Every routine and trigger on a team's page now has an Edit button. Change its days, time, instructions or name in place. Its history stays, and its next run moves to the new schedule.
Teams search the web, with nothing extra to connect
Teams on Claude, OpenAI or Grok now search the web with that provider's own search, on your own key: a few cents a search, counted in the team's budget. There's no search service to sign up for.
A search comes back as a short write-up with its sources, and the team can open any of those pages to read them in full.
DeepSeek has no web search of its own, so teams that run only on DeepSeek can still read pages a job names but can't search.
Fewer reviews, and long work that finishes
A fully autonomous team publishes documents in its own library without waiting for you. Changes to your organisation's library still come to you, and so does anything from a job that flagged a suspicious email or page.
A team can correct its own draft before you've reviewed it, so you see one draft, not two.
On Claude, a team can write a long document or think through a hard step in one go instead of being cut off part-way.
Help desks on call, and access levels across every tool
A team set to ask before any change asks before every change in every tool it uses, including the tools it shares with the rest of your organisation.
At the default level, anything that sends, deletes or shares asks you first in every connected tool, including servers you added yourself. With full access it runs, after the same check against the job.
Help desks get a one-click on-call trigger: Watch new tickets, for Zendesk, Help Scout, Gorgias and Plain.
The Project Management and Product Analytics teams now suggest Linear or Jira, and PostHog, Amplitude or Mixpanel, connected in one click.
The daily summary goes out to every organisation on time, however many there are.
Hearing about it sooner, and links that just work
When a team flags that an email or page tried to instruct it, or that a job asked for something risky, the team's lead and approvers hear about it straight away by email and chat. The daily summary counts anything still unseen.
A link to a team or a job in your other organisation now offers to open it there, instead of saying it couldn't be found.
A page for something that isn't there says so at once, instead of loading for half a minute.
A decision you make on an approval reaches the team within minutes, even if something on the way drops it.
Tools a team builds for itself follow the same safety rules as the team when they run inside a job.
A team that proposes a new way of working no longer stops to wait for you. It carries on, and the proposal waits in Needs you for up to two weeks; your yes puts it live.
Teams that read the web — and can't be talked into leaking
Teams can read the web: a worker opens public pages and PDFs on its own — sites the job names, results from a search connection, and the pages those link to. Exa and Tavily searches no longer ask for approval; they only read.
A second pair of eyes on fully autonomous teams: before a team that never asks sends, deletes or shares something, it's checked against what the job is actually for. Anything that doesn't fit — mail to an address the job never mentions, a file shared with a stranger — waits for you, with the reason.
When a team flags that an email or page tried to instruct it, that job asks you before every change for the rest of its run, and grants itself nothing.
GitHub now asks before anything that isn't a read — pushes, merges, comments and new issues — unless you give the connection full access.
Agent-to-agent, safe retries, and teams that speak up
Jugg speaks A2A, the agent-to-agent protocol: an agent built on any A2A framework finds Jugg at jugg.ai/.well-known/agent-card.json and hands a team a task. Questions from the team come back as "input required", the report comes back as the result — and with push notifications, Jugg tells the agent when the task is done or needs an answer instead of being polled.
Safe to retry: an agent that resends a request after a timeout gets the first answer back, never a second job. Every error carries a stable code, and rate limits say how long to wait.
Teams speak up without stopping: when an email or page tries to give a team instructions, when a job asks for something risky, or when the numbers look wrong, the team flags it under Needs you and carries on with the rest. Agents using the API see the same flags on the job.
Jugg for agents — and a new home page
Connect from Claude.ai, ChatGPT and other apps by signing in: add jugg.ai/api/mcp as a custom connector, sign in to Jugg, choose what the app may do. It shows under Settings → API keys, where you can disconnect it.
Agents can hand over standing work, not just one-off jobs: schedules ("weekdays at 9"), watches that act only when there's something new, and private webhook URLs — each can start in shadow mode and be taken live later.
