What’s new in AgentHub.

Connect Claude or Claude Code to AgentHub, pick the clients it may read, and ask about open requests, your own promises and published updates without copy and paste.

• In AgentHub, open Settings, then AI apps. It gives you the address to add as a custom connector in Claude, and the commands to run in Claude Code.
• You pick the clients each app may read. Change them or disconnect in AgentHub at any time.
• Claude can read open requests, published client updates and your own saved promises for those clients. AgentHub’s connector has no tools to approve, check, publish or send.
• Some text stays in AgentHub, such as internal notes, contact fields, private chats and drafts. Words repeated in a request, update or saved promise can still go out.
• Request text from before about October 31, 2026, and text from Slack, Google or the marketplace, stay in AgentHub too. Claude still sees that those records exist.
• A limited filter removes some links and masks some emails and phone numbers. Names and other personal details still go out, and it can miss things.
• Owners and admins can connect. Operators can if the workspace owner allows it. The terms and settings of your Claude account or organization apply to what Claude reads.
• Disconnect in AgentHub ends one connection. Claude Code can also read through your Claude connection, so disconnect both to stop both. Neither recalls what Claude already read.

A weekly or monthly client update can now start from what you wrote down during that period. Turn it on for a schedule, and on the due date AgentHub suggests items from your notes for you to review.

• Under an update you made in a client’s conversation, AgentHub offers “Prepare a draft like this on a schedule?”, with who it’s for and the dates each draft covers. Choose Every week or Every month to set it up.
• On the schedule, tick “Add items from my notes each week” (or each month). It’s off until you turn it on, and it reads only your own notes.
• On the due date, the items wait in a private conversation, and Briefs tells you they’re ready to choose.
• Review the suggested items, untick anything you don’t want, then prepare the draft. Your client sees the items as your report.
• An item that may change who decided something, or that says “last week” without the date, waits under Needs your check.
• It uses AI actions, usually one each time. The draft stays private until you publish it, and nothing is sent unless you send it.

In a client’s conversation, the update now opens beside the chat, so you can check the draft against what you wrote without losing your place.

• Choose Review beside this chat on the update card.
• The draft opens on the letter as your client will read it. Ready to publish? shows who it’s for, the dates and what’s left to check.
• Looks right saves the letter as shown. Edit the update changes it.
• Open as a full page when you want more room.
• On a phone, the update fills the screen, with a way back to the chat.

AgentHub now keeps track of who asked for what, who owns it and which work it belongs to.

• Improved how memory preserves who asked, decided or owns the work.
• Better handling of people with similar roles and details belonging to different projects.
• When conflicting notes affect your request, AgentHub can ask which information applies.
• For requests such as “the Friday MFA slide,” AgentHub asks which date you mean when matching notes put that Friday in the past.

When a note says “Rina will send the admin access by Friday”, AgentHub remembers it. Home lists those promises with who owes each one and when, so Friday’s logo files don’t quietly turn into next month’s problem.

• Each promise shows the client, when it’s due, who owes it and the note it came from. Only you see your promises.
• Anything unclear, like a date that moved with no new one, says Needs clarification instead of guessing.
• Mark a promise Done, edit it, or say it’s not a follow-up. Undo takes it back.
• For a promise someone else owes that’s due or overdue, choose Draft follow-up and AgentHub writes a short, friendly one. You send it yourself.

Ask for an update in a client conversation in the app. AgentHub suggests items from your notes and earlier work with that client. Choose what goes in, then review the draft before publishing.

• Prepare an update even when the client has no workflows.
• Inspect the sources behind suggested items. Your team can read the items you include in the draft; your source conversations stay private.
• Progress and open items appear as your report. Work you say is finished is labeled Reported done, separate from verified work.
• Items that need your judgment appear under Needs your check. If a source changes before publication, affected items need review again.
• Across your clients, Briefs and Slack continue to prepare updates from recorded work only. Nothing is published or sent automatically.

Some Word drafts that end on stray code-like text now get a warning. AgentHub keeps the file and asks you to check the ending.

• The reply and the file’s card say “Check the ending. This draft may be incomplete.”
• You can still preview, edit, download and share it.
• It doesn’t catch every unfinished ending, and it doesn’t check Excel files. Read the document before you send it.

We improved how AgentHub drafts a recommendation you ask for in slide content or a Word document, using the figures in your notes and a reason for its choice. Answers also show which of your notes they used.

