# May 2026 {% updates format="full" %} {% update date="2026-05-27" %} ## Shareable threads, live updates This release makes agent threads easier to share and follow in real time, alongside polish to AWS KMS, invitee, and environment workflows. ## 🚀 New features ### Copy link to thread from the agent thread header Added a "Copy link to thread" option in the agent thread header menu. One click grabs a shareable URL to the current thread. **What does it mean for you?**\ Sharing an agent thread is now a single click. *** ### View manifest action on the Environments page Added a "View manifest" action on environments, matching the existing blueprint flow. The inline viewer includes syntax highlighting, search, and copy. **What does it mean for you?**\ You can review collection deployment manifests in place, instead of downloading them and opening an external editor. *** ## ⚡ Improvements ### Realtime updates for the agent thread list The agent sidebar now consumes a dedicated WebSocket. Threads currently in progress, newly created integration threads, and thread metadata all refresh without a refetch. **What does it mean for you?**\ Active threads update live in the sidebar, and integration threads appear the moment they're created, without you having to refresh the page. *** ### Cleaner markdown rendering in agent responses Improved paragraph spacing, section hierarchy, and pre-formatted/code block rendering so long agent answers are easier to scan. ASCII and preformatted blocks now hug their content properly. **What does it mean for you?**\ Long agent responses are easier to read. Headings, paragraphs, and code blocks all have room to breathe. *** ### Re-selectable collection in the "Create env from cloud" drawer After picking a collection in the modal, you can now change it from inside the drawer instead of having to bail out and restart the flow. **What does it mean for you?**\ Switching collections mid-flow no longer means starting over. *** ### Dedicated AWS KMS (BYOK) card in Integrations BYOK configuration is no longer nested inside the Bluebricks API Key section. It now has a standalone AWS KMS card with its own detail page. **What does it mean for you?**\ Finding and managing your AWS KMS encryption configuration is easier, with a first-class card alongside the other integrations. *** ### Role and sorting on the invitee list The Members → Invited tab now shows each invitee's role and sorts the most recent invite at the top. **What does it mean for you?**\ You can see at a glance who's been invited with what role, and recent invites stay visible instead of getting buried. *** ### Thread list shown when the agent sidebar is minimized Clicking Alert/Recent Threads in the collapsed sidebar now reveals the full list of threads, instead of jumping straight to the latest thread. **What does it mean for you?**\ The collapsed sidebar is now usable for thread navigation. You can browse threads without expanding the full sidebar first. *** ## 🐛 Bug fixes * **Fixed lingering "response interrupted" notification**: Stopping a thread mid-run left the "response interrupted" notification on screen even after posting more messages or switching threads. The notification now clears once the next action happens. * **Fixed sidebar menu scrolling up on click**: Clicking items in the sidebar (Collections, Packages, Envs) caused the sidebar to auto-scroll up and push other items out of view. The sidebar now keeps its scroll position. {% endupdate %} {% update date="2026-05-20" %} ## Approve before you deploy This release adds human-in-the-loop confirmation for the agent's destructive operations, names alert investigation threads automatically, expands the context graph to EKS Pod Identity, and polishes a stack of agent and chat behaviors. ## 🚀 New features ### Confirmation gate for write tools in the agent The agent now pauses for human approval before running destructive tools like deploy, approve, reject, and rotate-secret. While a confirmation is pending it surfaces as a pinned card above the prompt. **What does it mean for you?**\ The agent stops for a human check before touching anything destructive. You're in the loop on consequential actions before they happen. *** ### Auto-generated names for alert-driven investigation threads Investigation threads created from alert integrations (e.g. Coralogix) now get concise, agent-generated titles based on the investigation summary, instead of identical placeholder names. **What does it mean for you?**\ You can scan the alert threads sidebar and tell incidents apart at a glance. *** ### EKS Pod Identity and Kubernetes ServiceAccount in the context graph The agent can now answer Pod Identity and least-privilege questions directly from the context graph, instead of cloning your IaC repo and grepping through it. **What does it mean for you?**\ Questions about who can access what come back from graph state rather than ad-hoc source-code searches. *** ## ⚡ Improvements ### Keep agent thread runs alive across navigation Switching between agent threads no longer aborts the run in the thread you just left. **What does it mean for you?