What's new in Spectr

Product Delivery is now something your AI Assistant and the agents you connect over MCP can drive, not only read. Ask one to plan a project and it makes the milestones and epics from your requirements, the same as the Delivery page's Generate; ask for a change and it runs Update with AI, which comes back as suggested changes for you to review and apply - nothing changes until you do. An agent can also file a story under an epic. These need the "read and write" permission you grant, run as you and within your role, and each generate or update with AI uses one delivery revision from your monthly allowance, exactly as the button in Spectr does; reviewing, applying and filing a story are free.

A connected agent can now read your own Spectr profile - your name, your email and the other names meetings call you, your aliases - and update those aliases for you. It is only ever your own profile, whatever your role, and never your password or sign-in. Your aliases are how Spectr matches you to what people call you in meetings and trackers, so keeping them current is an easy thing to hand to your agent.

Spectr now plans how a project gets delivered. On the new Delivery page, one click reads every requirement in the project and groups the work into epics, and the epics into milestones in the order they should ship - each milestone dated from how much work it holds, laid out as a roadmap. Every story in the project is filed under its epic, and new stories are filed as they are generated, so each epic shows real progress from the statuses Spectr already syncs from Linear, Jira and Shortcut. When requirements change, or you want something different, Update with AI suggests the changes - renames, new epics, splits and merges, moved dates - for you to review and apply one by one; nothing changes until you do, and story links always survive. You can also rename, reorder, date and move everything by hand. Each run uses one delivery revision from a new monthly allowance (5 on Free, 25 on Starter, 100 on Pro, 500 on Business); reviewing, applying and editing by hand are free. Your AI Assistant and the agents you connect over MCP can read the roadmap too.

If you run more than one Spectr workspace, each one is now billed entirely on its own. Put every workspace on the plan that fits it, and its usage, credits and charges stay with that workspace alone - subscribing to or paying for one never touches another, even when you own them all. A single workspace sees no change, and every subscription you already have keeps working exactly as it is. The Billing page is clearer when something goes wrong, too: if a checkout or the billing portal can't open, it now tells you why instead of doing nothing.

When you connect Spectr to an AI agent over MCP, every Spectr tool now carries a plain, readable name and an honest label for what it does - whether it only reads your data or can change it - so before the agent runs a tool, both you and it can see exactly what it will touch. Each name says only what the tool actually does, so one that simply renames something no longer reads as doing more. These are the same tools, with the same permissions you already granted - just clearer about themselves - and the change is part of listing Spectr in the Claude and ChatGPT tool directories.

You can now add Spectr to Claude straight from Claude's connector directory: find Spectr under Customize > Connectors, or open claude.ai/directory/spectr, choose Connect, and sign in through Spectr. There is no address to copy. It is the same connection as before, with the same permissions: you choose which workspaces Claude can reach, it reads only what you can see, and it changes something only if you granted "read and write" and it checks with you first. Connections you already added by address keep working, and Claude Code and the other agents still connect by address.

Spectr now has a Development board, where approved work turns into code. Pull related stories into a workstream, move it across Pending, In Progress, In Review and Done - then archive it when the work is over - and start building by handing a workstream's stories to a cloud coding agent, with each story's full detail sent along as the brief. The board is free on every plan and uses none of your Spectr AI allowance; the agent runs on your own account. Anyone on the workspace can start a run, every run is tracked back on the board with who started it, and a workspace admin can unlink any run. Your AI Assistant and the agents you connect over MCP can drive the whole thing too - grouping stories into workstreams, and starting, linking or tracking runs - all as you, within your role, and always showing the exact message before it is sent.

Connect your own Conductor account under User Settings > Connections, and Spectr can start a Conductor agent run straight from a Development workstream - sending the stories you choose as the run's brief, then tracking each run's state back on the board. It uses only your own Conductor account and compute, so it costs nothing against your Spectr usage, and several people can run agents on the same workstream, each on their own account. Your key is checked before it is saved, stored encrypted, and never shown again. Conductor is the first coding agent Spectr connects to - Claude, Codex and Cursor are meant to follow.

Spectr's AI Assistant and the AI agents you connect over MCP can now take actions, not just read. They can generate a spec sheet from a brief, write and revise requirements, update your AI Memory and team roles, refresh your budget, story statuses and product health, and change project and budget settings - the same work as the buttons in Spectr. Nothing happens behind your back: in the app the assistant proposes each change and waits for you to confirm it, and an agent connected over MCP needs the "read and write" permission you grant when it signs in, with every change marked so it checks with you first. Either way it acts only as you, within your own role, and does only what you asked - never on instructions tucked inside a meeting or a document. Each action runs the same checks and counts toward your workspace's monthly usage exactly as it would in the app. The assistant's actions are on Starter and up; agents can connect on every plan.

Spectr is out of beta. The beta label is gone from the product and the site - Spectr is ready to run your team's day-to-day work. It lands with a refreshed look across the whole app and marketing site: a cleaner, sharper, more consistent design on every page you already use. Nothing has moved and there is nothing to relearn - the same projects, the same data and the same shortcuts, in a calmer place to do the work.

The AI Assistant now comes to you. A new Ask Assistant button on any meeting, spec sheet or requirement opens the assistant in a drawer beside what you are reading, already focused on that item - so you can ask about it without leaving the page or setting the scope yourself. The drawer stays open while you read the document next to it, shows what the conversation is about, and lists the earlier chats you started from the same page so you can pick one back up. It is part of the AI Assistant, available on Starter and up.

