every update to Aldena, newest first

Six new benchmarks now shape the LLM leaderboard.

DeepSWE, LiveCodeBench, FrontierCode, FrontierSWE, Terminal-Bench, and MCP Atlas now contribute to coding and agentic tool-use rankings.

Adding an MCP server no longer starts with an empty URL field.

Open a room's Connectors tab, press Marketplace, and search a catalog of 90+ remote servers by name, address or keyword. Install one and it connects the same way a hand-added server does. Ask Aldena can search the catalog for you.

The LLM leaderboard is live, a free and ad-free ranking of 709 models built from public benchmarks.

One overall board spans coding, agents, math and vision. Nine categories have a page of their own: coding, agents, math, vision, design, image generation, video generation, text to speech and transcription.

Every model a category scored ranks in the one table.

• Coverage is printed on every row, as "7 of 9 benchmarks".
• Expand a row for every run behind the score.
• Search by model or vendor, and sort by any benchmark column.
• Switch a data source off and the board re-ranks in your browser from what is left.
• Switch on "available on aldena" to narrow the board to the models you can run here, then check what they cost on /models.

Add a remote MCP server to a room from the Connectors tab and your agents can use its tools.

Give it a short name and its URL. Aldena connects, lists what the server publishes, and caches it.

• Each server has its own status, with the reason underneath when something is wrong.
• Every discovered tool gets its own allow, ask or deny row in room permissions. Read-only tools default to allow, the rest default to ask.
• A tool you deny is never named to your agents at all, so it cannot be requested.
• Rename a server and your permission choices follow it. Delete one and its token is revoked.

MCP servers share the per-room connector limit on your team plan with the built-in connectors.

Every screen in Aldena has been reworked to be usable on a phone.

• The sidebar collapses into a drawer you open from the header, and it closes itself the moment you pick a room.
• Dialogs and drawers fill the screen, scroll on their own, keep keyboard focus inside them, and close on the escape key or a tap outside. Two stacked on top of each other no longer dim the screen twice.
• Tabs across settings, team pages, and agent details scroll sideways instead of wrapping into an unreadable pile.
• Long links, code blocks, and tables in a chat stay inside the message instead of pushing the whole page sideways.
• Buttons, toggles, and list rows carry a finger-sized tap area, even where the icon itself stays small.

Connect Notion from room settings and your agents can work in your workspace.

Agents can search your workspace, open a page and read it as markdown, and create pages in batches under a page or a database. They can update properties, rewrite page content, move pages, and set icons and covers. They can build a database with its property schema, rename and re-shape a data source, add views and change what they filter and sort by, and query either a data source or a saved view. They can read and post comments, look up workspace users, and query AI meeting notes where your plan includes them.

• You choose whether the connector may only read, or also write, or also comment, before you connect it.
• Notion's consent screen asks which pages to share, so what an agent can reach is what you picked there.
• Every Notion tool appears in room permissions, so reading can run freely while an edit asks you first.

Agents can also watch what Notion reports and speak up in the room when something happens: pages created, edited, moved, deleted and restored, data sources created and re-shaped, and comments added. Subscribe to a single page or to a whole data source, where a change to any row wakes the agent watching it.

The room chat has been rebuilt to be easier to follow while an agent works.

• A tool call and its result are now one line: the tool, what it ran on, and a fold that opens the full result. No more raw output in the middle of the conversation.
• Thinking blocks start collapsed, so a long chain of reasoning no longer buries the answer.
• Code blocks are syntax highlighted and come with a copy button.
• Attachments and generated images, video and audio sit in one consistent frame.
• A run that fails gets its own alert with a plain explanation of what went wrong.

Opening a chat now puts you at the newest message with no scroll jump, and fast streaming no longer knocks the view loose from the bottom.

Agents can now send mail from a connected Gmail account, either composed in one step or by sending a draft they prepared.

Sending stays off until someone turns on the "Send email" scope in the room's connector settings. After that every send asks you first in room permissions, unless you allow it. Sent mail cannot be recalled.

Connect Vercel from room settings and your agents can work in your deployment platform.

Agents can list the projects on your team, open one, and page through its deployments by state or environment. They can read a single deployment with its ready state and aliases, and pull the build logs behind it, tailing from the end or reading from the start and filtering down to just the failures. With deploy access they can ship files straight to a project, as a preview or to production.

• You choose whether the connector may only read, or also deploy, before you connect it.
• Every Vercel tool appears in room permissions, so reading can run freely while a deploy asks you first.
• A production deploy is refused unless the agent has read that project in the same conversation, so nothing replaces live traffic against a stale picture.

Agents can also watch what Vercel reports and speak up in the room when something happens: deployments created, succeeded, failed, and canceled, projects created and removed, and new domains. Watch a single project or the whole team, and narrow a subscription to production or preview so an agent wakes only for what you care about.

A Vercel team can be connected in one room at a time.

The learn center is live at /learn, a library of 54 questions about AI agents, each answered on its own page.

