Cloud agent productsUpdated 9 October 202612 min read

Kavela vs Manus

Manus is a capable general agent that gives every task its own cloud computer. Kavela runs a standing team of agents under a lead agent, on your own Claude or ChatGPT plan if you have one.

Yong Hong TanFounder, Kavela

Abstract

Manus is a general AI agent that works in a sandboxed virtual computer, one for each task, and turns a request into research, slides, websites, apps, data analysis or video 1, 2, 3. Manus 2.0, released on 28 September 2026, introduced Cue, a separate early-access app for groups of agents, and Manus Flex followed a day later for using your own API key 3, 4. This report describes how Manus handles six things a team of agents needs, then how Kavela handles the same six. The short version: both give each piece of work its own isolated computer, and both connect to most of the tools a business uses. Kavela adds a lead agent over a standing team, one calendar and one Approvals inbox across all of it, a workspace inbox agents use to sign up for services, and the option to run Claude Code or Codex on the plan you already pay for.

TopicKavelaManus
Your model planClaude or Codex on your own plan, or included Kav AI, chosen per agentManus 2.0 models on credits, or an API key from a supported provider through Manus Flex 5, 4
Managing many agentsA lead agent plans and assigns work to a standing team, on one calendarSub-agents within one task; agent group chats in Cue, a separate early-access app 6, 3
Agent emailA workspace inbox agents use to sign up for servicesMail Manus starts tasks from approved senders; Cue agents send and receive from their own address 7, 8
ApprovalsOne inbox across every agent, as typed cardsPrompts differ by surface; results stay with each task 9, 10, 2
Recurring runsSchedules or events, each agent session in its own cloud computerAn isolated VM per task; 20 concurrent and 20 scheduled tasks on paid plans 2, 11
ConnectionsOAuth, MCP, API keys, secrets or a live browser sign-in, set up by the agentOAuth connectors, custom MCP, API keys for apps it builds, cloud and local browser sign-ins 12, 13, 10, 14
Table 1. Kavela and Manus at a glance.

What Manus is

Manus describes itself as an autonomous general AI agent designed to complete tasks and deliver results. Each task runs in a sandbox: a virtual computer with internet access, a persistent file system and the ability to install software 1. You type a request and choose Agent mode, which plans and works on its own and uses credits, or Chat mode, which does not 5. While Manus works in its cloud browser you can watch every step, and when it reaches a CAPTCHA, an SMS code or a sign-in check it asks you to take over, then hand control back 10.

Manus 2.0, released on 28 September 2026, covers research, slides, websites and apps, data analysis, video and games, and Manus Studio on desktop adds editors for documents, spreadsheets, slides, websites, code and video 3. Manus runs on the web, desktop and mobile, and also takes work from Slack, by email and through an API 3, 15, 7, 16. Projects hold a master instruction and a knowledge base that every new task in the project inherits 17.

Manus has operated independently since 1 September 2026, after its acquisition by Meta was unwound 18, 19. Work is paid for in credits, which cover model tokens, virtual machines and third-party API calls; on 9 October 2026 the pricing page listed plans from $20 to $200 a month 20, 11.

1. Your own model plan

How Manus does it

Manus agents run on Manus's own model tiers: Manus 2.0 Lite, Manus 2.0 and Manus 2.0 Max in Agent mode, with the free plan's daily credits limited to 2.0 Lite 5, 20.

Manus Flex, announced on 29 September 2026, lets you bring an API key from a supported inference provider into the Manus workspace. The launch partners are OpenRouter, Fireworks and Modal, and you set a primary model and a reasoning effort. Model inference is then billed by that provider, while the other services and infrastructure in a Manus task still use Manus credits 4. Neither the announcement nor third-party launch coverage describes signing in with a Claude or ChatGPT subscription 4, 21.

How Kavela does it

Kavela runs on your choice of plan, agent by agent. With your own Claude plan, the unmodified Claude Code runs inside the agent's own cloud computer and you sign in to it yourself; with a ChatGPT plan, the agent can run Codex instead. Anyone who has neither uses Kavela's included Kav AI usage.

Because the choice is per agent, one team can mix Claude and Codex agents, and the subscriptions you already pay for do the model work.

If you already pay for Claude or ChatGPT, that plan can do your agents' model work, in place of a separate per-token bill.

2. Managing a team of agents

How Manus does it

Within a single task, Manus can split work across many agents. With Wide Research, the main agent breaks a request into independent sub-tasks, gives each one a dedicated agent with its own fresh context window, then collects the results and assembles them. Manus says it has been tested up to 250 items 6.

Persistent agents that hand work to each other live in Cue, a separate app launched with Manus 2.0: you put several agents in a group chat with a shared goal, they pass work between them, and you set the direction and make the final call. Cue is in early access and needs an invite code 3. Inside Manus itself, a sidebar lists tasks and projects with filters such as Favorites and Scheduled, and the Automations page has a month view of past runs and upcoming scheduled work 17, 22. The docs describe no standing manager agent that assigns and follows up on work across separate tasks 6, 17.