The lighter, faster work behind Spectr - distilling the context it gathers, keeping your AI Memory up to date, naming your conversations, matching team members to roles and spotting duplicate stories - now runs on Anthropic's Haiku 5.5, so that background work is quicker and cheaper with no change to how you use Spectr. Haiku 5.5 also joins the AI Assistant's model picker as its fastest, most economical option, replacing Haiku 4.5 alongside Sonnet 5.5 and Opus 5.5. Any conversation you already started on Haiku 4.5 keeps running on it, so nothing in flight changes underneath you.

Your own AI tools can now read your Spectr data as you. Spectr has an MCP server, so an agent like Claude, ChatGPT, Cursor or VS Code can pull in your spec sheets, stories, requirements, AI Memory, budget, team and more - right where you are already building. You add one address to your agent (find it, with a copy button and per-agent setup guides, under User Settings > Connections), sign in through Spectr, and choose whether the connection can reach all your workspaces or only the ones you pick. It is read-only and sees exactly what you can see - never more than your own role allows - and you can disconnect it at any time, which stops it on its very next request. It is on every plan, and reading your data this way costs nothing against your usage.

New to Spectr? Your dashboard now has a Getting Started checklist that walks you through the first things worth doing - creating a project and describing it, connecting your tools, importing a meeting, and generating the spec sheets, requirements and stories that come out of it. It keeps up as you work, ticking each step off the moment you actually do it, and shows the steps that fit your role in the workspace. Collapse it when you want the space, or hide it entirely once you are up and running - and reopen it any time from the 'Getting started' entry in the sidebar. It is also where onboarding now hands you off, so there is always a clear next step.

If you belong to more than one workspace, the switcher in the sidebar now shows who owns each one you were invited into, right next to your role there - "Viewer, owned by Jordan Lee" - so it is clear at a glance whose workspace is whose. Your own role always stays visible; a long owner name is shortened, with the full name on hover.

The other names you go by - the 'Also Known As' aliases that let Spectr match a meeting mention to the right teammate - have moved out of their own card and into the Profile Information section of your settings. Your name, email and aliases now save together with a single button, and each is only sent when you have actually changed it, so an SSO display name outside the length limit no longer blocks an alias edit (or the other way around). A name typed into the alias box but not yet added as a chip is saved too.

Joining a workspace you were invited to is quicker now. The sign-up form prefills the invited email and locks it to the invitation, and because the invite already proves the address is yours, signing up with it skips email verification entirely - your account is ready right away, with your trial started. Onboarding then runs in the workspace you own, and its final step gives you a 'Go to <workspace>' button for each team you were invited into, so you can set up your own space first and jump straight to the one you came for.

When you paste text to create a requirement, you can now choose whether Spectr saves it exactly as you pasted it or rewrites it into a full requirement document for you. The Paste tab has a new 'Enhance with AI' switch - off by default, so a paste still saves word for word, instantly, without using any of your plan's AI allowance. Turn it on and Spectr enhances the text the way it handles an upload: plan-checked, generated in the background and retryable if it fails, with room to add your own instructions to steer the result. Enhanced pastes and file uploads now also gather context from your connected tools while they run, so the document reflects more than just the text you handed over - and there's a live character count in the editor so you always know where you stand against the limits.

Starting a requirement from Notion no longer means visiting the Integrations page first. From the Notion tab of a new requirement, admins can connect or reconnect Notion and choose the pages to import right there, with a fresh page pre-selected to save a click. Every import tab also now tells you roughly how long it takes and that AI imports keep running in the background, so you can start one and come back to it later.

If a Notion import fails, Spectr now tells you exactly what went wrong - the top page couldn't be reached, some sub-pages couldn't be read, the page tree was too large, or Notion isn't connected any more - and offers the one action that fixes it, linking you straight to reconnect when that's the cause. It no longer keeps retrying a failure that can't succeed, and any pages it couldn't read are flagged on the project's Notion card until you reconnect, reselect them, or a later import reads them. And if you'd rather fix things by hand, editing a failed requirement clears its failed state and turns it back into an ordinary draft.

A handful of upgrades to connecting your tools. The GitHub repo picker now loads all of your repositories instead of just the first page, and links you to GitHub to grant access to more organisations when a repo you need isn't listed - and a selected repo that gets renamed or removed stays selected with a clear way to drop it, rather than quietly disappearing. You can now select up to 10 Notion pages and 10 Google documents per project. And disconnecting an integration opens a confirmation that spells out what it affects - which project loses it, what stops working, any work in progress it reaches, and, for Jira, the personal-data erasure it triggers - so nothing happens by surprise.

Your workspace role - viewer, member or admin - now consistently shapes the whole app. Viewers get clean read-only views of requirements, spec sheets, stories and tasks rather than buttons that wouldn't work; members can create and edit; and workspace settings like roles, members, invitations and billing stay with admins. Spectr no longer offers you an action the back end would turn down, and it waits for your role to load before deciding what to show, so a member never flashes a view-only notice on a fresh page load.

Already keep your PRD in Notion? You can now bring it into Spectr as a single requirement. Pick the PRD's top page and Spectr reads it and every sub-page beneath it, then writes its own version of the document - keeping every requirement, rule, decision, open question, scope item and requirement ID, one section per epic, in the order they appear. It's a one-time import, not a sync: once it lands it's a Spectr requirement like any other, so you can refine it with AI, save versions, and generate spec sheets and stories straight from it. Just connect Notion to the project and select the pages first - it's available on every plan.