Every page opens by answering the question in its title outright, then goes a level deeper: how the thing works, what the real numbers are, and where the usual answer gets it wrong. Each one ends with the related questions worth reading next, so you can follow a thread instead of going back to search.

Connect Sentry from room settings and your agents can work in your error tracker.

Agents can search issues and events across your organization, open a single issue with its events and tag values, and pull up session replays, profiles, and event attachments. They can list your teams, projects, and releases, and find the DSNs a project ingests through. With write access they can resolve, ignore, and assign issues, create teams and projects, and mint new DSNs.

• You choose whether the connector may only read, or also triage issues and provision projects, before you connect it.
• Every Sentry tool appears in room permissions, so reading can run freely while anything that writes asks you first.

Agents can also watch what Sentry reports and speak up in the room when something happens: issues created, resolved, assigned, ignored, and reopened, new errors, and comments. Watch a whole project or a single issue, so an agent wakes only for what you care about.

Connect Gmail from room settings and your agents can work in your mailbox.

Agents can search your mail, read whole threads and single messages, and list your drafts and labels. Reading a message shows the names and types of anything attached to it. With write access they can draft replies with real file attachments, create labels, and label or unlabel threads and messages.

Drafts wait for you. On the scopes this release shipped, an agent prepares a reply and leaves it in your drafts for you to read and send yourself.

• You choose whether the connector may only read, or also draft and label, before you connect it.
• Every Gmail tool appears in room permissions, so reading can run freely while anything that writes asks you first.

Agents can also watch a mailbox and speak up in the room when something happens: mail arriving, mail you sent, and labels added or removed. You can watch the whole mailbox or scope it to a single label, so an agent wakes only for what lands in a particular place.

Connect Slack from room settings and your agents work in your workspace the same way they already work in GitHub, Jira, and Linear.

Agents can read channels and threads, open shared files, look up people and channel members, and check who reacted to what. With write access they can post messages, schedule one for later, add reactions, and create channels. If you grant the search scope, they can search messages as the person who installed the app.

• You choose what the connector may do before you connect it, and you can narrow or widen that later without starting over.
• Every Slack tool appears in room permissions, so reading can run freely while anything that posts asks you first.

Agents can also watch Slack and speak up in the room when something happens: a message posted, a mention, a reaction added or removed, someone joining or leaving a channel, a file shared. At the workspace level they can watch for channels created or archived and for new people joining.

Disconnecting the connector removes the app from your Slack workspace, and so does deleting the room. If another room is still connected to the same workspace, that room keeps working.

Connect Linear from room settings and your agents can work in it the same way they already work in Jira.

Agents can read issues, search them with filters, and look up teams, projects, workflow states, labels, and people. With write access they can create and update issues, move them between states, comment, and link issues to each other.

• Reading is allowed by default. Anything that writes asks you first, until you say otherwise in room permissions.
• An agent has to read an issue before it can update it, so it never overwrites a change it has not seen.

Agents can also watch Linear and speak up in the room when something happens: an issue created, updated, or commented on. Point a subscription at a single issue or at a whole team.

Enable Image generation under the new Tools tab in room settings, pick a model, and agents in that room can create images from a prompt.

• Ask for an aspect ratio, or several images in one call.
• Images render inline in the chat that made them, with a download button.
• Every image gets a public CDN link, so other agents can pick it up from there.

You are charged the provider's reported cost for each generation. Rates for every image model are on the pricing page.

Enable Video generation under the new Tools tab in room settings, pick a model, and agents in that room can create video from a prompt.

• Ask for a duration, resolution, or aspect ratio.
• Generation takes minutes, so the agent starts the job and checks back on it rather than blocking.
• Finished video plays inline in the chat, with a download button and a public CDN link.

Video is priced after the fact, because the charge depends on the delivered duration and resolution. Per-second rates, with and without audio, are on the pricing page.

Enable Speech generation under the new Tools tab in room settings, pick a model and a default voice, and agents in that room can read text aloud.

• The voice list comes from the model you choose, and agents can pick another voice from that same list per call.
• Audio plays inline in the chat, with a download button and a public CDN link.

You are charged the provider's reported cost for each generation. Rates are on the pricing page.

Enable Audio transcription under the new Tools tab in room settings, pick a model, and agents in that room can turn audio into text.

• Point it at a public link or at a file on the room's server.
• Speech an agent generated earlier is just a link, so agents can chain the two.
• Name a language when you know it, or let the model detect one.

You are charged the provider's reported cost for each transcription. Rates are on the pricing page.

New self-serve teams choose Starter or Business and begin with a card-required 14-day trial. Checkout charges nothing today, then renews monthly per active member at the price Paddle shows. Enterprise remains available through sales.

• Starter: 10 members, 20 rooms, 10 agents per room, unlimited connectors.
• Business: 50 members, 100 rooms, 20 agents per room, unlimited connectors.
• Enterprise: custom limits, priced per deal. Talk to sales.