Tell me when it's done: give a job a notify URL and Jugg sends a signed message when it finishes or needs an answer.
Your organisation now has a timezone, taken from your browser the first time you sign in and editable in Settings. Teams use it for "today", and new schedules default to it.
New pages for every integration and every ready-made team, so you can see what a team would do in your tools before you sign up.
Teams ask for the tools they need: when a job needs a service that isn't connected, the team asks, and the request waits under Needs you with a Connect button. Agents using the API get a link to hand to a person for the same thing.
Your AI agent can now use Jugg directly: add Jugg as an MCP server (Claude Code, Cursor and any MCP client) or call the REST API, and it can hire teams, hand them jobs, answer their questions and collect the results.
Agents can sign themselves up: one call gives an organisation, an owner key and a claim link for the person they work for. When that person signs in through the link they own it, and it gets the free starter turns.
API keys in Settings: make a key for an agent or your own code, choose its access (member, admin or owner), see when it was last used, and revoke it any time. A key never has more access than the person who made it.
Agent docs at /docs/agents, an OpenAPI description of every call, and llms.txt so assistants can find their way around.
A new home page: what a standing team does, how it learns the job and shadows your people before it goes live, and the teams you can hire today. Plus About and Contact pages.
Teams that learn the job before they do it
Learn the job: one click on a team and it studies how the work is done at your company — your documents, the help articles in your tools, and your people's past replies, tickets and records. It comes back with a team briefing, a playbook drawn from what your people actually do, the questions your documents don't answer, and a readiness score from real past cases. It reads only; nothing is sent or changed.
Supporting a new product? Import its help centre or manuals into Docs, then let the team learn it. Everything it writes waits for you to publish or approve.
Pinned documents: pin a short briefing in Docs and every job that can read it starts with it — what the product is, the plans, the words you use, the answers to the common questions. Up to three per library.
Shadow mode: put a trigger in shadow and its jobs draft exactly what they would do — the full reply, the update, the decision — while your people keep doing the work. Each day the team compares its drafts with what your people actually did, grades them, and proposes fixes where it got things wrong. The trigger shows how often the drafts matched; switch it live when you're happy.
Grade a draft yourself: on any shadow job, say whether you'd have done the same, with a note. The team learns from your note, and your grade always stands.
Draft only, any time: tick Draft only when you give a team a job, and it writes down what it would do without sending or changing anything.
Teams that work with your people — Slack, questions, schedules and more
Slack, both ways: connect your own Slack app (we give you the manifest), then @mention it with a job. Its questions and its result come back in that thread, and mentioning it in the thread answers the question or steers the job.
Agents can ask you: when a team needs a decision only you can make, it asks — by email, chat or Slack — and waits for your answer. Approvals now take a note the agent reads, an edit of the wording before it goes out, and "don't ask again" for an action you trust. A reminder arrives before anything expires.
Steer a running job: send guidance from the job's page and the director changes course on its next step, without starting over.
Triggers on a clock: "weekdays at 9am, London time", and routines that simply run every time — a weekly report every Monday. Plus a private webhook URL and a team email address: anything that can call a URL, and anyone who can send an email, can give a team work.
People on teams: name a team's lead, approvers and members. Approvals go to them, and whoever gave a job hears when it finishes. Choose what you're emailed about in Settings. Mark a team private so only its people see its jobs, memory and documents.
Company basics: change someone's role, hand the organisation to someone else, leave one, and send invites by email. Settings now shows the full activity log, with a CSV download, and the sign-in methods on your account.
Monthly budgets: a team's budget can refill every month, and a team on autopilot paces its own work across the month.
New connections: HubSpot, Pipedrive, Asana, Zendesk and Shopify, with a private token from each.
Start free: every new organisation gets 200 free turns for its first team. Failed jobs have a Run it again button, and the team designer now suggests an on-call routine and a search connection for research work.
