Roundup

Best AI agent apps in 2026: 17 apps compared

Agent apps now range from crews on your own computer to hosted assistants and coding agents. Here are 17, grouped by kind, with what each is for, where it runs, which models it uses and what it costs, all from official pages.

Sources checked October 30, 2026Sources

The short answer

If you want a crew with roles and a task board on hardware you control, try APX; for one assistant across your chat apps, OpenClaw or Hermes Agent. If you'd rather not install anything, the hosted assistants fit: Claude Cowork or ChatGPT dots if you already pay for Claude or ChatGPT, Manus for finished deliverables, Grok Bot or Viktor for teams. For code, Claude Code, Codex, Cursor, Antigravity and Devin are built for it, and APX can run Claude Code and Codex for its agents.

At a glance

Scroll the table sideways to see every column.

Best AI agent apps in 2026: 17 apps compared — At a glance
App	Best for	Runs on	Models	Price
APX
APX	An ongoing crew for your projects	Your computer or server; desktop, Android, web, terminal	Your ChatGPT plan, APIs, Claude Code, Codex, ACP, local	Free, MIT. Model and service usage can cost extra.
Crews and personal agents
OpenClaw	A personal assistant across your chat apps	Your own hardware; control UI, terminal, desktop and mobile apps	Hosted and local providers; swappable model plugins	Free, MIT. Model and service usage can cost extra.
Hermes Agent	An assistant that improves its own skills	Your computer or a server; terminal UI, desktop app	Multiple model providers; MCP, skills and plugins	Free, MIT. Providers and hosted execution can cost extra.
OpenBot	Persistent AI teammates on your desktop	Desktop app for macOS, Windows, Linux; mobile companions	Codex, Claude Code, ACP agents, compatible and local endpoints	Source available under PolyForm Noncommercial; provider costs separate.
Paperclip	Running agents as a company with governance	Self-hosted web dashboard on your laptop, Docker or server	Claude Code, Codex, Gemini CLI, OpenCode and more	Free, MIT. Agent and model usage can cost extra.
Hosted assistants
Claude Cowork	Documents, research and files inside Claude	Anthropic's cloud; Claude Desktop, web and mobile apps	Anthropic's Claude models; MCP connectors and plugins	Paid Claude plan, from Pro at $20/month; usage limits apply.
ChatGPT dots	One always-on agent for ChatGPT users	OpenAI's cloud computer; ChatGPT apps, Slack and Teams	OpenAI models; ChatGPT plugins	Included in ChatGPT Pro (from $100/month) and Business Premium.
Manus	Finished reports, slides, sites and videos	Manus' cloud; web, desktop and mobile apps, Slack, Telegram	Manus' own model options; MCP connectors	Free with 300 daily credits; Pro from $20/month.
Grok Bot	Hosted Bots that work while your laptop sleeps	A cloud computer in the US; desktop, mobile, X	Picks the model per task; apps and custom MCP	Included with X Premium ($30/month), Super ($100) and Ultra ($200).
Viktor	One shared AI teammate in Slack or Teams	Viktor's cloud; Slack, Microsoft Teams, Discord and web	Claude, GPT, Gemini and others; MCP and integrations	Free with up to $100 in credits; Team from $50/month.
Muse	Personal errands and life tasks	A cloud VM run by Meta; mobile, web, WhatsApp, Mac	Meta's Muse Spark model; built-in and custom connectors	Free with a weekly limit; Power $20/month; Maximum $100/month.
Instinct	An assistant you text or call	A hosted US service; text, calls, web and Mac	Unnamed third-party language models; Apple, Google, Facebook accounts	Some features are paid; no public price list.
Coding agents
Claude Code	Coding with Claude in your terminal or IDE	Terminal, IDEs, desktop, web and mobile; local or cloud	Anthropic's Claude models, also via Bedrock, Google Cloud, Foundry	In Claude Pro ($20/month), Max, Ultra, Team, Enterprise; or API.
Codex	Coding with OpenAI models, included in ChatGPT	ChatGPT apps, CLI, IDE extension; local or cloud	OpenAI models; Responses API-compatible custom providers	In ChatGPT Free, Go, Plus ($20/month) and Pro; or API.
Cursor	An AI code editor with cloud agents	Editor and CLI on macOS, Windows, Linux; cloud VMs	Grok and Composer included; Claude, GPT, Gemini at API price	Free; Plus $8, Premium $30, Super $100, Ultra $200 a month.
Google Antigravity	Agent teams for coding, free to start	App, IDE and CLI on macOS, Windows, Linux	Gemini, Claude and GPT-OSS models, by plan	Individual $0 with weekly limits; business from $30 per seat.
Devin	Turning tickets and alerts into pull requests	Devin's cloud VMs; Desktop IDE, CLI, Slack, Teams	Anthropic, OpenAI, Google, Cognition and open models; MCP, ACP	Free; Pro $20/month; Max $200/month; Teams from $80/month.

