Claude Cowork alternatives: what to pick, by what you're missing
Ownership, model choice, usage limits, or just the coding half — the honest map from Cowork to OpenClaw, Hermes, Claude Code/Codex, and the DIY routes.
The right Claude Cowork alternative depends on what’s actually sending you looking. Want an agent you own — open source, any model, your machine? That’s OpenClaw (or Hermes). Want mainly the coding half? Claude Code and Codex do that deeper, from the same vendors. Chafing at usage limits or the subscription? The open agents let you pay per token or reuse a plan you already have. There is no straight swap that is also closed-source-but-different — Cowork’s real competition is the open-agent world.
First, be fair about what Cowork is
Claude Cowork is Anthropic’s agentic mode — it plans multi-step work across files and tools, runs scheduled tasks, drives the screen, and comes included with paid Claude plans ($20 Pro to $100–200 Max). It’s polished, and if you’re happy inside it, switching costs you defaults that mostly just work. People go looking for alternatives for four concrete reasons: Claude-only models, usage limits that agentic work burns through quickly, sessions that live in Anthropic’s apps and cloud rather than on a machine you control, and policy risk — one vendor deciding what your agent may do (a risk 2026 made concrete — the full comparison covers that story).
The alternatives, by what you’re actually after
- You want ownership → OpenClaw. The most-adopted open-source personal agent (MIT, foundation-governed, ~385k GitHub stars): persistent memory in plain files, 25+ messaging channels so the agent lives in your chat apps rather than a vendor’s, and any model behind it — including a ChatGPT subscription you already pay for. The trade: it runs on a computer someone must operate.
- You want ownership with a different philosophy → Hermes. The other serious open agent, differing from OpenClaw on architecture and defaults more than on mission — we’ve compared the two honestly.
- You mostly used Cowork for code → Claude Code or Codex. The dedicated coding agents go deeper on repositories than a general desk-work agent, and both now have cloud/web modes. If what you miss is “an agent that works while my laptop is closed,” the DIY route is a coding agent on an always-on VPS.
- You wanted a teammate for documents and busywork, minus the subscription math → the open agents again, honestly: that category’s closed alternatives are mostly enterprise workflow suites, which solve a different problem than a personal agent does.
The real decision underneath
Every row above reduces to the same fork: rent a finished agent inside a vendor’s walls, or own an open one and take on (or buy) its operations. Renting is genuinely better for zero-setup desk work in one ecosystem. Owning wins when the agent is meant to be yours — always on, reachable from your messaging apps, remembering everything in files you can read, surviving any vendor’s policy change. If the operations part is what’s stopping you, that piece is buyable without giving up the ownership: a managed OpenClaw pod is the open agent with the sysadmin included, $29/mo with model usage in the box. Either way, start from the reason Cowork stopped fitting — that reason picks your alternative better than any feature table.