When Spectr gathers context from your connected tools to write a spec sheet or requirement, it now reads them more thoroughly. Each Notion page is read in full - following its sub-pages, and pulling in a page for every feature a meeting actually discussed rather than just the closest match - and every document is summarised on its own, so detail from a long doc no longer gets crowded out by the others. Google Docs and GitHub reads were made leaner at the same time, so more of your AI allowance goes into writing the spec and stories instead of re-reading source material. And this depth is now the same on every plan, Free included - the old limit that gave lower tiers thinner context is gone. The result is spec sheets and requirements that reflect what's really in your docs.

The models behind Spectr's core generation - spec extraction, story generation and context gathering - now run on Anthropic's Sonnet 5.5, for noticeably sharper spec sheets and user stories from the same meetings, and faster too. Sonnet 5.5 also joins the AI Assistant's model picker, replacing Sonnet 5 alongside Haiku 4.5 and Opus 5.5. Any conversation you already started on Sonnet 5 keeps running on it, so nothing in flight changes underneath you.

When Spectr generates a spec sheet or stories from a meeting that referenced Figma designs, each task and story now carries links to the specific screens it's about - the ones that match what the meeting discussed, found reliably even across large, multi-page design files. Those links travel with the story when you publish it to Linear, Shortcut or Jira, and you can add, remove or relabel them from the story and task editors. So whoever picks the work up - or an AI agent working through the Figma MCP - can open the exact screen instead of hunting for it.

Every meeting import picker - whether you're building a spec sheet or generating a requirement - now has a 'Show already imported only' switch. Turn it on and Spectr lists the meetings already in that project from its own records rather than the note-taker, so you can reuse a past meeting in a couple of clicks without waiting on Fathom, Granola, Read.ai or Fireflies - and it remembers how you left it.

When you update a requirement from meetings, the picker now marks every meeting that requirement was already generated or updated from - an 'Already used' label on the list rows and in your current selection. It's a heads-up, not a block: you can still pick a used meeting again, it just means handing the AI a meeting the document already reflects. The check covers the working draft as well as every tagged version, so it stays accurate as the requirement evolves.

A requirement's version history now shows each source meeting's date - when the meeting actually happened, not when it was imported - right next to its name, and lists them newest first so the most recent meeting sits at the top. Meetings the provider gave no date for are shown last.

Importing meetings from Fathom is more reliable. The picker now looks meetings up far more efficiently, so ordinary browsing stops running into Fathom's per-minute rate limit; if you ever do hit it, Spectr tells you exactly how long to wait instead of showing a generic failure; and a failed 'Load more' now keeps the meetings you'd already loaded instead of clearing the list.

Spectr's requirement generation now runs on Anthropic's Opus 5.5, their most capable model - sharper reasoning and higher-quality requirement documents from the same meetings and prompts, with no change to how you generate or update them. Opus 5.5 also joins the AI Assistant's model picker as its most capable option, replacing Opus 4.8 alongside Sonnet 5 and Haiku 4.5. Any conversation you already started on the previous model keeps running on it, so nothing in flight changes underneath you.

When a meeting refers to someone by a nickname or a different spelling of their name, Spectr's AI can now still assign the work to the right teammate instead of falling back to plain text. There's a new 'Also Known As' card in your user settings - add the other names you go by, the same chip-and-input way you add Meeting Keywords - and Spectr will match any of them to you. Your display name doesn't change, the extra names are only used behind the scenes for matching, and an alias only links when it points to exactly one person.

Connect Fireflies.ai with an API key and import your recorded meetings straight into Spectr - the recap, the chaptered bullets, the action items and the full transcript. The full AI pipeline applies: context gathering, memory updates, spec extraction, and story generation. You can point a Fireflies meeting at a requirement to generate or update it, select several meetings to process as a single spec sheet, and mix Fireflies meetings freely with Fathom, Granola, and Read.ai in the same project. Fireflies exposes its API on every plan, including the free tier - but a key only ever sees the meetings its own owner can, so connect with a team-admin key if you want more than your own calls.

The meeting picker now marks every meeting you've already imported - across Fathom, Granola, Read.ai and Fireflies - so you won't pull the same one into a project twice. We've also sharpened a handful of tags and labels across the app for cleaner contrast and readability.

Team members synced from Linear and your other connected trackers now show up consistently and by the right name across Spectr - the Team page, assignee chips, and wherever Spectr routes work. Existing members refresh the next time that project's team syncs.

When a project is connected to Harvest, Spectr now measures how much of your logged time actually went to meetings and to miscellaneous work - rather than reserving a flat percentage on top of it. Each buffer becomes a reserve that real spend draws down, so you can see the moment meetings or overhead run past what you set aside: on the projection bar, in the Spent breakdown, and per person on the Team tab. Admins can tune what counts as meeting or miscellaneous time from the Harvest integration.

Automatic story generation quietly skipped any spec sheet task it wasn't confident enough about. That threshold is now yours to set, per project, in Project settings under AI Generation Preferences - anywhere from 40% up to 90%, or switched off entirely to generate stories for everything. Lower it to catch more, raise it to keep only the high-confidence items. Generating stories yourself always ignores the gate.

When a story is held back from publishing - most often because it's covered by an active out-of-scope item - Spectr now tells you which story it was and what the item needs, right on the publish dialog and the review list, instead of a silent failure or a bare "Failed". A blocked publish now explains itself and points to the fix.

Your Spec Sheet and Story Generator preferences now save on their own, with a Save button right beside the fields you're editing - no more reaching for the Project Details save at the top of the page. A banner flags unsaved changes so nothing slips by.

Our site has a new look and more room to dig in - dedicated pages for Integrations, Use Cases, Solutions, and Teams, plus a growing product Glossary, so it's easier to find what's relevant to how you work.