Legacy teams with complimentary Starter keep Starter access without a trial or renewal date. Nothing they already built goes away.

During a trial, plan and seat changes apply immediately without restarting the trial. After renewal, upgrades and added seats are prorated immediately while downgrades and seat reductions wait for the next billing period. Cancellation remains scheduled for the current trial or billing-period end and can be undone before it lands.

Paddle remains the source of truth for trialing, active, payment-recovery, paused, and canceled states. Payment-recovery teams remain usable while Paddle retries. Paused or canceled teams are restricted until an owner reactivates them; already-running work may finish, and room servers keep using prepaid credits until they are deleted or credit suspension takes effect.

Owners manage all of this under Team settings in Billing. Admins can see the plan but not change it.

Credits are unchanged and still separate. They pay for model and server usage while the plan pays for the team.

Hire Orion from the agent catalog and hand it a bug. Where a code review leaves you comments, Orion carries the work to the end: it reproduces the failure, traces the root cause, makes the smallest fix that removes it, and adds a regression test.

Give it a bug report, a stack trace, or a failing test, or point it at a branch to sweep before it ships. It chases only defects that bite, never style opinions, and works anywhere in the stack.

Ask an agent to keep an eye on a pull request, a build, or a project and it will. When something happens on the other end, the event arrives in the same chat where you asked, and the agent decides whether it is worth saying anything.

• GitHub: pull requests, issues, workflow runs, and jobs. Watch a whole repo or a single PR.
• Bitbucket: pull requests, pipelines, and deployments, with an optional filter for one environment.
• Jira: issue updates and comments, on a single issue or an entire project.

Tell the agent what you actually care about in plain language. "Let me know here if CI fails or someone requests changes" is enough. That instruction is what the agent uses to judge whether an event deserves a reply or silence, so most events pass by without noise.

A new Subscriptions tab in room settings lists everything the room's agents are watching: the resource, the events, the instruction they were given, and which agent owns it. Remove any of them from there.

Subscriptions are created by agents, not from this tab. If you would rather an agent could not set them up at all, turn off Event subscriptions in that connector's scopes.

Every tool an agent can reach now runs under a policy you set: ask, allow, or deny.

• In a room, the new Permissions tab in Settings covers the tools agents use to do the work — reading and writing files, running commands, browsing the web, and their own memory — plus the tools that come with GitHub, Bitbucket, Jira, and Google Drive once you connect them.
• In your account, a new Permissions page covers what Ask Aldena may do on your behalf: teams and members, rooms, agents, skills, technologies, env vars, scheduled tasks, billing and credits, and account settings.
• Set a whole section in one click with Allow all, Ask all, or Deny all, then override individual tools from the same list.

When a tool is set to ask, the agent stops and shows you the exact call it wants to make. Answer allow once, always allow, or deny, and the run picks up from there. Always allow writes the policy back so you're not asked twice.

Asking you a question is never gated. Agents can always reach you.

Custom skills used to be stuck in the room you built them in. Now they can belong to the whole team.

• Promote any room skill from that room's Skills settings. It moves into the team catalog, and its version history and agent assignments come with it.
• A new Shared skills page for the team is where you create, edit, and delete them. Everything listed there shows up in every room's catalog.
• One team-wide active version. Switch it and every room serves the new version immediately.
• Ask Aldena can do all of this for you — create a shared skill, promote an existing one, switch the active version.

Agents schedule work for themselves — a nightly check, a follow-up in an hour. Until now there was no way to see what was on the books.

A Scheduled tasks tab now sits in every room's settings. It lists each task the room's agents scheduled, recurring or one-time, with the schedule spelled out, when it next runs, and which agent owns it. Unschedule any of them from there.

Tasks are still created by agents, not from this tab. It's visibility and an off switch, not a scheduler.

Aldena is out of 0.x. Here's what's working today.

• Rooms of agents with real roles: staff a room from a marketplace of ten agents connected by an org chart that delegates work down manager lines on a live canvas.
• Any model, per agent: every agent runs on the model you pick from a 300+ models catalog at transparent per-token pricing.
• Real tools: connect GitHub, Bitbucket, Jira, and Google Drive — agents open real pull requests, move real tickets, and work with real documents.
• You stay in control: agents pause to ask you questions mid-run, and engineering work lands as a pull request for your review — agents never merge their own work.
• Custom agents and skills: create your own agents with custom traits, accent colors, and per-agent skills, all versioned.
• Ask Aldena: a built-in concierge that sets up rooms, hires agents, and answers questions about your team, credits, and billing.
• Plus teams and invites, credit-based billing, scheduled agent runs, and the public changelog you're reading now.

You can now follow what's new in Aldena.

• The public changelog page lists every update in chronological order, newest first.
• The newest updates appear as dismissable cards at the bottom of the in-app sidebar.

Click a card to read the full entry, or dismiss it to reveal the next one.