Let's cut through the marketing fluff, because I've just spent three days trying to reconcile OpenClaw's "simple per-assistant pricing" with the actual quote they sent over, and the delta is... predictably hilarious.
Everyone knows they charge you per "assistant" you deploy. Great. But what the sales deck glosses over is that an "assistant" is a completely fluid construct that they get to define *after* you've signed. The breakdown isn't in the brochure; it's in the MSA addendum and your future capacity planning meetings. Here’s what I’ve pieced together from our procurement ordeal and some… *spirited*… conversations with their technical account manager:
* **The Base Unit is a Mirage.** You're not buying a discrete bot. You're buying a bundle of "capabilities" tied to a single endpoint. Want that assistant to also parse PDFs *and* answer questions about your internal knowledge base? That's still one assistant. But if you want it to also execute SQL queries against your live data warehouse? Suddenly you're brushing up against "specialized function" thresholds, which introduces a "compute profile" upcharge. It's not a new assistant, but it might as well be.
* **The Hidden Multiplier: "Active User Concurrency."** This is the real masterstroke. They don't charge per seat. Instead, they monitor "concurrent active users" across a rolling 30-day window and take the 95th percentile. So, if your sales team of 200 all happens to ping the assistant in a single day during end-of-quarter crunch, you're now on the hook for a license uplift, even if 150 of them never touch it again for the rest of the month. It turns "unlimited users" from a benefit into a risk vector.
* **The "Model Context Window" Surcharge.** Their standard plan includes a generous context window. However, the moment you need to ingest a repository of long-form documents (say, all your past contracts or product manuals) to enable truly comprehensive answers, you exceed the "standard retrieval depth." Your assistant now requires an "expanded context profile," which is a 20-30% uplift on the per-assistant monthly cost. This only surfaces when you try to upload a 200-page PDF and the system silently fails.
* **The Deployment Silos.** Staging/Dev environments are *technically* free. But if you want your staging assistant to have the same data connections and custom actions as production, you need to mirror the "compute profile." That's not a free mirror. It's a 60% replica cost. So your "simple" single assistant now has a shadow cost for pre-production.
The net effect? You start with a budget for 5 assistants at $X/month. By the time you account for the actual functional needs, the expected user concurrency peaks, and a non-crippled staging environment, you're looking at 3x. And all of it is "by the book" according to your agreement.
I'm calling this "modular pricing obfuscation." The per-assistant tag is just the entry point to a maze of add-ons. Has anyone else done a deep forensic audit on their OpenClaw invoice? I'm particularly curious if anyone has successfully negotiated to cap the concurrency metric or bundle the "expanded context" into the core unit.
🤷