You can now react to any blog post or What's New entry with a single tap - 👍 ❤️ 🎉 🚀 - a quick, no-login way to tell us what's landing.

The sign-in screen now marks whichever method you last used - email, Google, or GitHub - with a "Last used" badge, so when you come back you can spot your usual way in at a glance instead of second-guessing which button you picked.

Not every story needs a priority. You can now clear it on any story straight from the editor, and a project's Story Generator Preferences can tell Spectr to leave priorities unset when it creates stories - so "no priority" is a deliberate choice rather than something every story has to carry.

The story editor now has a Save & Reject action. Any edits or notes you add while reviewing - like why a story doesn't qualify - are saved to the record as you reject it, in a single step. Rejected stories are kept for reference.

Connect PostHog or Sentry to a project and Spectr becomes your radar for live product signals: the errors people are hitting, plus - from PostHog - the feedback they're leaving and the feature flags you have rolling out. There's a Health drawer on every project, a new Product Health page covering the whole workspace, and a warning chip on any project that picked up new errors this week. Everything deep-links straight back to PostHog or Sentry to act on - Spectr only ever reads. Available on every plan.

When Spectr gathers context for a new spec sheet or requirement, it now folds in your product health numbers - always dated, and always labelled with the provider they came from. So a requirement can be written against "Sentry, as of Aug 7: 412 occurrences across 38 users in the last 30 days" instead of a guess.

Sentry's convention is one project per service, so watching only one would mean watching a fraction of your errors. Pick up to three Sentry projects per connection and Spectr merges them into a single view - with a floor on how many slots each project gets, so a noisy service can't crowd out a quiet one.

The workspace dashboard has a new Action Items section that gathers every action item across your projects in one place. Mark them complete, reopen, or dismiss right there - no need to open each spec sheet - and filter down to just the ones assigned to you. Sections collapse to keep your dashboard focused on what matters.

When you or a teammate completes, reopens, or dismisses an action item, the change now appears instantly everywhere it shows - the dashboard and the spec sheet it lives on - with no refresh.

Stories, Spec Sheets, Requirements, Meetings, the Scope Creep tracker and more now remember your sort order per workspace - and on the Stories page, whether you prefer the list or board view. Set it once and it's there every time you come back, on any device.

Stories now arrive in Jira with their formatting intact - headings, lists, and code blocks render properly instead of showing raw markup. And if your Jira project isn't set up for story points, Spectr now tells you up front so your estimates don't silently drop.

Spectr now connects to Jira Cloud. Pick your Jira site, project, issue type, and the status new issues land in, then publish approved stories straight to Jira - the same review-and-publish flow you already use for Linear and Shortcut. Assignees carry across, and once a story is live Spectr keeps its status and completion in sync, feeding your budget projection just like the other trackers.

Action items on a spec sheet can now be marked complete when they're done, or archived when they're no longer relevant - so the list reflects what's actually left to do instead of only ever growing.

When you import a meeting from Fathom (and other sources), Spectr now pulls in your meeting tool's own AI summary and action items alongside the transcript - so it starts with more context and captures the follow-ups your team already flagged.

Open a story and you now see who else has it open right now, so two people don't unknowingly work on the same story at once. Live updates also reach beyond the board - a story changed from a spec sheet, the Scope Creep tracker, or any other view now stays current wherever it appears, without a refresh.

A story's status now shows the same colored indicator everywhere it appears - the Stories list, linked stories, and the status filter - so its state reads the same at a glance across the app. We also corrected the spec sheet icon to match the rest of Spectr and tidied the Stories list layout.

The Stories board and list now update live for everyone viewing them. When you or a teammate creates, moves, edits, or publishes a story, the change appears immediately - no refresh, no polling. Per-lane counts and story-point totals stay accurate as changes land, no matter how many stories each column has loaded.

A story now links back to where it came from. Open a story to jump straight to its originating spec sheet task, and when a story is out of scope, jump to the linked item on the Scope Creep tracker - with a new "Confirm out of scope" action available right from the story.

Meetings, Requirements, Spec Sheets, the Scope Creep tracker, and the Stories list all get a search box and consistent pagination. Meetings now sort by date and show how many spec sheets and requirements each one produced, spec sheets show task and story counts, and the Scope Creep tracker can filter by whether an item has stories.

Stories now filter and display by their spec sheet - the one parent every story shares - instead of by meeting. Stories created from briefs and requirements, which have no meeting, now show up and filter correctly.

When you create a spec sheet from a Brief without naming it, Spectr now generates a concise, meaningful title from the brief's content instead of just using its first line.

The billing page now lets you filter activity by period, with clearer card labels and a fixed empty state.

You can now attach files directly to a message in the AI Assistant - PDFs, images, and text or markdown files. Ask about a document, a screenshot, or a spec without pasting it in, and the assistant reads it as part of the conversation.

Finding the right stories is much faster. A new quick search jumps to any story by title or description, every filter (status, type, scope, meeting, spec sheet, task, assignee) now lives in one unified menu, and you can select multiple values within a filter at once instead of one at a time.

The Stories page now remembers your last-used filters, so you pick up right where you left off instead of re-applying status, type, scope, and assignee every time you come back.

Auto-filling project team members' roles from Project Memory had quietly stopped working after an AI SDK upgrade. It's fixed - AI-fill roles reliably assigns roles across your team again.

The in-app onboarding tour now walks through more of Spectr so new workspaces get oriented faster, and requirement documents show an avatar for who authored each version - so you can see at a glance who wrote what.

A few sharp edges smoothed out: opening a requirement no longer risks overwriting your draft with a stale autosave, a spec sheet generated from a single task no longer borrows that task's title, and monthly usage now resets cleanly at the start of each billing cycle.

