OpenClaw

OpenClaw is the personal harness for people who want an assistant that lives on their machine and answers where they already talk: Telegram, WhatsApp, Slack and Discord. The trait that decides it is the daemon. openclaw onboard --install-daemon leaves a process running that you own, and it drives hosted or local models rather than other coding agents, so it is a companion, not an orchestrator.

Pick it if you want an assistant you control and are willing to run and patch it yourself. Pick Hermes Agent if you want skills that grow with use, and Claude Code if all you want is a coding agent in a terminal.

The risk is the default. The sandboxing page sets agents.defaults.sandbox.mode to off, so a fresh install runs exec, write and edit on the host with the user's permissions. Turn it on and the same page says it is not a perfect security boundary and that bind mounts bypass it, which is where most people put the repository.

Read the page before the first run, not after. What it does right: with the sandbox on, the network defaults to none and mounts from ~/.ssh, /etc and the Docker socket are refused, so the blast radius is small if you accept the defaults.

The architecture is documented and legible. 1. The gateway process stays on the host; native plugins and MCP tools run in-process with it. 2. Tool execution is routed to a backend chosen from Docker, Podman, SSH or OpenShell, scoped per agent, per session or shared. 3. The browser runs in its own container on a dedicated network. 4. MCP is served and consumed.

No benchmark is published, which is appropriate for a harness whose capability is whichever model it routes to. The observation: a system that lets four different backends execute the same tool call has already survived one model change and will survive the next.

The money question has an unusual answer. The OpenClaw Foundation took stewardship on 14 February 2026 when Peter Steinberger, the creator, joined OpenAI, so the roadmap is no longer one person's calendar. There is no pricing, no paid tier and no cap table, and 388,758 stars is distribution a lab would pay for if it were for sale.

Likely pivot: none; a foundation does not pivot, it drifts or it holds. The dependency is contributor attention, and attention follows the creator, who now works elsewhere. Position: long the project, short the assumption that the founder's departure changes nothing.

The demo is an assistant answering from a phone. Procurement: there is no vendor, so there is no contract, no support line and nobody to answer the questionnaire. Sixty engineers means sixty daemons on sixty laptops with access to messaging accounts, and China's state enterprises and agencies were prohibited from using it in March 2026 on security grounds, which our own security team will find in ten minutes.

There is no headless mode, so nothing runs in CI and nothing is centrally logged. Onboarding is a shell script and a key. Not yet. Revisit if a managed offering with an audit trail appears.

MIT, and the whole thing is readable. MCP servers go in an mcp.servers block with a transport, an enabled switch and a toolFilter that whitelists read_*, which is more control than most clients give me. Providers include Ollama, LM Studio, vLLM and llama.cpp, so it runs on my box with nothing leaving the LAN, and it exposes itself as an MCP server so my other tools can call it.

The npm install needs --allow-scripts=openclaw, which tells me the postinstall does real work and I should read it. I did. A fork would survive the foundation without noticing. That is ownership.