29 best Manus alternatives in 2026
By Marcell Havlik, founder of TODO for AI · Updated
A general-purpose agent that researches and builds documents, slides, sites and videos in its cloud, with a desktop app that can work in approved local folders. Below: what to switch to, scored on the six things that matter, and an honest word on when to stay.
Key takeaways
- Best overall. TODO for AI scores 29/30 against Manus's 12/30 on our six-dimension scale: any model provider on one plan, your own PC, a cloud VM and your logged-in Chrome, and one board your team and agents share.
- Models. Manus model choice: Vendor-selected, or your own key via Manus Flex. TODO for AI runs any provider, Claude, GPT, Gemini, Grok or a local Ollama model, and lets one review another's work.
- Benchmark. TODO for AI scored 89.9% on Terminal-Bench 2.1 with Opus 5.5 (high) (own harness, 80/89 tasks, 2026-10-08; not an official leaderboard submission).
- Free options. TODO for AI has a free plan (Claude Sonnet only). Notion, ClickUp, Linear, Asana and 7 more also have free plans. OpenHands, OpenClaw, Hermes Agent, Aider and 2 more are free software, but you pay your model provider per token.
- Open source. The fully open-source Manus alternatives are OpenHands, OpenClaw, Hermes Agent, Aider and 2 more. TODO for AI publishes its edge runtime and CLI.
- When to stay. Stay on Manus if you want a polished consumer agent for research, slides, sites or video.
Manus
TODO for AI
Any providerWhy people look past Manus
Manus popularised the general-purpose cloud agent: give it a goal, it browses, researches and hands back a document, a deck, a site or, since Manus 2.0 in September 2026, a first-cut video. It needs no setup. For a non-technical user who wants a finished artefact from one prompt, it is a strong choice.
It is no longer cloud-only. The desktop app's My Computer feature works in folders you authorise and runs command-line tools, with approvals you can grant once or always, and Browser Operator drives your local Chrome. Manus Flex (September 2026) lets you bring your own API key through partner providers such as OpenRouter. What remains: credit pricing that is hard to predict, local access that goes through a desktop app on macOS or Windows, and projects where colleagues share instructions and files but not each other's tasks.
Cost per task is hard to predict and heavy work is expensive. Flex moves inference to your own provider bill, but the credit plans remain the default.
Local work goes through the Manus desktop app on macOS or Windows, in folders you add one at a time. There is no Linux machine or server of your own to run it on, and the always-on option is a paid Cloud Computer.
By default Manus routes between vendors for you. Flex adds bring-your-own-key through its partner providers; there is no cross-vendor review of one model's work by another.
A project holds shared instructions and files and you can invite colleagues to it. Tasks are private by default and shared one at a time, so the project shows what everyone agreed to publish rather than what everyone is working on.
Most of this list solves a different problem
Manus gets the task done.
TODO for AI organises the whole company’s work.
Manus answers one question: how do I get this task done, mostly in their own cloud? So do the others — Cursor inside the editor, Claude Code in one person's session, Lindy by asking its assistant.
We answer a different one: where does the work live? One board per project, and a project for each part of the company — engineering, marketing, ops, finance, one per client. Your colleagues work on it. So do your agents, as many as you want to run, each with its own model, tools and permissions. Items carry the people who worked them and the agent that ran them, and the board filters by person, so what the company has in flight is one screen instead of a Slack thread.
That is not a team plan bolted onto an AI tool. It is the product. The agents are how items get closed; the board is what the company runs on.
How do the Manus alternatives compare?