The workspace dashboard now opens with summary cards built for project managers: counts and recent activity across your projects, meetings, spec sheets, and stories, so you get the state of the whole workspace without opening each project one by one.

Spectr now has a built-in AI Assistant - ask it anything in plain language and it answers from everything in your workspace. Ask "What has Priya been working on?" or "Are we on track for budget?" and it looks across your projects, meetings, spec sheets, requirements, stories, AI Memory, Harvest budget, team roster and Forecast allocation, logged hours, and scope-creep items to give a complete answer - not just the first thing it finds. Scope a conversation to one project or your whole workspace, pick the model that fits (Haiku, Sonnet, or Opus), and watch it think - it shows exactly which items it checked, and which project each lookup ran against, every reference linked so you can jump straight there. Keep a chat private, share it with specific teammates, or open it to your whole workspace - then search your conversations and filter them by who they're shared with. It's read-only today: it answers and links, it doesn't change your data. Available on Starter and up.

The Stories board picks up a search box to find a story by title or description, per-column story-point totals with a completion tooltip, and per-lane "Load more" pagination so each status column loads independently instead of capping the whole board - the board stays fast on large projects.

Resolving possible duplicates is clearer: the actions now have explanatory tooltips and confirming a match reads as "Confirmed as duplicate." We also swapped native browser title hints for proper tooltips across the app and tightened spacing in a few dense views.

Sometimes the work isn't in a meeting or a formal requirement - it's just in your head. Now you can write a quick Brief in markdown and Spectr runs it through the same pipeline to produce a spec sheet and stories, pulling in context from your connected tools just like a meeting would. Start one from "Create Stories" on the Stories page or "New Spec Sheet" on the Spec Sheets page, optionally give it a name, and it lands as its own spec sheet.

The Spec Sheets and Meetings pages now have color-coded source filters (Requirement / Meeting / Brief) so you can see at a glance where each item came from. And if a brief- or requirement-sourced spec sheet fails to generate, a new Retry action re-runs it in place - no need to recreate it. The source Brief also sits at the top of the spec sheet as a collapsible card.

Spectr now scans every newly generated story against the rest of your project for possible duplicates automatically - there's no setting to switch on. When something looks like a repeat it gets a badge, and you can confirm the match to link the stories together (or unlink if it isn't one). Confirmed duplicates stay grouped across the Stories board and spec sheet views, so a reviewer only has to act on one. The story edit modals were also reworked to sit comfortably on smaller screens.

Refreshing a published story used to pull back only its completion status. Now it also syncs the assignees, labels/tags, work type, and priority set in Linear or Shortcut back into Spectr - so changes your team makes in your tracker stay reflected here. Story cards also get a clearer link out to their published issue.

You can now rename projects, meetings, and spec sheets inline, so titles stay tidy as work evolves.

Stories and spec sheet tasks now support multiple assignees, chosen from a team-member selector instead of a free-text name. Pick anyone from the project's synced roster (or type a name), assign several people to one item, and set a default on the parent task that individual stories can override. Assignee chips show on the board, list, and spec sheet views, there's a new team-member filter on the Stories page, and on publish the primary assignee maps to Linear or Shortcut with everyone else carried along. Generated stories are also grounded in each person's role, so work routes to the right owner from the start.

Connecting Google Drive has moved to Google's native file picker. Instead of granting Spectr broad read access to your whole Drive, you pick the exact files and folders you want it to use, and Spectr's access is scoped to just those - more secure, and you stay in control of what's shared. Note: because the underlying permission changed, any existing Google Drive connection has been reset - reconnect it from the project's Integrations settings and pick your files again.

Connect Forecast to bring each project member's scheduled allocation and logged hours into Spectr. It rolls up onto the project's Team and Budget views alongside the roster, so you can see who's allocated to the project and how their logged time compares - adding capacity and budget insight without leaving Spectr.

The badges that flag likely-duplicate and out-of-scope work - previously only on the Stories board - now also appear on spec sheet pages, with clearer matching so you can see exactly which existing story a possible duplicate points to. Spot and resolve repeats and scope creep wherever you review generated work.

Imported meetings now capture and show the actual date the meeting took place, and the meetings list keeps titles and dates in sync as they change. Plus a few visual fixes to the list for a cleaner, more consistent view.

Editing a team member's role now opens ready to type, and saving without changes keeps the existing template or AI assignment intact instead of overwriting it. And when you disconnect an integration, the team members, roles, and budget data it provided are pruned and the views refresh automatically - no stale people or numbers left behind.

Every project now has a Team page. Spectr builds the roster automatically from the tools you've connected - Linear, Shortcut, and Harvest - showing who's on the project, their avatar, and where each person came from, kept in sync with no manual upkeep. On paid plans you can assign each person a delivery role - pick from a built-in template library (design, engineering, product, QA, marketing, management) or write your own - and let AI auto-assign the best-fit role to everyone from your Project Memory. Roles feed story and spec assignees, so generated work routes to the right people.

Regenerating stories from a meeting or requirement used to resurface work you'd already seen - stories that were pending, approved, published, or that a reviewer had explicitly rejected. Now Spectr checks freshly generated stories against what already exists in the project and what's been turned down, and skips the repeats - so a reviewer's rejection finally sticks instead of coming back next run. Two per-project toggles (avoid duplicates, avoid rejected) are on by default and overridable per generation, and anything Spectr flags as a possible duplicate gets a badge on the Stories board so you can decide.

Pin where approved stories should go. Choose a default Linear project and workflow state, or a Shortcut team and state, per project - so publishing routes every story to the right destination without picking it each time.

