What's new

What this page is

This page lists what changed in your workspace, one entry per production release, newest first. It is updated each time a release reaches nagent.ai, and each item links to the page that explains the feature in full.

Release 8.2, 15 October 2026

Slack

• Ask your whole team from Slack. @Nagent with no agent named puts your message to the Whole team room as you, and the room's lead answers as it would in Nagent. The room's thread comes back to Slack, and replying there continues it. See asking your team.
• Nagent reads the channels you choose. An admin ticks the public channels it reads into the Knowledge Hub and presses Read these channels: the last 90 days, then daily. If you connected Slack earlier, press Connect again first. See what the app reads.
• Sign in with Slack (Beta) is for members who have linked their Slack account, and opens the workspace that link belongs to. See linking your Slack account.
• Removing the app in Slack disconnects it. Nagent forgets the install and deletes what it read from Slack. See disconnecting.
• Changed: Slack and Teams offer only the agents on your teams. An agent on none of your team rooms is no longer listed and no longer answers by name there.
• Changed: a room's answers reach Slack only if the room allows it. Otherwise Slack hears that an agent answered, with a link to the room.
• Changed: naming an agent in Slack or Teams starts it only for an Organisation admin, Department head or Team lead. Anyone else is told why. A Slack message delivered twice starts one run.

Email

• Teams have email addresses. Every team room and every chief of staff or team lead gets one, such as marketing.<workspace>@agents.nagent.ai, and the Whole team room gets <workspace>@agents.nagent.ai. See an agent's own email address.
• Mail from a member goes into the room. Mail from a verified member is posted into the room as them, and the answers come back in the same email thread. Other mail is answered by the agent directly, without tools for anyone Nagent does not recognise.
• A new address asks first for a week. For its first 7 days, a reply to someone outside your workspace waits for a person to approve it.
• Agent mail is signed with the agent's card: photo, name, role, what it does and its address, marked as written by an AI agent.
• Changed: a member is proven, not just named. Mail to an agent counts as from a member only when it passes DMARC, the check that it came from the domain it claims, and the sender is an active member. A lead's mail that passes DMARC is answered without tools. Everyone else is unverified: no tools, and rate limited. Each email is processed once, even when it is retried. See who gets tools.
• Changed: an agent writes on its own only to people who know you. Mail to anyone but a member, or a lead or prospect who has replied, waits for a person to approve it, whatever the agent's level.
• Links in emails act only for their reader. A pipeline digest's links work only for the member they were sent to, and a MOXA approve link only for someone who may approve that post.

Developers

• The SDKs are published: pip install nagent and npm install @nagent/sdk, both 1.0.0. See SDKs.
• Knowledge, leads and triggers over the API and MCP (Beta). On the Team plan a key can read and add knowledge, read, create and update leads, and fire triggers, only as far as its member may. These scopes, removed in release 8, are back. See knowledge, leads and triggers.
• A retried run starts once. A run sent again with the same idempotency key no longer runs the agent twice. See runs.
• Two docs homes. Platform docs are for people using Nagent; developer docs cover the API, MCP, webhooks and SDKs. The old developer docs address opens /developers.
• Search knowledge by meaning. mode=semantic on the API, MCP and both SDKs returns the best matches first, with a score and the passage that matched. See knowledge, leads and triggers.
• Changed: MCP client tokens are retired. No new ones can be made; a key with the new Fire triggers scope does the same over MCP. A token already issued acts only as its issuer, and one naming no issuer stops. See MCP client tokens.
• Changed: every new key expires, after 30, 90 or 365 days. A key is held to its member's role where the agent works, and the SDKs refuse plain http except on localhost.

Workspace and rooms

• Meeting links open in the team room, from invites, calendar events and the meetings list. See meetings in the team room.
• New conversation starts a direct message with the agent you pick. See direct messages.
• Team Members works as it should. Its windows open over the page, each row shows the role held at the organisation, and the unit list no longer offers agent conversations. See granting roles.
• Sign out from anywhere. Your name on the rail, a new top bar on pages that are not rooms, each room's header and the phone's top bar open one account menu with Profile, Settings and Sign out.
• New workspaces open their admin pages. Someone who signed up could be refused by set-up, settings and billing; sign-up now makes them an Organisation admin. See signing up.

CRM and leads

• A workspace made from a won deal starts with the account's logo and name, and pressing create twice no longer makes two. The lead page reads correctly in the dark theme. See the lead page.