• Ask for a recommendation in your draft. AgentHub is guided to give one with reasons, keep your figures and identify its own suggestions. When the facts aren’t enough, it should say what’s missing. You still make the decision.
• From your earlier work lists your own notes first, with their opening words, so you can see what an answer used.
• Answers now receive today’s date and the Monday-to-Sunday week in your workspace’s time zone.
• A client update draft says it’s made from what’s recorded in AgentHub. Your notes aren’t added automatically, and an empty record no longer reads as an all-clear.
• Details now shows saved contacts, not just the main contact. Add someone from the update card, and a title from your notes, like “Rina Shah, IT manager”, is filled in for you.
• Review every draft before you share it.

Turn on setup reading for an n8n account, and answers about a client can explain how its assigned workflows are set up.

• Off until an owner, admin or operator turns it on for an n8n account.
• AgentHub keeps summaries of how each assigned workflow is set up: what starts it, its checks, the fields it writes and its steps. Earlier summaries stay as history.
• Code is never read or kept. Values in fields named like secret or personal data are withheld, though a sensitive value in a plainly named field can still be kept.
• Turning it off or disconnecting deletes the summaries.

In a client conversation, AgentHub can use what you told it in your other conversations with that client.

• It keeps a short list of facts from your own messages: people, numbers, dates, decisions and open items.
• Answers say where each one came from, such as “from your notes, October 14.”
• It uses only your own conversations with that client, never a teammate’s or another client’s.
• See what an answer used under From your earlier work, and stop any item from being used again.

In the app, say this week, last month, Q3 or two dates, and AgentHub works out the dates for the update in your workspace’s time zone.

• A plain request with a period makes the draft right away.
• With no dates, or anything more in your message, it shows the dates it would use so you can confirm or change them.
• If the person you name isn’t a saved contact, you can add them from the answer.

Long answers are no longer discarded by the old length limit. If an answer runs out of room, AgentHub asks you to request one part at a time.

• Bold text, bullets and numbered steps show as formatting, not asterisks.
• Copy keeps that formatting for an email or document, or copies plain text.

Under each recommendation in an answer, Make this a request opens the request form with that recommendation filled in.

• Owners, admins and operators see it while the workspace can be edited.
• The form shows it as suggested by the assistant, with its assumptions.
• Nothing is saved until you save the form.
• Creating a request does not approve its scope or start any work.

Ask for a client update in a client’s conversation or in Across your clients, and AgentHub makes the private draft for you to review.

• It is the same draft you would get from Briefs, New brief.
• Nothing is published or sent. You review the draft and choose what the client sees.
• Ask for a Word document instead if you want a file.

In a client conversation in the app, when your message mentions the client’s website, industry, the work you do for them or their main contact, the answer can offer to save those details.

• Save in one click, or review each detail first.
• You choose what is saved.
• Saved details are shared with your team and used in later answers. Your notes stay private.

In a client conversation, in the app or in Slack, AgentHub reads the details you saved: website, industry, the work you support and the main contact.

• What came from those details shows About this client.
• Internal notes and the contact’s email are never read.
• AgentHub does not visit the client’s website.

The assistant can make an editable Word or Excel file in a conversation, and you revise it by typing.

• Get an editable file in a conversation and revise it by typing.
• Every version is kept; restore an earlier one at any time.
• Share one version by private link, and revoke it when you want.
• Review the file before you send it. AgentHub does not check that figures or wording are right for your client.

Ask on Home about one client, across your clients, or in general, and keep the conversation.

• Ask on Home about one client, across your clients, or in general.
• Answers about your clients link to the records they used.
• Choose which parts of a conversation to share into a job; nothing else is shared.
• Conversations are private to you.

Invite people by email to work in your workspace.

• Invite people by email as a viewer, operator or admin.
• Plans are priced by clients, not people.

Client briefs now read like a letter from you, the library shows each item as itself, and a new workspace begins by pasting one client request.

• A client brief renders as a letter: letterhead in your name and colors, a greeting to the person it is for, the summary in your voice, your sign-off, and the changes enclosed with who approved each one and how it was checked.
• The catalog and your library draw each item as itself: a workflow as its own map, an AI skill as its page. The catalog opens with license-checked workflows and AI skills from public repositories, each with its author and license.
• A new workspace starts with one pasted client request. The confirm step fills the title, who asked and when from the message, and names anything still missing in one sentence.
• On the record, the status color follows the state, the approval wait panel notices the client's decision on its own, the client approval page says who is asking, and the verification checks go into the brief.
• The homepage opens with what AgentHub is, and shows one record filling as four requests arrive.

Each workspace has its own address for forwarded client emails, and Drafts to review is one list grouped by client.

• Settings and Drafts to review show your workspace's forwarding address. Forward a client email from the address you sign in with and it lands as a draft with the sender, the date and the ask filled in.
• Drafts to review groups every draft by client, whether it came from a forwarded email, a client link or Slack. Each one becomes a request, joins an open request as a client message, or is dismissed. Nothing is on the record until you decide.
• When you file a draft you can keep the sender's address as a client contact, so the next email from them lands under that client.
• The homepage is new. It opens with the moment the product is for: a client asking what changed, who approved it, and how it was checked.

