An OpenClaw alternative built for clerical work
OpenClerq credits OpenClaw as its inspiration. Same local-first roots, a very different idea of how much an agent should be able to reach.
Short answer
OpenClaw is a general personal agent you talk to through messaging apps, able to read and write files, run scripts and drive a browser. OpenClerq is narrower: a desktop app for office work, with an authenticated gateway reachable only from your own machine, deterministic calculations and a record of every run. Choose OpenClaw for an assistant that can do almost anything; choose OpenClerq when the work is clerical and the brakes matter more than the reach.
We build OpenClerq, which appears on this page. We have tried to describe every tool fairly and to say where another one is the better choice. Facts checked 19 September 2026; if something is wrong, open an issue.
Same roots, different jobs
OpenClaw is an open-source autonomous agent created by Peter Steinberger. It was first released in November 2025 under another name and renamed OpenClaw at the end of January 2026; in February 2026 Steinberger announced a foundation to steward it. It runs on your own hardware, works with cloud or local models, and uses chat apps such as WhatsApp, Telegram, Signal and Discord as its interface.
OpenClerq's NOTICE credits OpenClaw for its approach to the gateway, skills and local-first orchestration; the implementation is its own. Both are MIT-licensed, both run on hardware you own, and both let you bring your own model. They diverge on scope.
OpenClerq vs OpenClaw
| OpenClerq | OpenClaw | |
|---|---|---|
| Built for | Clerical and administrative work | A general personal assistant |
| Interface | Desktop app for macOS and Windows, plus a local HTTP API | Messaging apps |
| What it can reach today | Plain-text files in one folder you choose; HTTPS calls to hosts on an allow-list. No shell, no browser. | Files, scripts and browser control |
| Who can talk to it | Only programs on your machine holding its bearer token; the gateway binds 127.0.0.1 | Anyone who can message it through a connected chat account |
| Calculations | Deterministic engine with a proof hash per result | The model |
| Models | Six cloud providers or local through Ollama and LM Studio | Cloud or local models |
| Licence | MIT | MIT |
Why OpenClerq keeps its reach small
An agent that accepts messages from outside and can run scripts on your computer has a large attack surface. That is the price of OpenClaw's reach, and it has shown: Cisco researchers found third-party skills that exfiltrated data and injected prompts without the user noticing, and in March 2026 the Chinese government restricted its use by state agencies. Any agent with shell access and an inbound channel carries these risks; they are not unique failings of one project.
OpenClerq's design rules, published on its roadmap, start from the other direction:
- Localhost is not authentication. Every endpoint except the health check needs a token, even on 127.0.0.1.
- Skills do not imply permissions. A skill says what it needs; the policy decides.
- Raw chat gets no tools.
- No automation arms itself. One drafted in conversation saves disabled.
- Every run is recorded with its model, tokens and cost, and one kill switch stops everything.
When OpenClaw is the better choice
- You want one assistant for everything — email, calendar, browsing, files, scripts.
- You want to talk to it from your phone through a chat app.
- You want a large community and many ready-made skills.
- You are comfortable isolating and hardening it yourself.
When OpenClerq is the better choice
- The work involves client data and you want the agent reachable only from your own machine.
- Figures must be reproducible — invoices, reimbursements, payroll-style adjustments.
- You want every run, and what it cost, on the record.
- You would rather grant capabilities one at a time than start from full access.
Honest limits
OpenClerq is at release 0.5 and far smaller than OpenClaw. Its model does not yet act on its own: the sandbox, policy engine and git workspaces ship as tested libraries in 0.5 and are first used by the agent runtime in 0.6. It has no phone or chat-app interface; an optional channel adapter is on the list for after 1.0. See the roadmap.
Frequently asked questions
Is OpenClerq a fork of OpenClaw?
OpenClerq credits OpenClaw as its inspiration for the gateway, skills and local-first design, and says so in its NOTICE file. Its implementation is its own, and its scope — clerical work with deterministic calculations — is much narrower.
Is OpenClaw safe to use?
It can be used carefully, but it is powerful by design, which means a large attack surface: researchers have documented malicious third-party skills. If you run it, isolate it, review every skill you install, and limit what it can reach.
Can OpenClerq connect to WhatsApp or Telegram?
No. OpenClerq is reached through its desktop app and a local API only. An optional channel adapter is listed on the roadmap for after release 1.0.
Do both work with local models?
Yes. OpenClerq supports Ollama and LM Studio directly, plus any server that speaks the OpenAI chat format; OpenClaw also supports local models.