Your teams in your pocket, a daily retro, and a new look
Telegram: connect your own bot in Settings and link your chat with a one-time code. Type a job in plain words and the team starts on it; /teams, /team and /status from anywhere. Owners and admins hear there when a team needs a decision, finishes a job or stops. Approvals stay in the app, behind your sign-in.
New Daily retro trigger: on days a team finished or failed jobs, it reviews them, files the lessons to the Hive, and proposes playbook changes for you to approve. Quiet days cost nothing.
A new look across the site and the app: paper and ink, with the Jugg.ai wordmark. Dark mode is now warm ink, and the live team views draw on paper in the light theme.
Nearly a hundred connections, knowledge import, and a support desk that learns first
Connections: close to a hundred services now connect with one click and a sign-in — Stripe, PayPal, Zapier, Supabase, Sentry, PostHog, Help Scout, Calendly, Apollo, Greenhouse and many more — grouped into tabs by kind, with search.
Each connection has its own autonomy: ask before any change, ask only before critical actions, or full access with no questions — set per connection, on top of the team's own setting.
Import into Docs: bring in your help centre or docs site by address, or upload PDF, Word, Markdown, text and HTML files. Large sources are split into parts; importing again refreshes what changed and never overwrites a document someone wrote or edited. Website imports run in the background with live progress and a Stop button.
Teams can ask for an import: when a team needs a help centre or docs site that isn't in Docs yet, it asks — and the request waits under Needs you, one click to start.
New Support Desk team: its first job studies your product — docs, code, internal notes and past tickets — and proposes a knowledge base for you to approve, with a playbook and escalation rules. Its on-call trigger comes switched off until you're ready.
Datastores your teams create now scale to any number of teams and tables.
Meetings, hands-off teams, job ordering, and your data under your control
Meetings: put several teams' directors around one table. Watch them talk it through live, add a note to the room, read or download the transcript — and the meeting ends with a summary, decisions and jobs for the teams that agreed them. Directors can call a meeting themselves when they need another team to agree something.
Job ordering: an agreed job can wait for another team's job to finish, then start on its own with that job's report in hand — and is cancelled if the job it waited for fails.
Ask me before…: each team has an approval setting — any change, critical actions only (the new default: sending, deleting, sharing and inviting still ask) or fully autonomous. Reads never ask. The same setting now lets a team create its own datastores, and a fully autonomous team its own tools and sub-teams. New skills always come to you.
Fully autonomous teams with autopilot on take work from each other without asking; anything else handed between teams sideways comes to you first.
Plans now set how many agents your organisation can staff at once (5 on Pay as you go, up to 200 on Company), with the next step up shown when you reach it.
Your data: an owner can export everything (one JSON file, never including secrets) or delete the organisation from Settings → Your data — deleting cancels billing first. Owners and admins can delete any memory from a team's Hive.
Google Docs your teams write now keep real headings, lists, bold, italics and links.
Billing history in the app: every invoice (with its PDF) and top-up receipt, refunds included, under Settings → Billing.
Anthropic's Fast tier runs on Claude Haiku 5.5 — quicker and cheaper for workers — and OpenAI's GPT-6 family is supported through OpenAI's own Responses API.
Workers pick up their next task the moment they finish one, so big fan-outs finish sooner and spend less.
A new Security page explains, in plain English, how your data is kept apart, encrypted and governed.
Grok, thinking and context controls, new pricing, and in-app support
Contact support from inside the app: open a ticket from Support in the sidebar, follow the conversation there, and get an email when we reply. You see the tickets you opened, and your organization's owners and admins see all of its tickets; your teams' agents never can.
New pricing, in US dollars: Starter US$29, Business US$99, Company US$399 a month (previously Crew, Studio and Group, in NZ dollars). Top-ups are US$20 for 4,000 turns.
Pay as you go: no plan needed. Buy turns when you need them (2,000 for US$20, never expiring) and run one team at a time.