| Tool | Interface | Models | Open | From | Score /30 |
|---|---|---|---|---|---|
| TODO for AI | Web, desktop, mobile, CLI, IDE (Zed, JetBrains, VS Code), voice | Any provider: Claude, GPT, Gemini, Grok, Ollama | Partly | Free | 29 |
 n8n | Web, self-hosted or cloud | Any, you supply the keys | Partly | Self-host free | 23 |
 OpenHands | Web, CLI | Any provider | Yes | Free (self-host) | 22 |
 OpenClaw | Chat apps, CLI, desktop | Any provider (BYOK) | Yes | Free | 21 |
 Hermes Agent | Chat apps, CLI, desktop app | Any provider (BYOK or Nous Portal) | Yes | Free | 20 |
 Notion | Web, desktop, mobile | Claude, GPT, Gemini, Grok, you pick | No | Free | 19 |
 ClickUp | Web, desktop, mobile | ClickUp Max, Claude, GPT, Gemini, you pick | No | Free | 18 |
 Linear | Web, desktop, mobile | Linear Agent: Claude, GPT or GLM, picked per workspace; delegated agents bring their own | No | Free | 17 |
 Aider | Terminal | Any provider (BYOK) | Yes | Free | 16 |
 Lindy | Web builder | Most major models, you pick | No | $30/mo | 16 |
 Asana | Web, desktop, mobile | Vendor managed for AI Teammates; AI Studio rules can pick OpenAI, Anthropic or Google | No | Free | 16 |
 GitHub Copilot | IDE, GitHub.com, CLI | OpenAI, Anthropic, Google, xAI, more | No | Free | 16 |
 OpenCode | Terminal | Any provider (BYOK) | Yes | Free | 15 |
 Cline | IDE extension | Any provider (BYOK) | Yes | Free | 15 |
 Copilot Cowork | Web, Copilot desktop app, mobile | Anthropic and OpenAI, picked by Microsoft | No | $18/mo + usage | 15 |
 Make | Web | Vendor managed, or bring your own LLM key | No | Free | 15 |
 OpenAI Codex | Terminal, ChatGPT web, IDE | OpenAI only | Partly | $8/mo | 14 |
 Cursor | IDE | Multi-vendor, metered | No | $20/mo | 14 |
 Claude Cowork | Claude Desktop, web, mobile, Chrome side panel | Anthropic only | No | $20/mo | 14 |
 monday.com | Web, desktop, mobile | Vendor managed | No | Free | 14 |
 Gumloop | Web, Chrome extension | Vendor managed, several available | No | $37/mo | 14 |
 Claude Code | Terminal, IDE, desktop app, web | Anthropic only | No | $20/mo | 13 |
 Zapier | Web, Chrome extension | Vendor managed | No | Free | 13 |
 Airtable | Web, desktop, mobile | Vendor managed | No | Free | 13 |
 ChatGPT | Web, desktop, mobile | OpenAI only | No | Free | 12 |
 Trello | Web, desktop, mobile | Vendor managed | No | Free | 12 |
 Lovable | Web builder | Vendor-selected | No | $25/mo | 11 |
 Base44 | Web builder | Vendor-selected, choice on paid plans | No | $16/mo | 11 |
 Motion | Web, desktop, mobile | Vendor managed | No | $19/seat | 11 |
| Manus (baseline) | Web, desktop (macOS, Windows), mobile | Vendor-selected, or your own key via Manus Flex | No | $20/mo | 12 |
How we scored
Each tool gets 0 to 5 on six dimensions: model freedom, long-running work, beyond code, openness, team & permissions, cost predictability. The sum is the score in the table. Facts (pricing, interface, models, licence) were checked against each vendor's public pages on 2026-10-08.
The ratings are ours, and we are on the list, so weigh them accordingly. Every fact behind a rating is checkable, and where a competitor is the better pick for a job we say so in its card.
Each alternative, honestly
TODO for AI
Any model provider on one flat plan, a persistent cloud VM, your real Chrome, and CLI tools for marketing, SEO, email and ops. 89.9% on Terminal-Bench 2.1, all 89 tasks, same model. And one board your team and your agents share, with a project for each part of the company.
- + You want a Manus alternative that runs on your own Linux, macOS or Windows machine, plus an always-on cloud VM included in the plan
- + You want to pick and mix models on one plan, with cross-vendor review
- + You want colleagues and agents working one shared list, not private tasks
- - You want a polished consumer agent for research, slides, sites or video
- - You mostly work in its cloud and touch local files occasionally
n8n
Fair-code workflow automation you can self-host, priced per whole workflow run.
- + Self-hostable, so data and cost stay yours
- + Executions count whole runs, not steps, so complex workflows do not cost more
- + A Code node runs real JavaScript or Python inside a workflow
- + The strongest AI agent and LangChain node set of any automation tool here
- - Genuinely technical: most non-developers will not build in it
- - Fair-code, not OSI open source, and the Business tier is self-hosted only at €667
- - You still design the workflow before anything runs
- - Self-hosting means you own the uptime, upgrades and secrets
OpenHands
Open-source autonomous software agent with a self-hosted or cloud workspace.
- + Open source
- + Sandboxed workspace per task
- + Multi-model
- - Focused on software tasks
- - Self-hosting is real ops work
- - No real-browser or business integrations
OpenClaw
Open-source personal agent that runs on your own machine and answers in WhatsApp, Telegram, Slack and 20+ other chats, bring your own keys.