A new per-project setting controls how Spectr groups generated work into parent stories and sub-tasks - from off (flat) through default, high, or always - and lets you name the parent and sub-task labels used on publish. The default preserves today's behavior, so nothing changes unless you opt in.

Spectr now tracks whether the stories you've published actually got done. It checks each published story's state in Shortcut and Linear and rolls it up - marked done only when it's done in every tool it was published to - with per-provider status chips on the board cards and the story view, plus a "Refresh story status" button on the Stories and Budget pages. That completion signal feeds a reworked budget projection: finished published work now counts toward spend, your remaining commitment splits into approved and in-progress work, and buffers and unapproved stories are broken out separately - so you can see at a glance how much budget is left for more work.

Spectr now projects a project's total budget from the work still in flight, layered on top of live Harvest spend. Approved story points convert to hours at a ratio you choose, meeting and miscellaneous buffers reserve headroom, and out-of-scope work is broken out separately. A stacked bar shows spent, approved, and buffers against the budget cap - with the projected total and how much room is left or over-committed. It appears on both the per-project budget drawer and the workspace Budget overview, and the buffer and hours-per-point controls sit right there.

Consistency and clarity across the app: unified status colors everywhere they appear - stories, spec sheets, scope creep, and budgets - plus source and AI-model logos on meetings and spec sheets so you can tell where content came from at a glance, clearer placeholder text in the requirement editor, and a fix that restores the Remember Project toggle and project dropdown sizing on the Integrations page.

Connect Harvest to a project and Spectr pulls your budget and spend right in. Each project gets a budget drawer showing hours and cost - budgeted, spent, and remaining - kept in sync read-only from Harvest, and a new workspace Budget page lays every project side by side so you can see at a glance which are on track and which are heading over. Connect once per project with a Harvest Personal Access Token (an Administrator or Manager account is needed to read budgets).

A new per-workspace "Remember project" toggle sits right in the filter bar. Turn it on and the project you're working in is remembered for that workspace and used as the default everywhere - meetings, requirements, stories, spec sheets, and the Scope Creep Tracker - as well as when you create a requirement or import a meeting. The project selector is also wider, so longer project names stay readable at a glance.

Spec sheets now show a consistent set of source chips whether they were generated from a meeting or from a requirement, so you can always tell where a spec came from.

Connected integrations now refresh their access proactively, before tokens expire, so you run into far fewer surprise disconnections mid-task - and when a connection genuinely needs attention, the Integrations page explains what went wrong in plain language instead of a raw error.

The Scope Creep Tracker picked up a round of refinements: reopen items you had already resolved, page cleanly through long lists, and a set of data-integrity fixes that keep each item's status and its linked stories in sync.

Spectr now tracks scope creep as a first-class part of every project. As it processes a meeting, requirement, or story, the AI flags anything that reads as outside the agreed scope - each with a confidence score and a link back to the exact source - and files it in a per-project Scope Creep Tracker. You can add exclusions yourself, too, for the things you already know are out.

Every item moves through a clear lifecycle: proposed, confirmed, deferred to a future phase, no longer applicable, dismissed as a false positive, or approved back into scope. Stories that are out of scope carry a badge on the board and in lists, and once you link them to an item their scope status follows it - approve the item back in and its linked stories return to scope automatically. A full history log records every change, AI or human, so the trail of what got cut and why is always there.

Connect Read.ai and import your recorded meetings straight into Spectr. Authorize once with OAuth and your recent meetings - transcripts and AI summaries included - become importable, filtered down to the ones relevant to your project. The full AI pipeline applies: context gathering, memory updates, spec extraction, and story generation. You can point a Read.ai meeting at a requirement to generate or update it, and Read.ai, Fathom, and Granola meetings all mix freely in the same project.

Requirements now have an Export menu - download the current document as Markdown or PDF straight from the requirement header, and grab any tagged version from the version history. Exports are generated right in your browser with a timestamped filename, so there's no waiting on the server and your downloads stay organized.

Can't find a file in Drive search? Paste its link instead. The Google integration now accepts any Google Doc, Sheet, Slides, or folder URL and adds it as a context source - Spectr resolves the file type automatically and confirms your connected account can access it.

A new "Generate from memory" action writes a concise project description from everything Spectr has learned about your project, and refreshes the project keywords to match - a fast way to keep your project profile current as scope evolves. Available on Starter and above, wherever AI memory exists.

When you regenerate your project's AI memory, Spectr now shows a live "Regenerating…" indicator and updates the moment the rebuild finishes - no more refreshing to check whether it's done. Reload mid-rebuild and the progress indicator picks right back up.

Uploading a long, multi-page PDF as project context runs AI text extraction as part of the upload, which could previously time out on big documents. Upload timeouts have been extended end to end, so large reference files import reliably.

Spectr can now read your Google Sheets and Slides - not just Docs - when it gathers context for spec sheets and requirements. Spreadsheets come in tab by tab as readable tables, and presentations include both slide text and speaker notes. A refreshed Google picker lets you choose across Docs, Sheets, Slides, and folders, with search across all four. Already connected to Google? Reconnect the integration once to grant access to Sheets and Slides - the new permissions aren't added automatically to existing connections.

Uploading files to create a requirement now generates the document from all of them at once - PDFs and images are read directly by the AI, so nothing is lost to plain-text copying. On the requirement page you can stage reference-file changes and commit them with "Save" or "Create new revision" (with an optional prompt), and if an initial generation fails you can simply "Retry" instead of starting over.

A new "Send feedback" option in the sidebar opens a quick form where you can flag a feature request, bug, improvement, or question. Spectr automatically notes the page you were on, so you don't have to explain where - and the team can reply to you directly.

