewing for non-technical users

#### Bug Fixes

* **Fix missing deployment node edges**
  * Restore missing edge connections in deployment graph visualization
  * Ensures deployment pipeline structure is correctly displayed
* **Fix dataset version links with trailing slash**
  * Correct URL generation for dataset version references
  * Prevents broken links in dataset navigation

***

### February 19, 2026

#### Features & Enhancements

* [**Add output preview grid tab to Compare Executions view**](https://docs.valohai.com/changelog/announcements.md#output-preview-grid-for-compare-executions)
  * Add an image grid tab with synchronized hover state and a filter field for comparing execution outputs visually
  * Lets users compare output artifacts (images, plots) side by side across executions
* **Find dataset by UUID in addition to name**
  * Accept dataset UUIDs when creating dataset versions, preventing accidental creation of a new dataset named after the UUID
  * Reduces errors when referring to datasets programmatically by ID
* **Disallow UUID and 'latest' as dataset version names**
  * Reject reserved names during dataset version creation to avoid ambiguity with lookups
  * Prevents confusing collisions between version names and system identifiers
* **Improve pipeline creation performance**
  * Create pipeline edges in bulk and skip redundant FK validation and sanity checks
  * Speeds up creation of large pipelines with many edges
* **SME: Revalidate commit list after repository fetch**
  * Refresh the commit list and update selection after fetching the repository in the SME create view
  * Ensures users always see the latest commits when creating a new workflow

#### UI/UX Improvements

* **Allow any value for grouped plot filter values**
  * Remove restrictions on filter value input in the grouped metadata plot
  * Gives users full flexibility to filter on arbitrary metadata values

#### Bug Fixes

* **Fix project URL resolution for projects with 'datasets' in the name**
  * Match URL fragments properly instead of the first occurrence of reserved keywords
  * Fixes broken navigation for projects whose names contain words like "datasets" or "dataset-version"

***

### February 11, 2026

#### Features & Enhancements

* [**Add execution selector to Compare Executions view**](https://docs.valohai.com/changelog/announcements.md#execution-selector-for-comparing-executions)
  * Add an execution selector sidebar for choosing which executions to compare, with filtering support
  * Makes it easier to pick and switch between executions in comparison workflows
* **Automatically use referenced task blueprints on task nodes**
  * Resolve task node steps through their associated task blueprint instead of requiring manual configuration
  * Simplifies pipeline setup when task nodes reference existing blueprints
* **Support runtime config preset YAML override in API**
  * Convert hyphenated YAML configuration keys to underscored format when processing runtime config presets
  * Ensures YAML-defined presets work correctly through the API without manual key conversion

#### UI/UX Improvements

* **Improve grouped metadata plot**
  * Add filtering by metadata values and support grouping by metadata values in addition to keys
  * Brings pivot-table-like interactivity to the grouped chart view
* **Use trace-specific colors for min-max shadows in statistics graphs**
  * Replace hardcoded gray with per-trace colors (blue for CPU/Memory, orange for GPU)
  * Makes it easier to distinguish overlapping resource utilization ranges

#### Bug Fixes

* **Fix multiple metadata charts sharing the same configuration state**
  * Scope metadata chart configuration per view to prevent charts from interfering with each other
  * Eliminates unexpected behavior when switching between views with metadata charts
* **Prevent filter state leakage between DatumBrowser views**
  * Assign unique view names and sanitize hidden filter keys when navigating between output views
  * Stops filters from one view carrying over into unrelated datum browsers
* **Fix input focus styles**
  * Restore correct focus ring styling on form inputs
  * Ensures keyboard-navigating users can clearly see which input is active

***

### February 3, 2026

#### Features & Enhancements

* **Add grouped metadata plot**
  * Introduce a grouped chart view with box plots and a statistics table for metadata, supporting grouping by categorical keys
  * Helps users compare distributions across experiment groups at a glance
* **Add SME workflow blueprint status frontend**
  * Allow managing workflow blueprint status from the SME interface and only show valid workflows in the start flow
  * Gives teams direct control over which workflows are visible and launchable
* **Allow store owner to delete project-scoped stores**
  * Grant project store creators the ability to delete their own stores from the UI
  * Further reduces dependency on organization admins for store lifecycle management
* **Support runtime config preset YAML override in Create Pipeline UI**
  * Apply runtime-config-preset overrides when creating pipelines from the UI
  * Ensures pipeline creation respects preset configurations consistently

#### UI/UX Improvements

* **Improve table loading and empty states**
  * Fix and refine the Tablor component's loading and empty states to avoid layout shifts
  * Delivers a smoother experience when tables are loading or have no data