**\ You can kick off a long-running agent task, switch over to another thread to follow up on something else, and come back to find the first one still working. *** ### In-progress thinking state survives reloads The agent's in-progress response is now persisted while it's running, so the thinking state recovers correctly after a page reload or thread switch. When the run finishes (success, failure, or callback) the in-progress draft is promoted to the final message. **What does it mean for you?**\ Reloading the page during a long agent run no longer loses the thinking context. You pick up exactly where you left off. *** ### Elapsed time on active agent runs Active agent runs now show elapsed working time inline, not just the current working phrase. **What does it mean for you?**\ Long agent runs feel less stuck. You can see at a glance how long the agent has been working. *** ### Side menu indentation polish Bumped the indentation between collapsible side-menu sections and their child items, and restored the section icons. **What does it mean for you?**\ Stronger visual hierarchy in the side menu makes it easier to scan and find what you need. *** ### AWS orchestration and discovery indicators in new cloud creation When creating a new AWS cloud account inside a collection, the Role ARN fields now show the orchestration/discovery indicator next to each role. **What does it mean for you?**\ No more guessing which ARN field is for which role. *** ### Clearer error messages in the Cloud Import Assistant Import-job failures in CIA now surface clearer, more actionable error messages instead of generic failure copy. **What does it mean for you?**\ When an import fails, you get a real explanation and a hint at what to do, instead of staring at a vague error. *** ## 🐛 Bug fixes * **Fixed stopped agent tasks that kept running in the background**: Manually stopping a thread mid-run didn't always propagate to the agent. You could navigate away, come back, and find the "stopped" task had completed. The stop signal now reaches the agent and runs actually halt. * **Fixed threads that silently stopped with only a spinner**: Some agent threads stopped processing with only a loading spinner visible, no error and no resolution. The thread now surfaces an explicit state instead of silently hanging. * **Polished agent chat scroll behavior**: Three related scroll fixes. The reasoning panel no longer yanks you back down when you scroll up to read it. The chat respects your scroll position while the agent is working. And sending a new user message now auto-scrolls to it correctly. {% endupdate %} {% update date="2026-05-13" %} ## Connect GitHub, share threads, see Kubernetes This release adds GitHub OAuth connect inside the agent, view-only access to shared agent thread URLs, and live Kubernetes object indexing in the context graph, alongside a batch of account and UI fixes. ## 🚀 New features ### GitHub repository connect in the agent When the agent needs Git access to a repository it now surfaces a permissions card right in the chat. Authorize via OAuth, the agent ingests the code, then continues. Success and failure banners are shown inline with a retry option. **What does it mean for you?**\ You can hand the agent access to a repo without leaving the conversation. The agent picks up your code on the spot and keeps moving. *** ### View-only access to shared agent thread URLs Share an agent thread URL with anyone in your org and they can open and read the conversation, even if they didn't start it. Viewers can read but not post. For collection-scoped threads, viewers need access to the same collection. **What does it mean for you?**\ Easier to share investigation context with teammates without forwarding screenshots or summaries. They see the full thread, and the agent stays in your hands. *** ### Kubernetes object indexing in the context graph The context graph now ingests live Kubernetes objects, not just static Helm templates. The agent can inspect failed Jobs, correlate Kubernetes objects with their Helm releases, and trace them back to the IaC that deployed them. **What does it mean for you?**\ When something breaks in a cluster, the agent can answer with real, live data: what's running, why it failed, and which blueprint owns it. *** ## ⚡ Improvements ### Per-message user info for multi-user agent threads Agent thread responses now carry user info per message (for integration-created threads) or at the thread level (for user-created threads). In integration threads, message avatars now show the actual sender rather than the logged-in viewer. **What does it mean for you?