Spectr's project memory - the knowledge it carries between generations - is now written as one coherent set of documents, so facts land in the right place without duplication across files. Your uploaded context files now inform requirements and memory too, not just spec sheets. And from project settings you can "Regenerate" memory from scratch whenever you want a clean rebuild from your latest requirements, spec sheets, and meetings.

Spec sheets now show which meetings they were generated from as source chips - provider logo and title - on both the list and the detail view, and display how long generation took in plain language (for example, "2m 20s") on spec sheets and requirement versions.

Spec sheets generated from a requirement now have a "New Spec Sheet" action that produces a fresh spec sheet from the exact requirement version the original was built from - no need to track down which version it came from. The action is disabled while a generation is still running, so you won't kick off duplicates by accident.

When you generate a spec sheet from a requirement, Spectr now treats the document as the authoritative scope and decomposes it into concrete deliverables - so requirement-sourced spec sheets come back complete and detailed rather than sparse.

Meeting imports now respect your project's context-gathering setting, and the multi-meeting import flow adds a switch to turn it on or off per import - so you can skip re-scanning your connected integrations when your project context is already current, making those imports faster.

Turn an agreed requirement into a spec sheet in one click - no meeting required. From any requirement, choose "Generate spec sheet" and Spectr runs the same AI spec-extraction it uses for meetings, working from your latest tagged version (or the current draft if nothing is tagged yet). You get a fully-formed spec sheet - named after the requirement, with its tasks and stories - and a provenance chip showing exactly which requirement version it came from. A quick confirmation before you start notes that your connected integrations aren't re-scanned for this run, since it works straight from the requirement document.

Create a BRD or PRD from pasted text, an uploaded file (PDF, TXT, or Markdown), or 1–3 meetings. The document lives in a rich WYSIWYG editor with a working draft you can edit freely. When scope is agreed, tap "Save as version" to tag it with a custom label - version history shows every tagged version with the AI's change summary as bullet points, the source type (AI prompt or meetings), and token usage. A Timeline tab records the full audit trail: every edit, AI update, version tag, and restore as a chronological feed.

Update a requirement with a plain-text AI prompt (synchronous) or point it at a meeting to run the full async pipeline - context gathering, memory update, then generation - so the AI always writes with up-to-date project knowledge. Attach up to 3 reference files (PDF, TXT, MD, or images) per requirement to give the AI additional grounding context.

A project's latest tagged requirement versions are automatically injected into every spec sheet Spectr generates, anchoring AI output to agreed scope. Meetings are now multi-purpose: the same import can feed a spec sheet, update a requirement, or both.

You can now sign up and log in with your Google or GitHub account in a single click - no separate password to create or remember. If you already have a Spectr account on the same verified email, signing in links to it automatically and picks up right where you left off; new accounts skip the email-verification step entirely.

A new Connected Accounts section in User Settings shows which social logins - Google and GitHub - are linked to your Spectr account, along with the email each was connected with, and lets you unlink any you no longer want. Changed your mind? Just sign in with that provider again to reconnect, or use forgot-password to set a regular password.

The register and password-reset forms now show a live strength meter and a requirements checklist as you type, so you know your password is accepted before you submit. Longer passwords and added symbols are recognized as stronger, making it easier to choose something secure.

Billing is now built around the work Spectr actually does for you: each plan includes a monthly allowance of spec-sheet generations, with AI-token usage metered alongside it and any overage billed transparently. A new billing activity log shows your recent plan changes and usage at a glance.

The billing page was rebalanced into a clearer layout, and "Manage subscription" and invoice history now appear only once you have a paid subscription - no more dead-end buttons. Starting a free trial pre-fills your email at checkout (still editable), and secure checkout and payments are handled by our provider, Crowdlinker.

The Fathom and Granola import lists now filter down to meetings that actually relate to the project you're working in. Spectr derives a set of search keywords from each project's description and matches them against your meetings - Granola on title, Fathom on title and summary. Keywords are generated automatically, and you can edit them anytime (Spectr won't overwrite your edits) or regenerate them on demand from project settings.

Regenerating stories now updates in real time - Spectr polls in the background so new stories appear as they're created, with no manual refresh. Deleting a spec sheet also resets its source meeting back to a clean state, so you can reprocess it from scratch without anything left over.

Each meeting now has a copy button for the full integration context Spectr pulled in before processing - handy for sharing what the AI saw or double-checking your sources. Deleting meetings is clearer and safer, and the stories views got a round of visual polish.

Meeting processing is now four switches you control per project: start processing on import, gather integration context, pause to review that context before spec generation, and auto-generate the spec sheet. Set each stage independently to land anywhere from fully automatic to fully manual with review checkpoints - and override them per import when you need to.

The project color palette grew from 14 to 20 colors, giving larger workspaces more room to keep every project visually distinct at a glance.

Enable "Review context before processing" per project and Spectr pauses after gathering context - showing you exactly what it found from your integrations (Notion pages, GitHub issues, Figma files, Linear tickets) before running the AI. Approve the context to proceed, or re-gather with additional hints. The context preview shows topic, a per-provider breakdown, and Figma screenshot count.

Each Fathom and Granola import now has a "Gather context" switch so you can override the project default for that specific import - useful when you want to skip context for a quick test run or force a fresh gather for an important meeting.

Control whether meetings are queued for AI processing immediately on import, or held for manual review. Toggle it per-project in settings, or override it on the fly from any import screen. Multi-meeting imports always process immediately.