By the numbers

Coding Agent Index v2.0 (FrontierCode v1.1, Terminal-Bench 4.0, SWE-Atlas-QnA-AA, equal weights)
Agent and model	Index score (higher is better)	API cost per task (lower is better)	Time per task (lower is better)
Claude CodeClaude Code - Claude Opus 5.5 (Max, Default Fallback)	62	$11.42	61 min
DevinDevin Fusion CLI - Claude Opus 5.5 Max + GPT-6 Luna High	62	$8.00	51 min
CodexCodex - GPT-6 Astra (Max)	55	$7.04	25 min
CodexCodex - GPT-6.1 Sol (Max)	54	$1.50	24 min
Google AntigravityAntigravity CLI - Gemini 4 Argon (High)	53	$11.73	35 min

Index: Coding Agent Index v2.0 (FrontierCode v1.1, Terminal-Bench 4.0, SWE-Atlas-QnA-AA, equal weights)
Source: Artificial Analysis Coding Agent Index ↗
Methodology: Artificial Analysis coding agents methodology ↗
Read on: October 30, 2026

Coding agents only: APX, Cursor and the other apps here are not in the index. Cost is pay-per-token API pricing, not a plan price; scores are averages across three public benchmarks and the index changes between versions.

The apps

01APX
A local agent office: a team of named agents organized by projects, roles and a shared task board, with persistent memory, skills, scheduled routines and chat channels such as Telegram and Discord. It runs on your computer or a server you control and works with model providers, local models and coding agents such as Claude Code and Codex.
Choose it if… You want a crew with distinct roles and a task board, on hardware you control.
Best for
An ongoing crew for your projects
Runs on
Your computer or server; desktop, Android, web, terminal
Models
Your ChatGPT plan, APIs, Claude Code, Codex, ACP, local
Price
Free, MIT. Model and service usage can cost extra.
Sources (15)
02OpenClaw
An open-source personal assistant that lives across your devices and a broad catalog of chat channels, including WhatsApp and Telegram. It has workspace memory, skills, scheduled jobs and multi-agent routing, and runs on your own hardware.
Choose it if… You want one assistant reachable from many chat apps, with a large plugin ecosystem.
Best for
A personal assistant across your chat apps
Runs on
Your own hardware; control UI, terminal, desktop and mobile apps
Models
Hosted and local providers; swappable model plugins
Price
Free, MIT. Model and service usage can cost extra.
APX vs OpenClaw →
Sources (5)
03Hermes Agent
An open-source assistant built around a skill-learning loop: it curates memory, searches past conversations and refines its skills as you use it. It has a terminal UI, a desktop app, built-in cron and a messaging gateway for Telegram, Discord, Slack and more.
Choose it if… You want one assistant that learns skills from experience, reachable from your chat apps.
Best for
An assistant that improves its own skills
Runs on
Your computer or a server; terminal UI, desktop app
Models
Multiple model providers; MCP, skills and plugins
Price
Free, MIT. Providers and hosted execution can cost extra.
APX vs Hermes Agent →
Sources (3)
04OpenBot
A desktop app for persistent AI teammates with their own instructions and workspaces, stored on your computer. Agents message each other and hand off tasks, with scheduled routines and channels such as Slack, Discord and Telegram.
Choose it if… You prefer its teammate workspaces and its noncommercial license fits your use.
Best for
Persistent AI teammates on your desktop
Runs on
Desktop app for macOS, Windows, Linux; mobile companions
Models
Codex, Claude Code, ACP agents, compatible and local endpoints
Price
Source available under PolyForm Noncommercial; provider costs separate.
APX vs OpenBot →
Sources (3)
05Paperclip
An open-source orchestrator that runs a team of agents as a company, with goals, an org chart, tasks, budgets and approvals. It is self-hosted, drives coding agents such as Claude Code and Codex, and has heartbeats and routines on schedules or triggers.
Choose it if… You want business goals run through an org chart of agents, with budgets and approval gates.
Best for
Running agents as a company with governance