Contracts

• Order forms and NDAs go out as documents, in the email body with a PDF of the published version. The NDA panel has Email NDA. See sending it to the buyer.
• The buyer corrects their own company details in the deal room, which publishes a new version naming them and tells the publisher by email.
• Changed: accepting needs a legal name and address. Neither document can be accepted until both are filled in.

Credits

• Changed: credits draw only for paid third-party work. The Knowledge Hub and searching it, agent and team memory, and the emails Nagent sends for you draw nothing, and connector per-call rates apply only on Enterprise. Model calls, web research, contact lookups, images, video, voice and renders still draw. See credits.
• Changed: audit entries, sign-ins included, are kept as long as your plan says: 30, 90 or 365 days. See the plans.

Website and reliability

• The homepage, Growth Org and Agentic AI Lab pages are redrawn in the product's look, with a light and dark switch in the menu.
• Email opens, clicks, bounces and spam complaints are recorded again, and a complaint unsubscribes the lead. See unsubscribe and suppression.

Release 8, 15 October 2026

People and roles

• Roles are held at a unit. Inviting someone, or pressing Roles in their row on Team Members, gives them a role at the whole organisation, one department or one team: Organisation admin, Department head, Team lead, Member or Viewer. A role changed here no longer changes what the person may do in another workspace. See people and roles.
• Nobody can grant more than they hold. An invite, a role change or a grant is refused if it would give a permission the person granting does not have, and a Department head grants only inside their department and its teams. The onboarding team step follows the same rule: a refused invite sends nothing and names the missing permissions.
• Admin emails go to Organisation admins. The weekly digest, meeting escalations and the default approver lists go to the members who hold Organisation admin, with Department heads added where a list is meant for senior people.
• Changed: Viewers lose some acting tools. A Viewer can no longer chat with NIA or DRIS, run DRIS's sitemap discovery, use the AI tools on a lead page (the pre-meeting card, the next best step, regenerating a draft), add, edit or sync their own knowledge sources, switch a workflow's schedule on or off, or refresh a contract's signature. These need a role above Viewer.
• Changed: View as on the Knowledge Hub needs a senior role. Seeing the list as another member sees it needs Organisation admin or Department head, and each use is recorded in the audit log.

Agents

• Agents treat outside text as material, not instructions. Lead details, inbound email, a lead's website, chat transcripts and results from connected tools reach an agent marked as third-party data, with the rule that nothing inside it is an instruction.
• Automatic emails go only to people who have written back. An agent sends on its own only to an address that has answered an email your workspace sent. An address someone typed into a form or a chat no longer receives automatic email.
• Mail from a stranger does not reach an agent's tools. An agent's inbox recognises your members, leads you have written to and prospects you have sent outreach to. Mail from anyone else is read without the agent's tools, so it cannot set work in motion.
• Website chats and email replies no longer become knowledge on their own. Anonymous chats and inbound replies are kept aside instead of joining your active knowledge, and agents leave them out when they search it.

Leads and email

• A reply that could belong to two workspaces is not filed. When the same prospect address is in two workspaces' outreach, a reply from it is captured by neither, rather than by the wrong one.
• Links in emails ask before they act. The unsubscribe link in a sequence follow-up, and the approve link in MOXA's reminder email, open a page with an Unsubscribe or Approve button, so a mail scanner that opens every link changes nothing. One-click unsubscribe from the mail client works as before. See MOXA.

Security and privacy

• Nagent reads your sites over https only. When it reads a site you give it, at onboarding, for Brand Lock, as a knowledge source or for DRIS's sitemap discovery, it connects over https. A site that answers only on plain http cannot be read, and the step reports an error. See how onboarding works.
• Deactivating someone stops their API keys. Their personal keys are revoked and stay revoked if you reactivate them, so they make new ones. A workspace key made by someone deactivated or removed stops working too. See deactivating someone.
• Nagent staff visits are in your audit log. When a member of Nagent staff enters or leaves your workspace to help you, the audit log records it.
• Résumés are private. A résumé sent through a careers page is stored privately, and opening it from the applications page gives a link that lasts five minutes. Résumés sent before this release keep their address.