New accounts can try the workspace for 30 days and prepare workflows and AI skills independently of a paid workspace plan.

• The workspace trial covers up to three clients. No card is required and there is no automatic charge. Existing permanent Free accounts keep their one-client access.
• When the trial ends, client records remain readable and exportable. Consultant remains US$49/month for three clients; Agency remains US$99/month for ten.
• Save workflows and AI skills with supporting JSON, Markdown, text and CSV resources in your private library. Review and export your package without starting a workspace trial.
• The request workbench, library and navigation have been updated. Marketplace publication, purchasing and payouts are not open yet.

The updated workspace brings clearer navigation, a dedicated My skills area and improved subscription handling.

• The refreshed workspace and sign-in experience are available in the app.
• An implementation note is required before a change moves to verification.
• Billing shows USD plan prices and identifies UTC dates. Subscription updates handle renewals and cancellations more reliably.
• Marketplace listings and purchasing are not open yet.

The legal pages identify AgentHub AI LLC and describe the website and account services used alongside the application.

• Privacy now covers website analytics, contact-form processing, authentication protection, and browser storage.
• Terms clarify paid checkout, monthly renewal, and what happens when a payment fails.
• Both pages include the business mailing address and a contact for questions.

New brief drafts include recorded approval names and dates so the client can trace the decision behind the work.

• Briefs distinguish a client response through an approval link from outside approval recorded by a team member.
• Missing approval stays visible. Private approval email addresses and response notes are excluded.
• Existing frozen briefs retain their original content. Publication and sending remain separate decisions.

The app has a refreshed design with improvements for smaller screens and clearer guidance through a client change.

• Navigation, forms, sign-in, and Help use the updated AgentHub design.
• The first-change form preserves entered details when a save fails, so you can correct the issue and retry.
• Saved-request messages and the next step toward a client brief are clearer.
• Client briefs and workspace records are easier to read on smaller screens.

AgentHub records when a secure approval page is first opened and requires a human scope decision before a passed change can close.

• The change record shows when the current approval page was first opened successfully. This records a page open, not who opened it.
• After a passed verification, the operator records one scope outcome: the result matches the recorded scope, revised approval was recorded, or the change closed with an unresolved scope difference.
• Revised approvals require a reference. Unresolved differences require a note. The outcome and attribution remain in the audit record and client brief.
• AgentHub does not send approval links, identify viewers, infer scope matches, perform implementation, or verify work automatically.

AgentHub now starts on a permanent Free plan for one managed client, with capacity plans for consultants and agencies managing more client relationships.

• Historical release: this one-client Free plan remains available to existing accounts. Since September 14, new accounts use a workspace trial; the package library remains free.
• Consultant is US$49 per month for up to three managed clients. Agency is US$99 per month for up to ten. Larger workspaces contact AgentHub.
• When a lower limit takes effect, the owner chooses which clients remain managed. Other client records become historical and read-only; nothing is deleted and published briefs remain available.
• Paid checkout never grants access by itself. AgentHub waits for signed payment confirmation before activating a paid plan.

Each client can have one reusable secure request link, and every submission arrives as a draft for consultant review.

• A named client contact can submit a request through the standing link without signing in.
• The request enters Request inbox as a draft until a person chooses the workflow and confirms or dismisses it.
• AgentHub creates but does not send the link, and submission never approves scope or starts implementation.

Start with the request, reuse the client and workflow when they exist, and create only the records that are missing.

• Create → Client change is now the single path for new and existing client work.
• Start with the client request, then select an existing client and workflow or create only the missing context inline.
• A blank workspace can still reach a formal Requested change from five required decisions.
• Existing workflow owners are preserved. The signed-in operator fills only missing technical or support ownership.
• One transaction creates the minimum accountable record or creates nothing.
• No connector is required, and nothing is sent, approved, implemented, verified, or published automatically.

New workspaces can create the minimum accountable record from five inputs without connecting an external tool.

• Enter the client name, workflow name, runtime platform, client request, and client decision owner.
• AgentHub creates the active client and workflow, records the decision owner as business owner, assigns the signed-in operator as technical and support owner, and files the formal request.
• The new record opens in Requested state so approval, implementation, verification, and brief publication remain explicit decisions.
• No connection is required, nothing is sent to the client, and AgentHub does not run or edit the workflow.
• This new-workspace path was superseded August 7 by the unified Client change path for new and existing client work.

Users can now follow current setup guides or ask AgentHub how to complete a product task without first creating workspace records.