Thinking and context, per tier: every tier can now think from Low up to Max (whatever its model supports) and keep Standard, Deep or Max context. Deep thinking takes longer, so a turn at Extra high counts as 2 against your plan and at Max as 4. Defaults are tuned to get work done — Genius thinks hard with Deep context.
xAI's Grok is a first-class model provider: add an xAI key and run any tier on Grok 4.7 or Grok 4.3, with prices and tier defaults built in.
Built on xAI's own API rather than a compatibility layer, so each tier's effort setting reaches Grok, its reasoning carries from one step to the next, and repeated context is billed at the cached rate.
Google Calendar and Drive now connect straight to Google's standard Calendar, Drive, Docs and Sheets APIs. Calendar asks for event access only, and your team can write Google Docs and Sheets it creates. A Calendar or Drive connection added before today needs one click on Reconnect.
Live view, Docs, Microsoft 365 connections, and turn allowances
Live: your organisation as it works — every team at its standing capacity, agents lighting up the seats they take, and every delegation, tool call, memory write and approval drawn as it happens.
Sign in with Google, GitHub or Microsoft (work, school or personal). Google and GitHub sign in to your existing account when they confirm the same verified email.
No more free trial: set up your key, tools and teams first, and they start work once you pick a plan.
Every plan now includes a monthly allowance of agent turns (one model call by one agent; a trigger's check counts as a tenth). It is a hard limit: when it runs out your teams pause until the plan renews or you add a top-up, so a bill can never run away.
Top-ups: NZ$15 buys 3,000 more turns, paid up front, and they never expire.
Docs: a written library for your organisation and for every team, in folders, with full history. Teams search it by section and read what applies before they act. Each team reads the organisation's library, its own, and those of the teams above and below it.
Agents can draft new documents or corrections, but nothing they write reaches another agent until a person publishes it. Drafts wait for you under Needs you.
Connect Outlook mail, calendar, OneDrive and SharePoint with Microsoft sign-in — work or personal accounts. Each person grants access for themselves; most organisations need no admin approval.
Any API, newest Claude models, and effort per tier
Connect any REST API by importing its OpenAPI (or Swagger) spec, by URL or upload. Your team gets search, describe and call tools for every operation, plus typed tools for the ones you pin. Calls are checked against the spec and writes ask you first.
Anthropic teams now run Claude Opus 5.5 (Genius) and Sonnet 5.5 (Smart and Fast).
Set effort (medium or high) separately for each tier in Settings → Model provider.
Bring your own key, everywhere: the Workforce Architect and Hive search now run on your own provider account too. Hive search uses your OpenAI key for embeddings if you add one, and full-text search otherwise.
Invite links are single-use, so a forwarded link can't let in a second person.
Plans, Google Calendar and Drive, and the team at work in 3D
Three plans — Crew, Studio and Group — priced by how many teams run at once, with a 30-day free trial and no card needed. (The free trial ended on 7 October 2026.)
Google Calendar and Drive connections through Google's official servers, alongside Gmail.
A live 3D view of the team at work on every job and team page, drawn from real activity.
DeepSeek, and a simpler model setup
DeepSeek is a first-class model provider, with prices and tier defaults built in. It suits cheap experiments; note that DeepSeek processes data in China.
Model setup is one key for the whole team, with cost and data-residency facts shown per provider.
Bring your own key, Skills, and long-term memory
Teams run on your own model provider account. You pay your provider for tokens, and us a flat monthly fee for the orchestration.
Skills: curated, versioned procedures for how your team operates. You approve each one.
The Hive keeps long-term memory: every finished job's report is filed and recalled at the next kickoff.
Every customer's data is isolated at the database level with row-level security.
Standing teams, Needs you, and tools teams build themselves
Needs you: one place for every approval, proposal and paused team, with email notifications when a hands-off team stops for you.
Raise a team's budget and resume it in one click when it runs out.
Teams can write, test and schedule their own sandboxed tools, with your approval.
Teams can create, see and delegate to other teams; a company-level CEO can run whole departments.
Use any REST API with a stored token, with no MCP server required.