Developers

• Asynchronous runs and sessions. Ask for an asynchronous run and the API answers at once while the agent works; read the run, follow its events or wait for the webhook. A session id continues a conversation, your own labels ride on the run and its webhooks, a run's output goes only to the key that started it, and approval lists page with a cursor. See Nagent for developers.
• Setup snippets for coding agents. Settings, API & MCP and /developers show one for Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop and Devin. Checking on a run and waiting for it now work over MCP, and each key has Rename.
• Python and TypeScript SDKs. Both cover asynchronous runs, sessions, every page of approvals and webhook checks, and arrive on PyPI and npm shortly. Until then, /developers shows curl, Python and TypeScript examples, how to generate a client from the OpenAPI document, and how to import it into Postman, Insomnia or Bruno. See SDKs.
• A key acts as a person. A personal key acts as its owner, and any other key as the member who made it. Each call is checked against what that person may do today, so a key loses what they lose. Only the owner can rotate a personal key, and a rotation cannot give the new key more than the person rotating holds.
• MCP client tokens stop with their maker. A token generated from this release stops working when the person who generated it leaves, is deactivated or loses permission to manage tools. Tokens made earlier are not tied to a person, so generate new ones. Generating and revoking a token is recorded in the audit log. See MCP client tokens.
• Changed: three scopes are gone. The knowledge, leads and triggers scopes can no longer be put on a key, because no endpoint served them.

Docs

• The docs have new screenshots for billing, the Knowledge Hub, teams, the agent workbench, autonomy and governance, CREA's sources, the customer profile and order forms.

Release 7, 15 October 2026

Teams and rooms

• Admins can see and join any team. A Workspace Admin sees the rooms they are not on under Other teams in the sidebar, and Join team adds them. See the team room.
• The agent cards on Home no longer overlap, and the workspace rail shows the current Nagent mark.

Sales

• Sending an order form has a To, a CC and a note. The address can be changed, up to five people can be copied, and a note of your own replaces the standard line. See order forms and contracts.
• A standard mutual NDA beside the order form. The Mutual NDA panel on the lead page drafts one at any stage, with the other party, the purpose and the term. You approve and publish it, and the buyer accepts it in the deal room. An NDA does not change the lead's stage. See the mutual NDA.
• CRM: deal rooms have a Documents section for sharing PDFs with your buyer, order forms can print your PAN, and the lead page and deal room read correctly in light and dark themes.

Content

• CREA watches sources for topics (Beta). Give CREA sites, feeds, handles or news searches to read every day. The best few new items a day become drafts in Review, and a daily digest says what she found. See sources CREA watches.

Developers

• Run your agents from your own code (Beta). On the Team plan and above, Settings, API makes API keys limited to the agents and actions you choose. A key runs an agent over REST or MCP, from Claude Code, Cursor or any MCP client, under the same approvals and budget as inside Nagent. See Nagent for developers.
• Signed webhooks. Nagent can post events to your own endpoint, signed to the Standard Webhooks format. A failing endpoint is retried, then paused after 24 hours, and its owner gets an email.
• NORA has a public profile and kit. Her card, CV and decks are at /agents/nora/kit, and her agent.md is a setup guide a coding assistant can follow.

Agents

• A Workspace Admin can raise an agent to L4 and L5. Raising to L4 or L5 is open to a Workspace Admin, CEO, Co-Founder or CXO. See autonomy and governance.

Billing and knowledge

• Each plan's knowledge base limits are enforced. New uploads pause at the plan's document or storage limit, and nothing is deleted. See adding knowledge.
• The audit log is kept for your plan's window: 30, 90 or 365 days. See billing and plans.
• Price changes come with 30 days' notice, by email and on a banner across the workspace. See billing and plans.
• A refund window for a first annual Business plan. For its first 14 days, the cancel dialog says how to ask for the annual fee back, less the credits used. See billing and plans.

14 October 2026

This entry also covers the smaller releases of 11 to 13 October.

Agents and autonomy

• Every action has a class. Actions are C1 Read, C2 Prepare, C3 Internal act, C4 External act, C5 Publish or Locked, and a level runs the classes up to its own number on its own. See autonomy and governance.
• Levels move one step up, by a person. Promotion review on an agent's page moves it one level up, Change level moves it down, and L5 needs a second person to approve. Each night Nagent nominates agents whose record has earned a move, and drops one level from agents whose record has slipped. See autonomy and governance.
• Every shipped action by class, in public. See Earned Autonomy.