• Seven focused guides cover the first client change, client and workflow setup, connected tools, request intake, approval and verification, client briefs, and product boundaries.
• Ask AgentHub answers setup questions from the current Help Center and links to the full guide.
• Operational questions continue to use stored workspace records and show the evidence behind the answer.
• Empty workspaces receive a clear first-change path instead of a missing-record dead end.
• Unsupported connections and unrelated questions are declined rather than guessed.

Work built with Claude Code, Codex, Cursor, or Copilot can keep GitHub and deployment references attached to the client workflow and its change record.

• Choose the platform where the automation actually runs. Work built with an AI coding tool can still run in n8n, Make, or Zapier.
• If the code lives in GitHub, add its repository and deployment links without connecting the GitHub account.
• Attach a pull request, commit, test or CI result, deployment, or other reviewed reference while a change is in progress.
• Implementation evidence stays private by default. Only references explicitly marked as client-safe enter new client briefs.
• Client workflows are now separate from optional Connected tools, and manual setup starts with only the client, workflow name, and runtime platform.
• AgentHub does not ingest source code or AI conversations, run tests, merge changes, deploy code, or treat a technical artifact as passed verification.

AgentHub can prepare a private client brief draft on an agency-controlled schedule, while review, publication, and email remain explicit consultant decisions.

• Set one weekly or monthly schedule for a client and name the contact who should receive the finished brief.
• Scheduled runs prepare private drafts from the client's recorded work. Nothing is published or emailed automatically.
• The consultant can edit the draft, publish a frozen brief, and then explicitly email its secure link.
• AgentHub records when the email provider accepts the message. It does not claim delivery, opens, or clicks.

When a change enters verification, AgentHub turns the recorded request and available safe structural difference into a short editable checklist.

• Suggested checks focus on changed workflow status, systems, steps, routes, order, and schedule where recorded evidence supports them.
• Partial or missing structural evidence produces an explicit source-platform review instead of invented detail.
• Consultants can edit and save the plan to the immutable change history before recording a human pass, failure, or insufficient-evidence result.
• The plan does not run a workflow, mark checks complete, or verify the change automatically.

n8n and Make connections now match the way consultants use shared automation accounts across internal and client work.

• Connect an n8n account or Make team once at the AgentHub workspace level.
• Assign only selected workflows or scenarios to one client at a time. Unassigned resources stay outside AgentHub.
• Later refreshes update assigned resources and surface new choices without importing them automatically.
• The n8n setup flow now explains that Cloud trial accounts do not provide API keys and points paid Cloud and self-hosted users to the supported paths.

After a baseline sync, AgentHub can compare supported workflow versions and show safe structural differences on the change record.

• The first supported n8n, Make, or Zapier sync for a workflow records a forward-only structural baseline.
• Later supported changes can show differences in steps, systems, status, schedule, order, and routing where provider data allows.
• AgentHub does not reconstruct workflow history from before the baseline was recorded.
• Credentials, parameters, field values, and execution payloads are excluded. Node and step names remain private workspace data.

AgentHub introduced 30 days of full access with no credit card. This offer was replaced August 13 by the permanent Free plan.

• New workspaces could start with one live client workflow without requiring an onboarding call.
• There was no automatic charge at the end of the access period.
• After 30 days, workspaces could continue for US$99 per month or remain read-only.
• The permanent Free, Consultant, and Agency plans superseded this offer on August 13.
• Authentication and billing screens now use the current AgentHub brand.

Selected Slack channels can feed likely client requests into a review inbox, and operators can ask questions against records stored in their workspace.

• Only explicitly selected Slack channels are monitored.
• Likely requests arrive as drafts and require human review before filing.
• Ask AgentHub cites stored records and leaves missing information visible.
• AgentHub does not approve, run, or edit a live workflow.

The main workspace now opens on the work that needs attention instead of a generic dashboard.

• A redesigned Operations Desk for active requests, approvals, incidents, and client commitments.
• Light mode is now the default application appearance.
• Navigation and record states were tightened around post-launch client operations.

Agencies can bring live workflow records into AgentHub even when a direct connection is not available.

• CSV workflow inventory import with validation before records are created.
• Clearer connection states when a provider token is revoked or stale.
• A refreshed client operations workspace for ownership, health, and open work.

The build gained additional ways to connect workflow and business context without handing outside tools database access.

• Read-only inventory connections for Make and Zapier.
• HubSpot connection flow for client context.
• Workspace API keys plus narrow endpoints for clients, workflows, and change requests.
• Zapier Manager alerts can open an AgentHub incident through a signed webhook.

A client change can now move through the accountable lifecycle and become a client-facing brief.

• Request, ownership, approval, implementation, verification, and closure stay on one record.
• Guided client brief creation from live workspace records.
• Read-only n8n workflow inventory connection.
• Consultant and client identity on branded briefs.