Runs on
Self-hosted web dashboard on your laptop, Docker or server
Models
Claude Code, Codex, Gemini CLI, OpenCode and more
Price
Free, MIT. Agent and model usage can cost extra.
APX vs Paperclip →
Sources (10)
06Claude Cowork
Anthropic's agent for knowledge work: research, analysis, documents and file tasks. By default it runs in an isolated sandbox on Anthropic's servers, with projects, memory, plugins and scheduled tasks; the desktop app can reach your local folders.
Choose it if… You already pay for Claude and want its agent for documents and research.
Best for
Documents, research and files inside Claude
Runs on
Anthropic's cloud; Claude Desktop, web and mobile apps
Models
Anthropic's Claude models; MCP connectors and plugins
Price
Paid Claude plan, from Pro at $20/month; usage limits apply.
APX vs Claude Cowork →
Sources (16)
07ChatGPT dots
One always-on agent per person that takes on ongoing goals and keeps working between conversations, on its own cloud computer and browser run by OpenAI. It starts from your ChatGPT memory and is reachable from ChatGPT, Slack and Teams; it is not offered in the EEA, Switzerland or the UK.
Choose it if… You live in ChatGPT and want one agent that keeps working with your ChatGPT memory.
Best for
One always-on agent for ChatGPT users
Runs on
OpenAI's cloud computer; ChatGPT apps, Slack and Teams
Models
OpenAI models; ChatGPT plugins
Price
Included in ChatGPT Pro (from $100/month) and Business Premium.
APX vs ChatGPT dots →
Sources (5)
08Manus
A hosted general agent that turns a request into a deliverable such as research, slides, a website or a video. It adds projects, skills, scheduled automations, parallel agents and a desktop bridge to folders you authorize, billed in credits.
Choose it if… You want finished deliverables from one request, with nothing to set up.
Best for
Finished reports, slides, sites and videos
Runs on
Manus' cloud; web, desktop and mobile apps, Slack, Telegram
Models
Manus' own model options; MCP connectors
Price
Free with 300 daily credits; Pro from $20/month.
APX vs Manus →
Sources (15)
09Grok Bot
Persistent Bots with names and roles that work through multi-step tasks on a shared cloud computer. They run in parallel, message each other, keep summaries of past work and run routines on schedules or triggers from Slack, GitHub, email and webhooks.
Choose it if… You want hosted Bots with event triggers that keep working while your computer is off.
Best for
Hosted Bots that work while your laptop sleeps
Runs on
A cloud computer in the US; desktop, mobile, X
Models
Picks the model per task; apps and custom MCP
Price
Included with X Premium ($30/month), Super ($100) and Ultra ($200).
APX vs Grok Bot →
Sources (10)
10Viktor
A hosted AI employee for operational work such as reports, dashboards, code and campaigns, shared by a whole workspace in Slack or Teams. It keeps workspace memory, records skills, runs scheduled tasks and asks for approvals.
Choose it if… You want one shared teammate inside your team's chat, hosted for you.
Best for
One shared AI teammate in Slack or Teams
Runs on
Viktor's cloud; Slack, Microsoft Teams, Discord and web
Models
Claude, GPT, Gemini and others; MCP and integrations
Price
Free with up to $100 in credits; Team from $50/month.
APX vs Viktor →
Sources (7)
11Muse
Meta's personal agent for goals, tasks and errands, such as sending emails or booking travel. It runs on a dedicated cloud machine per person, keeps memory you can inspect and edit, and you reach it in the Muse app or WhatsApp.
Choose it if… You want a personal agent for errands that you message in WhatsApp.
Best for
Personal errands and life tasks
Runs on
A cloud VM run by Meta; mobile, web, WhatsApp, Mac
Models
Meta's Muse Spark model; built-in and custom connectors
Price
Free with a weekly limit; Power $20/month; Maximum $100/month.
APX vs Muse →
Sources (8)
12Instinct
A personal assistant that takes actions on everyday tasks, reached by text message or phone call, with a web app and a Mac app. It indexes data from the accounts you connect and keeps a Vault for your materials.