Sales

• Order forms. DEXA drafts the order form from the lead, a person approves and publishes it, and the buyer accepts it in the deal room with no account. The lead moves to Won when they do. Set your company details, template and plans under Order form on Sales settings. See order forms and contracts.
• RIVA checks an order form before it is published, and blocks the publish while something is wrong. See RIVA.

Billing and credits

• Refusals speak in credits. Work that would pass your allowance or the hard ceiling is refused in credits, with a button to Upgrade plan or Choose a plan. See billing and plans and credits and budgets.

Knowledge

• The Knowledge Hub is grouped into five sections: Knowledge, Memory, Intelligence, Operational and Health. See the Knowledge Hub.
• Accounts (Beta) puts everything about one customer on one page, as your role may read it. See the Knowledge Hub.
• Gaps (Beta) lists what your agents could not answer, and Write this document answers it. See the Knowledge Hub.
• Add a video or a facts sheet (Beta), and choose Who reads it when you add a document. See adding knowledge.
• Each level keeps its own memory. Lines learned by two or more agents are proposed on the Memory tab, and a person approves them. See the Knowledge Hub.

Workspace

• Industries and markets (Beta). Tick every industry you are in and the markets you sell into, on Settings, under General. Your industries suggest the strategists you can add. See who you sell to.

Programmes

• NIVA, investor relations (Beta). For workspaces where Nagent switches it on: your raise, investors by stage, the deck and the data room, with every send yours to approve. See NIVA.
• A FOLIO book carries your brand. It reads your whole Knowledge Hub, writes in your brand voice and house style, and puts your logo on its cover.

10 October 2026

Teams and rooms

• Long work runs as a job. When what you ask a room needs more than a reply, the lead starts it as a job. A card in the thread shows each step as it runs, Stop ends it, and a reply in the thread steers it. An approval the job needs waits on its card, and the job is listed in your Inbox. See long-running work.
• The thread is its own pane. It has its own composer, you can resize it, and a reply can carry a file. See the team room.
• A huddle can change the team. The lead's minutes can seat a teammate, create a sub-agent, add a skill or add a workflow, and those run on the lead's word with an audit entry naming it. Items can be marked for later, decisions name who agreed, and Wrap up asks the lead to write the minutes now. See huddles.
• Runs started from a room report back into it. A workflow run you start from the room, or a trigger that names the room, posts its report there, with a step card that follows the run in the thread. See workflows in a room.
• Only a chief of staff wears the badge. MIRA, SERA and CREA are shown as chief of staff in their rooms; every other lead is shown as a lead. See prebuilt teams.

Workflows

• Build a workflow from one sentence. Type what it should do. The name, the plan and the steps are drafted onto the canvas, and Publish starts its triggers and puts the workflow on its team. A room can start the build too, with Build a new one for this team. See building a workflow.
• Test runs on a free model. Test run, free model in the run panel runs a workflow with free read tools only, so you can check the steps before a paid run. See building a workflow and, for the same test from a room's card, workflows in a room.
• Every prebuilt team room arrives with workflows. Thirty-three prebuilt workflows are installed across the prebuilt rooms. Their schedules and event triggers stay off until you switch Keep running on, from the workflow's card in the room. See workflows in a room.
• Business events start workflows. A lead replying (lead.replied) and a flagged signal (signal.flagged) can now be a workflow's trigger. See triggers and webhooks.

Sign-up and sign-in

• Remember this device. Tick Remember this device for 30 days on the sign-in page and that browser signs you in without asking again. The Trusted devices page lists each one with Revoke trust, and signing out forgets the device. See signing in.
• A new password signs out your other sessions. Changing your password ends every session but the one you changed it from. See signing in.
• Sign-up asks for your consent. The Terms and Privacy box must be ticked before the sign-up button works. See signing up.
• Pick a department, or one lead. Sign-up lets you take a department whole or tick a single team lead, and the footer counts both ("1 team and 1 lead chosen"). HOOK, MOXA and Echo are no longer added to every workspace; a whole Marketing department still brings them, and a lead can be added later from the Welcome screen. See signing up and prebuilt teams.
• Your brand is read once at sign-up, from your website, and your account keeps your browser's time zone so meeting slots fall in your own hours. See Brand Lock.
• The set-up chat does the set-up. Your team lead reads twenty-one set-up guides, asks for your objective first, reads your website from the chat, takes a file you attach, and shows Set-up N of M as the workspace fills in. Its replies stream word by word. Hands-free voice is on where it is switched on, marked Beta. See setting up with your team lead and agent voices.