- + Fully open source, MIT, run by a non-profit foundation
- + Any model through your own keys, including local Ollama
- + Reachable from the chat apps you already use
- + State, memory and credentials stay on your hardware
- - You run and secure it: gateway, exposed ports, plugin supply chain
- - Per-token bills per provider, no pooled plan
- - No hosted machine, so it stops when your box does
- - You install, configure and update it yourself
Hermes Agent
Open-source self-improving personal agent from Nous Research: skills it writes itself, memory that persists, reachable from 20+ chat apps.
- + Fully open source, MIT, 30k+ GitHub stars
- + Writes and refines its own skills from experience, memory persists across sessions
- + Any model via own keys, OpenRouter or Nous Portal
- + Cron scheduler, subagents, seven sandbox backends
- + Migrates from OpenClaw in one command
- - You host, secure and update it, or rent a VPS
- - No hosted machine bundled; cloud browsers are third-party providers you configure
- - No persistent task list, work lives in chat threads and cron jobs
- - No team roles or pooled usage
Notion
The docs-and-databases workspace, with Custom Agents that run on schedules and triggers inside it.
- + Custom Agents pick their own model: Claude, GPT, Gemini, Grok or Auto
- + Run on schedules and on triggers, in Notion or from Slack, without anyone watching
- + Shareable like teammates, with full access, edit or run-only permissions
- + MCP connections to 18+ apps plus custom MCP servers an admin enables
- + Activity logs showing what the agent did and when
- - No shell anywhere: not your machine, not a cloud VM, not a sandbox
- - Reach is the MCP connections an admin wired, plus optional web browsing
- - Cannot act in your logged-in browser session
- - Credit metering on top of seats, so a busy agent is a second bill
ClickUp
All-in-one work platform whose Super Agents can be @mentioned and assigned tasks like teammates.
- + Super Agents behave like team members: @mention them in comments, assign them tasks
- + Pick the model or let Brain route between Claude, GPT and Gemini
- + Autopilot Agents scoped to Spaces and Lists, on recurring schedules
- + MCP both ways, so external agents can work the workspace too
- + The board, docs, chat and whiteboards are already one product
- - No direct shell and no machine: agents act on workspace objects and connected tools
- - Two subscriptions stacked, workspace seat plus AI seat, then credits on top
- - Agent capability is a list of skills rather than an open-ended runtime
- - Notoriously heavy product if all you wanted was a list
Linear
The issue tracker engineering teams like, with its own Linear Agent and coding sessions in a Linear sandbox.
- + The best shared board in software, and agents are first-class citizens on it
- + Linear Agent built in: chat, @Linear in comments or Slack, Skills, MCP connections and Loops on a schedule or event
- + Coding sessions in a Linear-hosted sandbox that installs the toolchain, runs the app in a browser to test it and opens the PR
- + Admins pick the model: Claude, GPT or GLM, billed at provider rates
- + Triage Intelligence suggests team, assignee, labels and duplicates
- + 30+ outside agents assignable too, including Claude Code, Codex, Cursor, Copilot and Devin
- - Built around product and engineering: the sandbox needs a GitHub repo, and non-code reach is whatever MCP connections exist
- - The shell is Linear's sandbox, not a machine you own
- - Its browser is the sandbox's, for testing your app, not your logged-in Chrome
- - Model is a workspace setting, not a choice per task or per agent
- - Outside agents are a second subscription
Aider
The original open-source pair-programming CLI, git-native, any model.
- + Very token efficient
- + Git-native commits
- + Runs anywhere Python runs
- - Pair-programming, not autonomous: you drive each step
- - No browser, no cloud, no scheduling
- - Code-only
Lindy
No-code AI teammates for business workflows: email, CRM, meetings and support, built from triggers and steps.
- + The strongest no-code builder here for business workflows
- + Large catalogue of triggers and app integrations
- + You pick the model per task, and computer use is included
- + Credits pool across the whole workspace
- - Built around defined workflows rather than open-ended work
- - Credits are an abstraction, and the rate changes per plan
- - Lives in its own cloud: no shell on your machine, no repo
- - Closed source
Asana
Work management with assignable AI Teammates, Dash, and a no-code studio for cross-system workflows.
- + 30+ prebuilt AI Teammates you assign work to in plain language, plus your own
- + Teammates search, read and write across 10+ connected apps: Gmail, Slack, Drive, Microsoft, HubSpot
- + Included in paid plans since September 2026, with Dash as an assistant on top
- + AI Studio builds multi-step workflows, with a model choice per rule
- + Mature permissions, reporting and enterprise governance
- - Reach is Asana plus its connected apps: no shell, no machine, no browser session
- - AI Teammates run on models Asana picks
- - 5 AI requests per user per month, then billed per request
- - A work management product first; heavy if all you wanted was the work done
GitHub Copilot
GitHub's coding agent: completions in the editor, a cloud agent that takes an issue and opens a pull request.