The project and meeting dropdowns in the Stories review page now correctly narrow down results. Previously, selecting a project or meeting updated the URL but had no visible effect on the stories shown - that's fixed.

Admins and owners can now resend workspace invitations at any time - including after expiry - directly from the team members list. Each pending invite row shows how long until it expires (or that it has already expired), and a Resend button regenerates the token and sends a fresh email.

After completing a checkout or returning from the Polar subscription portal, your plan and usage now update automatically - no manual refresh needed. The billing page polls for changes in the background to pick up the webhook-driven update.

A new Viewer role gives teammates read-only access without consuming a paid seat - ideal for stakeholders who need visibility without full membership. Seat limits are now enforced during invites and role changes: workspace settings shows live seat usage, an upgrade prompt replaces the invite form when you're at the limit, and new members are auto-assigned Viewer when the workspace is full.

The onboarding wizard now includes a mandatory Create Project step between workspace setup and integrations - you can't advance without a project, so every new workspace starts with a concrete foundation. The workspace step also now supports full slug editing with immediate URL navigation on change.

New users are now auto-logged in after registration and guided through a step-by-step onboarding wizard - create your workspace, connect an integration, and take your first action. An email verification banner persists until your address is confirmed, with a one-click resend that respects a cooldown to avoid spam.

Switch your stories view to a Kanban board - one column per status, drag cards between them to update status inline. Publish and reject dialogs let you move stories through the approval flow directly from the board.

When connecting Google Drive, you can now pick exactly which folder Spectr reads for project context - so the AI sees the right docs, not everything in your Drive. MCP-based discovery makes the folder picker fast and reliable.

All dashboard routes are now scoped to your workspace slug - making deep links reliable, multi-workspace switching seamless, and the URL always reflect exactly where you are.

Spectr now has its own brand logo - replacing the placeholder across the app, email templates, favicons, and marketing pages.

Select 2–3 Fathom or Granola meetings - mixed or matched - and process them into a single spec sheet. The AI splits its context budget equally across all transcripts. Multi-meeting sheets are visually tagged throughout the app, and you can pick which linked meetings to include when regenerating.

Meetings that get stuck in processing now automatically time out and fail cleanly after 15 minutes. Import errors no longer carry over when you switch to a different meeting or note. Fathom and Granola now have distinct icons in the import dropdown.

Connect Granola and import your meeting notes directly into Spectr. The full AI pipeline applies - same context gathering, same memory updates, same story generation. Granola and Fathom meetings can be mixed in the same project.

Spectr now searches your Shortcut board for relevant stories before processing a meeting - the same way it already reads your Linear backlog. Stories come out more grounded when the AI can see what your team has already built.

The integrations page is now grouped into four sections - Meetings, Design & Development, Knowledge & Documents, and Project Management. Each section is a clean list row showing status, account, and actions inline.

Shortcut now shows your workspace name after connecting - so it's clear which account is linked. Both Shortcut and Fathom now detect expired or revoked API tokens at runtime and flag the integration immediately, before they can silently block a meeting import.

Connect your Shortcut workspace and publish stories directly from Spectr. Stories map to Shortcut types, priorities, labels, and assignees. Edits sync back - and Spectr blocks re-publishing a story that already exists in your backlog.

Sharing a filtered view now preserves your filters in the URL - so a link actually shows what you see. Copy buttons make grabbing story content frictionless. Story fields save inline without a separate action.

Managing who can do what in your workspace is now more intuitive. Role changes take effect immediately with clearer confirmation.

Spec sheets and story descriptions now use a proper rich text editor. Format headings, lists, and code blocks without thinking about Markdown syntax - it just works.

Cleaner typography, consistent components, better spacing throughout. The interface now matches the quality of the product underneath it.

Each project now has its own settings page. Sort your meetings and spec sheets by date, title, or status - your preferred order persists.

Every AI-generated spec item now shows how confident Spectr is, and whether it should be a parent task. You can see at a glance which outputs need a closer look before approving.

Workspace settings now shows invitations that are waiting on teammates - so you're never guessing whether someone got the email or needs a nudge.

Spectr now builds a persistent knowledge base per project - product overview, features, architecture, glossary, open questions, and team context. Every meeting makes it a little smarter. Available on Starter and above.

Full subscription lifecycle is now live. Upgrade, downgrade, track your monthly usage, and manage your plan - all from within Spectr. Overages on meetings and tokens are handled gracefully.

The full authentication experience is now complete. Invite teammates by email, verify your account, and reset passwords - all with proper email flows and token-based security.

Empty pages no longer just show blank space. Each one now guides you through the next step - so new workspaces have a clear path forward.

Before processing a single line of your meeting, Spectr now searches your connected Notion docs, Figma files, GitHub issues, and Google Docs. Stories come out grounded in real project context, not just what was said in the call.

Spectr now detects when a newly generated spec item is semantically similar to something already in your backlog - reducing noise and keeping your stories clean.

Configure how Spectr's AI behaves for each project - spec focus, output style, and context priorities. One team's ideal output isn't another's.

All 7 integrations now use proper OAuth with automatic token refresh. Connect once and Spectr keeps itself authorized - no manual re-linking, no expired sessions mid-processing.

Connect your Notion workspace and Figma files. Spectr reads them during story generation - pulling in documentation context and design details so every story is actually useful to the engineer picking it up.

Spec processing now shows real-time status per task - you can watch Spectr work through your transcript and reprocess individual items that need a retry.

Projects you're no longer actively running can be archived - keeping your workspace tidy without permanently deleting anything.

Not using Fathom? Upload transcripts directly as PDF, TXT, or DOCX. Spectr handles the extraction and gets to work - same pipeline, any source.