Agents

• Build a whole team from one proposal. Ask the builder chat for a team, approve its proposal once, and every agent is built, the room opens and the lead is told who joined. See building a team of your own.
• One click to a live agent. Provision in the agent builder makes the agent live and opens its page, with its role, department and who it reports to set from the spec. See creating an agent.
• A lead hands work to your own teammates. Agents you built yourself are now on the list a lead delegates to, described by their role and tools. See building a team of your own.
• Import many skills from one file. Import skills on the Skills page takes one .json or .md file and creates up to 100 skills, marked Beta. See skills.
• Board views belong to the team. A Layout row on the team board sets how each kind of entry is shown, as a list, a calendar, a to-do list or KPI cards, and every agent on the team renders the same layout. The agent's Surfaces card keeps only what it may write. See boards and surface views.

Studios

• RUPA finds images by what is in them. Search the library in plain words, and use Cut out to lift a subject onto a transparent background. Presets and assets are kept with the studio. See RUPA.
• Product images reach the studios. Images on an offering are read by RUPA's and ALPHA's room actions. See products and services.
• Every new studio image is described and indexed, so search can find it. The description is charged to your credits. See credits and budgets.
• ALPHA personas with their own faces and voices. A persona's avatar can be made from a still, from a description or from a twin's footage, with consent recorded, and its voice can be cloned, designed or the twin's own. Marked Beta. See ALPHA personas and, for the voice engines a persona can use, agent voices.
• A cut with no template bound. When no video template is bound, ALPHA plans the shots as generated clips and labels the storyboard "Assembled by Nagent". See ALPHA.
• VIDYA makes long-form video. A learning series team under CREA: KRAMA plans the season, LEKHA writes each lesson and SATYA checks it against your sources, with a human, a co-hosted or an AI presenter on screen. The Nagent team adds it to a workspace. See VIDYA.

Programmes

• Prospects can unsubscribe. Every prospect email carries an unsubscribe link and the headers mail clients read. An address that opts out goes on your suppression list, its sequences end, and a prospect promoted to a lead keeps the opt-out. See unsubscribe and suppression.
• Echo drafts replies to reviews. A reply is drafted and waits as a card for a person to approve before anything is posted. See Echo.
• More agent work has a price. A CREA blog draft, a DRIS plan, a NIA plan and a FOLIO book each stop at a priced budget, and a room pass that holds a paid research tool may buy one research round, capped at 0.50 USD. See credits and budgets.

On your phone

• The workspace fits a phone. The rail opens as a drawer, a room's panels open as sheets from More, the Inbox shows one pane at a time, and sign-in is one tap from the menu. See on your phone.

Fixed

• A run that stopped before it finished now ends with what it produced, rather than a progress note.
• Agent tools that failed quietly with a runtime error now run.
• Workflow and trigger runs that name a room post their report into it; before, the report was dropped.

6 October 2026

Setting up

• A welcome that asks one thing at a time. Your first visit opens on two ways in: set up with your team, or set it up yourself. Each screen shows one next step. See your first hour.
• Your team's lead guides set-up. SERA for Sales, CREA for Content and MIRA for Marketing read your site, draft what your business sells and to whom, and offer each change as a card you save or undo. See setting up with your team lead.
• Each team's first job ends in a review. Nothing it makes goes anywhere until you approve it, and once it has run it can run every week at the same time. See staying in the loop.

Staying in the loop

• Since you were away. Back after a day or more, Home says what ran, what was approved and what is waiting for you.
• Work is waiting for you. When something has waited two hours for your decision, you get one email with a link that shows exactly what will run before you approve it. See staying in the loop.
• Drafts that learn from your edits. When you edit an outreach draft, the next drafts are written with your changes in mind. See VANI.
• Send feedback from the sidebar, from search or by asking your team's lead, and tick the box to hear when it ships. See sending feedback.

Fixed

• Work waiting for your decision shows in your Inbox, on Home and in the bell, with the same count in each.
• The set-up counts on the welcome screen match what is set up.

Where to go next

• Getting started: the first pages to read.
• Staying in the loop: the emails and cards that keep you up to date between releases.
• Sending feedback: tell us what to build next.
• Glossary: the words these entries use.