How Kavela does it

In Kavela the lead agent is Kav. You tell Kav what you want, in the workspace chat or in Telegram, and Kav plans it, hands each part to the right worker agent, follows up and brings back the result. The workers are a standing team: named agents with their own skills, connected apps and run history, and you can hire a new one by describing the job.

Every task from every agent sits on one calendar, by day, week or month, with its start time, due time and live state, so a week of agent work reads like a week of a team's work: what is scheduled, what is running, what needs you and what is done.

Wide Research makes one large task faster. A lead agent with a standing team keeps many separate pieces of work moving, week after week.

3. Agents that can sign up for things

How Manus does it

Mail Manus gives each user a bot address, and you can add addresses for particular workflows. Forwarding or copying an email to one starts a task, and the result comes back by email. Only emails from pre-approved sender addresses can trigger tasks, which makes it a way to start work from your own mail; the docs do not present it as an address for receiving sign-up emails from other services 7.

Cue goes further. Each Cue agent can send and receive email from its own address, and also has its own phone number, wallet and computer 8. Cue is a separate app in early access 3, and its price after early access has not been announced, according to third-party coverage 23.

How Kavela does it

Every Kavela workspace has its own receive-only inbox, Kavela Mail. When an agent signs up for a service on your behalf, it uses a fresh sign-up address from that inbox, waits for the verification email, reads the code or the link and carries on. The wait shows in Approvals. A workspace can hold up to 10 live sign-up addresses at once, each address lasts 24 hours, and received mail is deleted after 30 days.

Kavela Mail does not send, and Cue's agents can. Its job is narrower: getting an agent through account verification inside the same workspace as the rest of your agents. With a live browser sign-in for the services that need one, that lets agents create the accounts a job requires and set those services up.

Most setup work starts with a new account and a verification email; an agent that handles both inside your main workspace finishes more of the job.

4. What agents need, and what they deliver

How Manus does it

Manus asks for permission in different ways depending on where it is working. The desktop app's My Computer feature asks before every local command, with Allow Once or Always Allow 9. Browser Operator needs a one-time Authorize for each task in your own browser 14. The cloud browser asks you to Take Over for verification steps 10. Mail Manus sends a confirmation email before starting a complex task, and you review an advanced automation before switching it on 7, 22.

Results stay with the task that made them. Its files are listed under View all files in this task, and a task can be shared by link or played back as a replay 2, 6. Automation runs appear in an Updates feed with their status and a short summary, and a run can be marked as needing attention 22. The docs describe no single inbox where requests from every task collect 22, 2.

How Kavela does it

Kavela collects everything your agents need into one Approvals inbox for the whole workspace. Each request is a typed card you answer where it is: connect an app, sign in to a site in a live browser, approve an action such as a post before it goes out, choose between options, or let an agent know the email it was waiting for has arrived. Other work carries on while a card waits.

What agents produce arrives as deliverables attached to the task that made them, from documents and reports to sites and files, listed under Delivered and opened in one click.

With every request from every agent in one inbox, checking on ten agents takes as long as checking on one.

5. Many agents, running on schedule

How Manus does it

Manus gives every task its own sandbox, which it describes as a fully isolated cloud virtual machine that does not affect other tasks and can run in parallel, with root access. Idle sandboxes sleep and are recycled after 7 days on the free plan or 21 days on Pro, and intermediate code and temporary files are not restored afterwards 2. Each paid plan allows 20 concurrent tasks and 20 scheduled tasks 11. For an always-on server, Manus sells Cloud Computer separately; if payment lapses, the machine and its files are deleted 24.

Automations come in three types: on a schedule, triggered by an event from sources such as Gmail, Outlook, Notion, GitHub, Shopify, a webhook or Mail Manus, and advanced automations built from plain language, which are in beta. You can pause, test, edit or delete each one, and some trigger sources are polled, so they may not fire at once 22.

How Kavela does it

Kavela isolates work in a similar way. Each agent session works in its own cloud computer, with a shell, files, git and a browser, which starts when work starts and sleeps when it stops. One agent's work never slows another's, and work keeps going after you close the tab. The workspace's files persist between runs, so tomorrow's run starts from today's output.

Automations run on a schedule or on an event, such as a webhook or a merged pull request, and can be limited to an active window with start and end dates. There is no cap on how many are switched on at once.

Both products isolate each piece of work, so for recurring work the choice comes down to limits, triggers and where the results land.

6. Connecting your tools

How Manus does it

Manus has prebuilt connectors that sign in with OAuth, including Gmail, Google Calendar, Google Drive, Notion, HubSpot, Stripe, Slack and GitHub 12. You can add a custom MCP server in Settings with its URL and an API key, bearer token or other credentials 25, and Zapier, premium data sources and the Manus API extend the list 26.

For web apps Manus builds, you can give it an API key in chat, and it stores the key as an environment variable on the server 13. You can sign in to your accounts in the cloud browser, where sessions persist, or let Browser Operator work in your own browser with its existing sessions and IP address. Manus notes that the cloud browser runs from data-center addresses, so some sites may ask for extra verification, and suggests your own browser in those cases 10, 14.