Choose it if… You want one assistant for errands by text or call, with no new interface to learn.
Best for
An assistant you text or call
Runs on
A hosted US service; text, calls, web and Mac
Models
Unnamed third-party language models; Apple, Google, Facebook accounts
Price
Some features are paid; no public price list.
APX vs Instinct →
Sources (5)
13Claude Code
Anthropic's agentic coding tool: it reads your codebase, edits files, runs commands and opens pull requests, from the terminal, VS Code, JetBrains, a desktop app or the browser. It has CLAUDE.md and auto memory, skills, hooks, subagents, MCP, scheduled tasks on your machine or in the cloud, and chat channels for Telegram, Discord and iMessage in research preview. APX can run it as the harness for an APX agent, so the two work together rather than compete.
Choose it if… You code with Claude models and want an agent in your terminal, editor or browser.
Best for
Coding with Claude in your terminal or IDE
Runs on
Terminal, IDEs, desktop, web and mobile; local or cloud
Models
Anthropic's Claude models, also via Bedrock, Google Cloud, Foundry
Price
In Claude Pro ($20/month), Max, Ultra, Team, Enterprise; or API.
Sources (7)
14Codex
OpenAI's coding agent, in the ChatGPT desktop, web and mobile apps, a CLI, an IDE extension and Codex Cloud, which runs tasks in parallel while your computer sleeps. It reads AGENTS.md, has optional memories, skills, plugins, MCP, subagents and scheduled tasks, and its CLI is open source under Apache 2.0. APX can run it as the harness for an APX agent, so the two work together rather than compete.
Choose it if… You already have a ChatGPT plan and want a coding agent locally and in the cloud.
Best for
Coding with OpenAI models, included in ChatGPT
Runs on
ChatGPT apps, CLI, IDE extension; local or cloud
Models
OpenAI models; Responses API-compatible custom providers
Price
In ChatGPT Free, Go, Plus ($20/month) and Pro; or API.
Sources (13)
15Cursor
A code editor and CLI with an agent that works locally, plus cloud agents that run in isolated VMs and can be started from the web, iOS, Slack, GitHub or Linear. It has rules and AGENTS.md, skills, subagents, MCP and automations on schedules or events; Cloud Agents, Automations and third-party models start at Premium.
Choose it if… You want an editor built around agents, with cloud agents for parallel work.
Best for
An AI code editor with cloud agents
Runs on
Editor and CLI on macOS, Windows, Linux; cloud VMs
Models
Grok and Composer included; Claude, GPT, Gemini at API price
Price
Free; Plus $8, Premium $30, Super $100, Ultra $200 a month.
Sources (11)
16Google Antigravity
Google's agentic development platform: a desktop app for orchestrating agents, an IDE, a CLI, extensions for other editors and a Python SDK. Agents run on your machine, which you can drive from a browser with Remote Control; it has rules, skills, MCP, conversation-scoped schedules and, on paid plans, multi-agent teams.
Choose it if… You want a free way to run coding agents with Gemini models on your machine.
Best for
Agent teams for coding, free to start
Runs on
App, IDE and CLI on macOS, Windows, Linux
Models
Gemini, Claude and GPT-OSS models, by plan
Price
Individual $0 with weekly limits; business from $30 per seat.
Sources (11)
17Devin
An autonomous software engineer that plans, codes, tests and opens pull requests, in cloud sessions on Devin's VMs or locally through Devin Desktop and the CLI. It runs parallel agents, keeps skills and shared Spaces, and starts automations on schedules or Slack, GitHub, PagerDuty and webhook events.
Choose it if… You want tickets, alerts and CI failures turned into pull requests.
Best for
Turning tickets and alerts into pull requests
Runs on
Devin's cloud VMs; Desktop IDE, CLI, Slack, Teams
Models
Anthropic, OpenAI, Google, Cognition and open models; MCP, ACP
Price
Free; Pro $20/month; Max $200/month; Teams from $80/month.
APX vs Devin →
Sources (16)

