London-based startup Scram is launching a free vibe-coding platform using China's DeepSeek models to eliminate the cost of AI prompts and charge users only when they publish an application. The platform will go head-to-head with other vibe-coding startups such as Lovable, which tend to rely on more underlying proprietary (and more expensive) models from frontier providers such as OpenAI and Anthropic.
Founder and CEO George Bevis, a serial entrepreneur who previously co-founded the Tide bank, told Pathfounders in an exclusive interview that he has personally invested $5 million in developing the platform over four years, betting that removing upfront development costs will attract enough users to sustain a subscription business.
Scram's new Free Mode allows users to build, test and refine applications without purchasing AI credits. Scram is betting that this will act as a lure to new users put off by the current offerings. Once they publish an app it will, however, cost them $9 a month.
The move challenges the credit-based pricing models of AI coding platforms such as Replit and Lovable, where developers can exhaust their free allowances before producing a working application.
"Nobody's got close to doing genuinely free vibe coding up until now," Bevis told me. He argues that taking an application to production quality can cost tens or hundreds of dollars in AI credits, discouraging experimentation and excluding potential users.
The breakthrough, according to Bevis, is DeepSeek's combination of capabilities and low operating costs.
"It does cost us money, right? There's no question," he said. But Scram expects enough users to convert to its $9 monthly publishing subscription to make the model viable.
Plus, the dats will not be going back to China via DeepSeek’s servers. The service uses DeepSeek's LLM on Cloudflare's servers in the US.
Users concerned about where their data is processed can instead select models from Anthropic or OpenAI, but (as with Replit et al) must pay for their use.
Scram also claims a technical distinction from conventional AI coding tools. Rather than allowing AI to generate application code directly, its system uses AI to produce instructions that Scram's own software then executes deterministically.
"If the instructions say make a blue box, then Scram will always make exactly the same blue box," Bevis explained.
He argues this makes applications more reliable and potentially cheaper to generate, although he acknowledged the company has not yet proved the latter claim.
The platform combines conversational prompting with a visual editor, allowing users to inspect and modify interfaces, backends, workflows and integrations, or add custom code.
Bevis acknowledged that competitors could eventually copy the free pricing model. "In the long run, nothing," he said when I asked what would stop them. However, he suggested rivals' existing relationships with AI providers such as Anthropic and OpenAI might slow their adoption of cheaper alternatives: “It'll be interesting to see how quickly they respond because they, I think, had big deals with Anthropic and OpenAI, where you know those companies expect them to use that, and I think they may be sluggish to move to an alternative provider.”
While Scram’s Free Mode initially targets individual users, Scram has broader ambitions in enterprise software. The platform remains in beta and lacks features such as multi-user access, which Bevis says are necessary for serious organisational use. These are planned over the coming months, with mobile app development capabilities expected next year.
Bevis also sees significant demand from emerging markets, where AI token costs can be prohibitive. He said children at a school he supports in Africa have already been testing Scram but have been constrained by the cost of AI usage.
"I actually think there'll be more demand from there," he said, suggesting developing markets could prove more important to Scram than the US.
Scram is targeting a launch of Free Mode later this week, subject to final development work.
For Bevis, the broader bet is that AI app-building will inevitably become free at the point of creation, shifting the business model towards charging users only when they have something worth deploying.
Bevis is joined by Tom Godber, co-founder and former CTO of Masabi (mobile ticketing and payments leader serving large public transport operators globally). Scram is now a 15-person team located in London and Bristol.