***

### January 27, 2026

#### Features & Enhancements

* **Add SME workflow run history view**
  * Allow SME users to view full workflow run history with context-aware headers, pagination, and workflow name display
  * Gives subject-matter experts visibility into past runs without navigating away from the SME interface
* **Add SME workflow blueprint status API**
  * Introduce a status field on workflow blueprints so blueprints can be marked active or inactive via API
  * Allows teams to control which workflows are available for launching
* **Validate required assigned inputs in SME workflow creation**
  * Prevent creation of workflow versions when required assigned inputs lack a value
  * Avoids workflow failures caused by missing mandatory inputs
* **Allow setting associated projects in execution output metadata**
  * Support a list of associated project IDs in the `valohai.dataset-versions` output metadata property
  * Enables automatic cross-project dataset association from execution outputs
* **Add `output_execution_counters` filter to datum API**
  * Allow filtering data by execution counter in the datum browser and API
  * Makes it easier to locate outputs from specific executions
* **Allow store owner to administer project-scoped stores**
  * Grant project store creators the ability to manage their own stores
  * Reduces dependency on organization admins for routine store management
* **Add UI to purge outputs when deleting pipelines and tasks**
  * Add purge-output toggles to pipeline and task delete dialogs and table actions
  * Lets users clean up outputs directly from the UI during deletion
* **Only show organization tokens for organizations that have them enabled**
  * Gate organization token visibility by the organization-level feature flag instead of user privilege
  * Prevents confusion by hiding token management from organizations where the feature is not active
* **Add audit logging to store secret administration**
  * Log changes made through the store secret configuration admin interface
  * Improves traceability and accountability for sensitive store configuration changes

#### UI/UX Improvements

* **Overhaul Compare Dataset Versions view**
  * Replace the cramped modal with a full-page view, add checkbox selection, a swap button, and improved comparison layout
  * Makes dataset version comparison more readable and easier to navigate
* **Improve DatumBrowser loading state**
  * Add integrated loading indicators for datum queries and a separate loader for the side panel
  * Provides clearer feedback when data is being fetched or filtered
* **Make maximum number of metadata property keys configurable**
  * Allow the property key limit to be adjusted instead of using a hard-coded cap
  * Supports projects with large numbers of metadata keys without truncation
* **Replace Zendesk chat with ClearFeed web chat**
  * Swap out the Zendesk widget for ClearFeed-based support chat
  * Provides a refreshed in-app support experience

#### Bug Fixes

* **Fix copying executions that use environment variable groups**
  * Resolve a regression where copying executions with EVGs failed due to strict typing
  * Restores the ability to reliably copy and re-run executions that reference variable groups
* **Return proper error when stopping an already-stopped resource**
  * Raise `InvalidStatusAPIException` instead of a generic error when attempting to stop a resource that is already stopped
  * Gives callers a clear, actionable error message
* **Block non-allowed email domains on social signup**
  * Enforce the email blocklist for social login (OAuth) signups, not just email/password registration
  * Closes a loophole where restricted email domains could bypass the allow-list via social auth

***

### January 14, 2026

#### Features & Enhancements

* [**Add transient environment variables**](https://docs.valohai.com/changelog/announcements.md#transient-environment-variables)
  * Allow users to define transient environment variables via API and manage them through a new editor in user and organization settings
  * Enables injecting short-lived, secret-capable variables into job payloads without persisting them in project configuration
* **Add API to purge outputs when deleting pipelines and tasks**
  * Provide a `purge_outputs` option on pipeline and task delete endpoints
  * Lets users reclaim storage in a single API call instead of manually cleaning up outputs after deletion
* **Support model version URL as deployment endpoint input**
  * Accept a model version URL when creating deployment endpoint files, automatically converting it to a dataset version URL
  * Simplifies deployment version creation by allowing direct model references
* **Add deployment version model input UI**
  * Add a model version input field to the Create Deployment Version page
  * Makes it easier to attach model versions to deployment endpoints directly from the UI
* **Default to execution hash v2**
  * Switch the default execution hashing algorithm to v2
  * Improves execution deduplication accuracy
* **Exclude interactive executions from hashing**
  * Skip hash computation for interactive (notebook/SSH) executions
  * Avoids misleading hash values for executions that are inherently non-reproducible

#### UI/UX Improvements

* **Debounce datum metadata query input**
  * Add input debounce and Enter-key flush to the metadata query field
  * Reduces unnecessary API calls while still allowing instant search on demand
* **Parse true/false as boolean in datum queries**
  * Recognize `true` and `false` as boolean values in metadata query input
  * Allows users to filter on boolean metadata without special syntax

#### Bug Fixes

* **Ensure deleted triggers do not run**
  * Prevent soft-deleted triggers from being evaluated and executed
  * Eliminates unexpected pipeline or workflow runs from stale trigger configurations
* **Catch missing job data exceptions gracefully**
  * Handle cases where job data is absent instead of raising an unhandled exception
  * Prevents 500 errors when viewing executions with incomplete backend data

***


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.valohai.com/changelog/archive/updates-2026.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.
