Why How We Run Changed: Dedicated Compute and Sustainable AI
September 2026 · by the Merciful team
When Merciful first launched, much of our processing relied on distributed community compute and crowdsourced backends. It was a wonderful experiment in decentralized access, and it allowed us to get off the ground with virtually zero overhead.
But as our community grew into hundreds of daily active users relying on Merciful for actual work, study, and creative writing, the limits of crowdsourced compute became painful: random dropped jobs, wild queue latency spikes, and nodes disappearing mid-response.
To deliver the speed and reliability our users expect, we made a fundamental shift: we now operate all our own unified text runners on dedicated cloud GPU clusters.
The Economic Reality of Dedicated Compute
Operating dedicated GPU clusters running state-of-the-art large language models means consistent, sub-second time-to-first-token. But GPU compute hours must be paid for in real currency.
Many AI products respond to this reality by sliding down the path of "enshittification":
- Cutting off conversations mid-sentence unless you enter credit card information.
- Cluttering chat streams with sponsored links or disguised advertisements.
- Harvesting and selling conversation logs to data brokers for behavioral profiling.
- Trapping users in confusing token micro-transaction schemes.
We built Merciful specifically to be an alternative to that playbook. We want to be "one of the good ones"—a straightforward, honest tool that respects your intelligence.
Our Compact With You: How We Stay Sustainable
Here is how we balance real infrastructure costs with our core commitment to open, accessible AI:
1. The Economy Safety Net (Never Cut Off)
Running out of daily tokens on our Advanced or Standard models does not lock your account or end your session. Instead, your room transitions smoothly to our lightweight Economy model. It continues responding for free, so you are never stranded in the middle of a thought.
2. Daily Grants With Meaningful Rollover
We don't punish you for having quiet days. Free verified accounts receive 50,000 tokens each day and can accrue up to 150,000 tokens (three full days of allowance). Pro subscribers can bank up to 1,500,000 tokens.
3. Pro Subsidizes Free
Our $5/month Pro tier exists for power users who want continuous, unrestricted access to the most capable models. Because our team is small and independent, revenue from Pro doesn't go toward marketing bloat or executive bonuses—it directly pays the GPU hosting bills that keep the free tier online for everyone.
4. No Surveillance or Data Brokering
We store your conversation history so you can reopen your rooms across sessions, not to sell marketing insights to third parties. We treat your prompts as your own.
Building for the Long Haul
We believe that small, independent, and reliable software is still possible on the modern web. You don't need manipulative psychological hooks or predatory subscription walls to build something people love. You just need to build a tool that works, be transparent when things change, and listen closely to the people who use it.
To see the full breakdown of new features in this rollout, read our companion post: What's New in Merciful: Tiers, ETAs, and Themes. Or to read about the automated engineering loop we dogfood behind the scenes, check out Meet Vesa: Inside the Agentic Loop Behind Merciful.
— The Merciful Team