If Claude already handles your documents and research, switching tools needs a reason. For this comparison, the reasons worth looking at are access to your own machine, a choice of models, and how you hand work over to a colleague.
Those are separate questions. Another cloud agent might change the model without changing where the work runs.
Disclosure: we build TODO for AI. This article adapts our September 2026 comparison notes; it is not a hands-on test of every tool.
Why people look past Claude Cowork
- Model choice. Cowork is a Claude product. If you want another model to review its output, that is a separate step to arrange, not a model switch inside Cowork.
- Its computer, not yours. Sessions run in Anthropic's isolated environment, not in a shell on your machine or your repo. Claude in Chrome reaches your logged-in browser from the side panel, but the rest of the run happens on their side.
- Connectors, not the tools you have. Reach is its browser plus whatever MCP exposes. The CLI you already use for deploys, ad accounts or your CRM is not a connector, so Cowork cannot just pick it up.
- Personal by design. Team plans can share a Cowork project with view or edit access, but there is no board where a colleague and an agent hold the same task.
Options to compare
| Tool | Best when | Models | Open source | From |
|---|---|---|---|---|
| Microsoft Copilot Cowork | Your company runs on Microsoft 365 | Anthropic, OpenAI (Microsoft picks) | No | $18/mo + usage |
| ChatGPT agent mode | One assistant for research, writing and everyday questions | OpenAI only | No | Free |
| Notion Custom Agents | Your work already lives in Notion | Claude, GPT, Gemini, Grok | No | Free |
| Lindy | Repeatable business processes, no code | Most major models | No | $30/mo |
| Manus | One-off research and slide decks | Vendor-selected | No | $20/mo |
| OpenClaw | A self-hosted agent in your chat apps | Any (your keys) | Yes, MIT | Free |
| Hermes Agent | A self-hosted agent that learns you over time | Any (your keys) | Yes, MIT | Free |
| TODO for AI | Several models, your own machines, a shared board | Multiple providers on paid plans | Partly | Free |
Prices are entry tiers. Agent features may require a paid plan; self-hosting can still involve model and hosting costs.
Copilot Cowork is the closest like-for-like for companies on Microsoft 365. It works inside Outlook, Teams, Word, Excel and SharePoint with your existing permissions, keeps running with the laptop closed, and has the admin controls IT expects. Outside Microsoft 365 its reach is thin, and it bills a Copilot licence plus metered credits.
ChatGPT combines chat, deep research and agent mode. It is OpenAI-only, agent mode runs in OpenAI's sandbox, and its browser is not your logged-in Chrome profile.
Notion Custom Agents let you pick the model (Claude, GPT, Gemini, Grok) and run on schedules and triggers inside your workspace. There is no shell anywhere, and they cannot act in your logged-in browser.
Lindy takes a different approach: build the triggers and steps for a repeatable process. That is useful if repetition, rather than one-off delegation, is the problem.
Manus needs no setup and is good at research and documents. You do not choose the model, and credit pricing is hard to predict.
OpenClaw and Hermes Agent are the open-source, self-hosted route: any model through your own keys, reachable from chat apps like WhatsApp, Telegram and Discord. You run and secure them yourself.
TODO for AI is a shared task board where agents close the items. On paid plans, Claude, GPT and Gemini can work the same task and review each other. It runs on your own PC, an always-on cloud VM and your real Chrome profile, and can use installed CLI tools, subject to authentication and permissions. Skip it if Anthropic models are all you need, or if you already pay for Claude and want no second tool.
Which difference matters?
- Documents, research and browsing, and Claude is enough: stay on Claude Cowork.
- Your company lives in Microsoft 365: Copilot Cowork.
- Want to self-host with your own keys: OpenClaw or Hermes Agent.
- Several models checking each other, your own machines, a board your team shares: TODO for AI.
Full scored comparison with 21 tools and the scoring method: Claude Cowork alternatives on todofor.ai.
Top comments (0)