How Kavela does it

Kavela connections belong to the workspace: connect a service once and every agent can use it. A service can be connected by OAuth sign-in, as an MCP server, with an API key, with custom secrets, or by signing in to it in a live browser you control. Passwords and keys go into the card, never through the model.

You can also leave it to the agent. With Let Kav set it up, the agent reads the service's documentation, works out the right way to connect it and does the setup, creating the account first if there is none.

Both products reach most of the tools a business uses. Kavela adds an agent that works out the setup from the service's own documentation, account included.

Where Manus is the better choice

Manus is the better choice for one large job you want done in parallel, such as researching hundreds of companies: Wide Research gives each item its own agent and assembles the results, and Manus has tested it up to 250 items 6. It also produces a wide range of finished work, including hosted websites and video edited on a timeline in Manus Studio 3.

Manus meets you in more places, with web, desktop and mobile apps, Slack, email and an API 3, 15, 7, 16, and its automations have event triggers with conditions and test runs 22. If you want agents with their own phone numbers and wallets, Cue offers that today, in early access 8, 3.

Conclusion

Manus is a capable general agent, and on two of the six points it matches Kavela: every task gets its own isolated computer, and it connects to most of the tools a business uses. The differences are in how work is organised around the agents. Kavela runs a standing team under a lead agent, puts every task on one calendar and every request in one Approvals inbox, gives the workspace an inbox for sign-ups, and runs Claude Code or Codex on the plan you already pay for.

If you want one capable agent to take a large task and come back with a finished result, Manus does that well. If you want a team of agents working across your business week after week, use Kavela.

Questions

Can I use my Claude or ChatGPT subscription in Manus?

Manus Flex lets you bring an API key from a supported inference provider, such as OpenRouter, Fireworks or Modal, billed by that provider. Its announcement does not describe signing in with a Claude or ChatGPT subscription. In Kavela, each agent can run Claude Code on your Claude plan or Codex on your ChatGPT plan, or use the included Kav AI usage.

Can Manus run several agents together?

Within one task, Wide Research splits the work across sub-agents. Agents that hand work to each other in a group chat are in Cue, a separate early-access app. In Kavela, Kav leads a standing team of named agents and puts all of their tasks on one calendar.

Do Manus agents have their own email address?

Mail Manus lets you start a task by emailing a bot address from an approved sender, and Cue agents send and receive email from their own addresses. Kavela gives every workspace a receive-only inbox that agents use to sign up for services and read verification emails.

Does Manus give each task its own computer?

Yes. Manus gives each task its own isolated virtual machine, and Kavela gives each agent session its own cloud computer. In Kavela the workspace's files persist between runs, and paid plans have no small cap on automations switched on at once.

References

  1. Welcome, Manus docs, accessed 9 October 2026.
  2. Understanding Manus sandbox - your cloud computer, Manus blog, 14 January 2026.
  3. Introducing Manus 2.0, Manus blog, 28 September 2026.
  4. Introducing Manus Flex: Use your own API key in the Manus workspace, Manus blog, 29 September 2026.
  5. What are the differences between Chat and Agent mode, Manus Help Center, accessed 9 October 2026.
  6. Wide Research, Manus docs, accessed 9 October 2026.
  7. Mail Manus, Manus docs, accessed 9 October 2026.
  8. Cue, Manus (Cue), accessed 9 October 2026.
  9. Desktop (My Computer), Manus docs, accessed 9 October 2026.
  10. Cloud browser, Manus docs, accessed 9 October 2026.
  11. Pricing, Manus, accessed 9 October 2026.
  12. MCP Connectors, Manus docs, accessed 9 October 2026.
  13. Website Builder: Third-Party Integrations, Manus docs, accessed 9 October 2026.
  14. Manus Browser Operator, Manus docs, accessed 9 October 2026.
  15. Slack integration, Manus docs, accessed 9 October 2026.
  16. Manus API, Manus docs, accessed 9 October 2026.
  17. Projects, Manus docs, accessed 9 October 2026.
  18. A note to our users, Manus blog, 11 August 2026.
  19. Manus resumes independent operations, Manus blog, 1 September 2026.
  20. What are the rules for credits consumption and how can I obtain them, Manus Help Center, accessed 9 October 2026.
  21. Manus Flex lets you bring your own API key, Tech City Authority (third-party), 30 September 2026.
  22. Automations, Manus docs, accessed 9 October 2026.
  23. Manus Launches Cue Agents With Phone Numbers and Wallets, Implicator.ai (third-party), 28 September 2026.
  24. Understanding Cloud Computer plans and billing, Manus Help Center, accessed 9 October 2026.
  25. Custom MCP Servers, Manus docs, accessed 9 October 2026.
  26. Integrations, Manus docs, accessed 9 October 2026.

Checked 9 October 2026. Products change; tell us if something here is out of date.