- + The widest model picker in a coding agent: OpenAI, Anthropic, Google, xAI, Moonshot, Microsoft
- + The cloud agent works from an issue and opens a pull request, in GitHub's infrastructure
- + Copilot CLI runs in your own terminal with per-command approval
- + Cheapest serious entry at $10, and a real free tier
- + Third-party agents, Claude Code and Codex, run inside the same GitHub surface
- + Already installed wherever your code already is
- - Everything is repo-shaped: the output is a pull request or it does not exist
- - The cloud agent runs on GitHub's runners, so long-running work leaves your machine
- - No business tooling, no inbox, no ad account, no logged-in browser
- - Premium requests are metered per model, so heavy agent use bills on
OpenCode
Open-source terminal coding agent, bring your own keys, any provider.
- + Fully open source
- + Any model via your own keys
- + Fast terminal UI
- - Pay-per-token bills you have to track yourself
- - No cloud machine, no scheduling, no team features
- - Code-only
Cline
Open-source VS Code agent extension, bring your own keys, plan and act modes.
- + Open source
- + Any model
- + Good MCP support
- - Runs only while VS Code is open
- - Token bills add up fast on frontier models
- - Code-only
Copilot Cowork
Microsoft 365 Copilot's agent: describe a task and it does the work across Outlook, Teams, Word, Excel, PowerPoint and SharePoint.
- + Does the work inside Outlook, Teams, Word, Excel, PowerPoint, OneDrive and SharePoint, with your Microsoft 365 permissions
- + Runs in Microsoft's cloud, so tasks continue with your laptop closed
- + Admin controls, spending limits and audit trail an IT department expects
- + Anthropic models at launch, OpenAI in the Frontier program, a Microsoft model coming
- - Built for Microsoft 365: your other tools, your servers and your code are outside it
- - A Copilot license first, then metered credits per task
- - No shell on your own machine, and web browsing is a Frontier preview through Edge
- - Microsoft picks the model lineup, and every task starts from a prompt
Make
Visual scenario builder with credit-metered runs and AI agents on every plan.
- + The visual canvas makes branching, iteration and error handling legible
- + Materially cheaper per run than Zapier at comparable volume
- + AI agents and AI Toolkit are on every plan, including Free
- + You can point it at your own LLM key instead of the vendor's
- - You draw the scenario before you get value, and a one-off never earns that
- - Credits are per module action, so the complex scenarios cost the most
- - Credits expire at the end of the billing term
- - No shell, no machine, no logged-in browser session
OpenAI Codex
OpenAI's coding agent: an open-source CLI plus a cloud sandbox inside ChatGPT.
- + Strong token efficiency on GPT-5.6
- + CLI is open source
- + Cloud tasks run in parallel from the ChatGPT UI
- - OpenAI models only
- - Cloud sandbox is ephemeral and repo-scoped
- - No browser automation, no business tooling
Cursor
A VS Code fork with built-in AI chat, unlimited tab completion and repo-scoped cloud agents.
- + Best-in-class inline editing and tab completion
- + Multi-model
- + Familiar VS Code experience
- - Cloud agents start from a repo and end at a pull request, so non-code work has no shape to run in
- - Agent usage is metered by model, so bills track habits not headcount
- - No business tooling: the browser is there to verify your app, not to run your accounts
Claude Cowork
Anthropic's agent, merged into Claude itself in September 2026: ask for anything and Claude decides whether it is an answer or a task it works through on its own.
- + Real autonomy on multi-step work, with scheduled runs on Anthropic's servers
- + Now simply part of Claude on desktop, web and mobile, no separate mode to learn
- + Built-in browser automation plus Claude in Chrome
- + Claude-grade output quality
- + Included on every paid plan, no extra tool to buy
- - Anthropic models only, so no second model checks the first
- - Runs in Anthropic's sandbox, not on your own machine or in your repo
- - Extends through MCP connectors rather than any CLI you install
- - Closed source
monday.com
Work OS whose AI agents sit in board columns and automations, metered in AI credits.
- + AI agents live in the board everyone already opens: columns, automations, assignees
- + Sidekick assistant and AI columns cover summarising, categorising and drafting in place
- + Separate products for CRM, dev and service, each with its own agents
- + Mature permissions, dashboards and reporting
- - Free plan has no AI at all, so the real entry price is a paid seat
- - AI credits run out and the agents stop; a busy month costs more than the seat
- - No shell, no machine, no session in your browser
- - You do not choose the model
Gumloop
AI-native automation for business teams: chat-driven agents plus flows, priced in credits.