**\ In threads coming from Slack, Coralogix, or other integrations, you can see exactly who said what, even when multiple people interact with the same thread. *** ## 🐛 Bug fixes * **Fixed org member search by email beyond the first page**: Searching organization members by full email address only returned matches if the user appeared on the first 10 results. Search now works for members anywhere in the list. * **Fixed invite members API silently failing**: Inviting someone already in the organization returned HTTP 200 with the failure tucked into an `errors[]` array, so the UI showed nothing. Failures now surface properly with appropriate status codes. * **Fixed code editor drawer closing on diff view switch**: On the deployment plan page, switching the diff view type from inside an open JSON chip's code editor was closing the drawer. The drawer now stays open through diff type changes. * **Fixed Create Collection submit button stuck on cloud error**: When creating a collection with a new cloud provider, if the cloud create failed (e.g., duplicate cloud discovery ID), the submit button got stuck in Processing and the user couldn't retry. The button now resets cleanly on error. {% endupdate %} {% update date="2026-05-06" %} ## Sharper agent context, alert-driven threads This release brings quote-reply and paste-to-quote to the agent, introduces Coralogix-powered alert threads, and ships a batch of agent and UI fixes. ## 🚀 New features ### Quote-reply chips and paste-to-quote in the agent Two new ways to attach context to an agent message. Select text on the page to get a "Reply" tooltip, or paste long or multiline content and it auto-wraps as a quote chip. The quoted text is sent with your message so the agent knows exactly what you're referencing. **What does it mean for you?**\ You can point the agent at the precise text you want it to consider, without copy-pasting paragraphs into the prompt. Answers stay grounded in what you actually meant. *** ### View thread button on user messages Each user message now has a "View thread" button that takes you back to the original conversation context. **What does it mean for you?**\ Jumping between message references and full threads is a single click. Easier to pick up a paused conversation or trace where a question came from. *** ### Coralogix integration setup flow We added a new integration that turns Coralogix alerts into agent threads. A new setup flow walks you through connecting a Coralogix workspace. **What does it mean for you?**\ Incidents and alerts from Coralogix can now flow into Bluebricks as investigation threads, so on-call work and infra context live in the same place. *** ### Alert threads section in the agent side menu Threads created from external alerts (starting with Coralogix) now have their own "Alert threads" section in the agent side menu, separate from regular conversations. Coralogix branding shows on the alert row and the thread header. **What does it mean for you?**\ You can spot alert-driven investigations at a glance and keep them visually separate from your day-to-day agent conversations. *** ## ⚡ Improvements ### Better handling of stale active threads When you open an agent thread that's marked active but has no live connection (for example, after a backend timeout), the UI no longer hangs on "Processing" forever. It now surfaces a "This response timed out" message with a retry option. **What does it mean for you?**\ No more waiting indefinitely on threads that are silently dead. You see what happened and can retry without leaving the page. *** ## 🐛 Bug fixes * **Fixed stream resume errors after multiple page reloads**: Refreshing the browser repeatedly during agent streaming would surface "Unable to connect to the agent due to an issue on our end." This was a multi-pod scoping issue and is now resolved. * **Fixed agent threads stuck on "Processing" indefinitely**: A thread whose tool call never produced a final answer would hang on "Processing" forever after a refresh. The thread now resolves to a proper timeout state. * **Fixed feedback submission after a stopped response**: After stopping a streaming response, clicking "No" in the helpfulness prompt and submitting feedback failed with "Something went wrong." Feedback now submits correctly even when the response was stopped mid-stream. * **Fixed run name and PR misalignment on the run plan page**: Opening a plan URL directly (with no back-button context) caused the run name and PR data to render misaligned. Now aligned regardless of how you got there. * **Fixed subdir and ref fields cleared on URL edit when adding artifacts**: In the "New Package" flow, pasting a git URL with subdir/ref and then editing the URL was wiping the parsed subdir and ref fields. They're now preserved. * **Fixed agent scope "select/unselect all" ignoring active filters**: With a filter active in the agent scope dropdown, "select all" and "unselect all" only applied to the filtered subset. They now consistently apply to all collections regardless of filter. {% endupdate %} {% endupdates %} --- # Agent Instructions: Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter: ``` GET https://bluebricks.co/docs/changelog/readme.md?ask= ``` The question should be specific, self-contained, and written in natural language. 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.