Questions

Which one runs on my own computer?

APX, OpenClaw, Hermes Agent, OpenBot and Paperclip run on your computer or a server you control. Claude Code, Codex, Cursor, Antigravity and Devin's Desktop and CLI work locally too, with cloud models. Claude Cowork, ChatGPT dots, Manus, Grok Bot, Viktor, Muse and Instinct run on their vendor's cloud; some can reach your computer if you allow it.

Source 1 ↗Source 2 ↗Source 3 ↗Source 4 ↗Source 5 ↗Source 6 ↗Source 7 ↗Source 8 ↗Source 9 ↗Source 10 ↗Source 11 ↗Source 12 ↗Source 13 ↗Source 14 ↗Source 15 ↗Source 16 ↗Source 17 ↗

Which one can use my ChatGPT or Claude plan?

Codex is included in ChatGPT plans from Free up, and ChatGPT dots comes with Pro. Claude Code is included from Claude Pro, and Claude Cowork needs a paid Claude plan. APX can use your ChatGPT plan and run Claude Code and Codex as harnesses, so their plans keep working inside APX; OpenBot and Paperclip also drive Claude Code and Codex, and Devin can link a ChatGPT plan.

Source 1 ↗Source 2 ↗Source 3 ↗Source 4 ↗Source 5 ↗Source 6 ↗Source 7 ↗

Which one is free?

APX, OpenClaw, Hermes Agent and Paperclip are free under the MIT license; you pay your model provider separately. OpenBot's license allows noncommercial use. Codex (in ChatGPT Free), Cursor, Antigravity, Devin, Manus, Viktor and Muse have free tiers with limits. Claude Code, Claude Cowork, ChatGPT dots and Grok Bot need a paid plan.

Source 1 ↗Source 2 ↗Source 3 ↗Source 4 ↗Source 5 ↗Source 6 ↗Source 7 ↗Source 8 ↗Source 9 ↗Source 10 ↗Source 11 ↗Source 12 ↗Source 13 ↗Source 14 ↗Source 15 ↗

Which one is best for writing code?

Claude Code, Codex, Cursor, Google Antigravity and Devin are built for code. Pick by the models and plan you already have and where you want the agent: terminal, editor or cloud. In the Artificial Analysis Coding Agent Index read on October 30, 2026, the best Claude Code and Devin configurations scored highest among the apps here, and Codex had the lowest cost per task; Cursor is not in that index.

Source 1 ↗Source 2 ↗Source 3 ↗Source 4 ↗Source 5 ↗Source 6 ↗

Can I use more than one together?

Yes. APX can run Claude Code, Codex and ACP agents as harnesses for its own agents, so a crew can code with the tool you already pay for. Paperclip and OpenBot also drive coding agents, and Devin Desktop can run ACP agents.

Source 1 ↗Source 2 ↗Source 3 ↗Source 4 ↗

Sources

Written by APX from the official pages listed under each app. Every cell is backed by those sources; plans, prices and features change, so check them before you decide. A missing detail is not a missing feature.

Sources checked October 30, 2026

APX's own docs

Coding agent index

OpenClaw

Hermes Agent

OpenBot

Paperclip

Claude Cowork

ChatGPT dots

Manus

Grok Bot

Viktor

Muse

Instinct

Claude Code

Codex

Cursor

Google Antigravity

Devin

Product names and logos belong to their respective owners. No affiliation or endorsement is implied.

Keep comparing

Meet your crewDownload APX v2