- + Built AI-first rather than bolted onto a 2012 automation tool
- + Agents take tasks in chat, Slack, Teams, Gmail and by voice
- + Nodes are AI steps by default: scrape, summarise, classify, write
- + Genuinely usable by marketing and ops people, not just developers
- - Entry price is $37 a month, the highest starting point in automation here
- - An 8% orchestration fee on top of credits
- - Reach ends at connected apps and the web
- - No shell, no persistent machine, no session in your own browser
Claude Code
Anthropic's coding agent, tied to Claude models and Claude subscription tiers.
- + Excellent code quality on Opus and Sonnet
- + Mature terminal UX, hooks, subagents
- + Chrome extension for browser tasks
- + Cloud sessions and routines run scheduled work without your machine
- - Anthropic models only, no second opinion
- - Cloud sessions are repo-shaped: GitHub repos, allow-listed network, claude/ branches
- - Built for repos, not for marketing, email or ops workflows
- - Closed source
Zapier
The largest integration network, plus Agents you can chat with on a separate activity meter.
- + Roughly 8,000 app integrations, far more than anyone else here
- + Zaps are the most reliable way to wire two SaaS tools together
- + Agents can browse and use attached knowledge without you building the flow
- + Non-technical staff genuinely ship automations in it
- - Zaps are trigger-shaped: something must fire before anything happens
- - Two separate meters, tasks and agent activities, each with its own plan
- - No shell, no machine, no session in your real browser
- - Per-action pricing punishes exactly the workflows that work
Airtable
Database-shaped workspace whose Omni agents build apps and act on records, metered in AI credits.
- + Structured records are a better substrate for agents than free text
- + Omni builds working internal apps and agents from a description
- + Building is free and unlimited; only running consumes credits
- + Credits pool across the workspace instead of stranding per seat
- - $45 a seat on Business is the steepest entry-level seat among the work platforms here
- - Everything must first be modelled as a base, table and field
- - No shell, no machine, no logged-in browser session
- - A document-heavy month burns credits fast at ~200 each
ChatGPT
The default AI assistant, with agent mode, deep research and connectors bolted onto the chat product.
- + The best general assistant for everyday questions
- + Deep research and agent mode are genuinely good
- + Connectors into Drive, email and calendars
- + A usable free tier, and the cheapest paid entry at $8
- - OpenAI models only
- - Caps are per feature, so you hit a wall in one place while credit sits unused in another
- - Agent mode runs in OpenAI's sandbox, not on your machine or in your repo
- - No shell on your machine, no CLI tools, and its browser is not your logged-in Chrome profile
Trello
The simplest kanban board, with AI that summarises and organises cards rather than working them.
- + The lowest-friction board anyone will actually adopt
- + Cheapest paid seats here, from $5 a month
- + AI summarises cards, generates checklists and schedules focus blocks
- + Free plan is genuinely usable
- - AI assists the card, it never completes the work on it
- - No agents, no shell, no machine, no browser session
- - Thin beyond kanban: reporting and structure are limited
- - You do not choose the model
Lovable
A prompt-to-app builder: describe a web app, get a hosted React front end with a built-in backend, edited by chat.
- + Fastest path from a sentence to a deployed web app
- + Backend, auth, database and hosting are built in
- + Two-way GitHub sync, so the code is yours
- + Workspace roles, per-member credit caps and SSO on Business
- - Builds its own apps: it cannot work in a codebase it did not generate
- - Every edit is a credit, and a wrong turn costs the same as a right one
- - The model is picked for you
- - Nothing outside the app: no browser, no email, no ops
- - Closed source
Base44
Wix-owned prompt-to-app builder with backend, auth and hosting included, metered in messages and integration credits.
- + Cheapest entry in the category at $16
- + Backend, auth, database and hosting included on the free plan
- + Model selection and GitHub sync on paid plans
- + Wix ownership gives it a long runway
- - Builds its own apps: no way into an existing codebase
- - Two credit meters, so a busy app spends integration credits after you stop prompting
- - Nothing beyond the app: no browser, email or ops
- - Closed source, Wix-hosted
Motion
AI calendar and task manager that schedules your work for you, credits included per seat.
- + Auto-scheduling genuinely solves the "when will I do this" problem
- + Tasks, calendar, docs and meetings in one place
- + Credits are generous per seat compared with the work platforms
- + Strong for individuals and small teams drowning in commitments
- - Plans your work, does not do it: the task still lands on you
- - No permanent free tier: $19 a seat after the trial, and that price needs annual billing
- - No shell, no machine, no browser session, no repo
- - You do not choose the model