26 Aug Meet Levama: Making AI Companions Safe by Default
Every mental health platform is being asked the same question right now: where is your AI? The pressure to add it is real and so is the risk of getting it wrong. Keeping a model on script, staying compliant and knowing what to do when someone in real distress opens the chat at two in the morning are not problems you solve in a sprint. That is the work Levama has taken on. Rather than every platform building the same safety layer from scratch, Levama offers a white label AI companion, shaped by clinical research, that a platform can launch under its own brand. As part of Vision Health Pioneers Cohort 8, the team is now bringing that infrastructure to its first partners in Germany, with the rest of Europe next.
Mental health platforms are under real pressure to add AI support, but building it safely is genuinely hard, stopping the AI from going off script, staying compliant, handling someone in real distress. Levama solves the how. A white label AI companion, shaped by real clinical research, that a platform can launch under its own brand in weeks instead of building from scratch.
A health platform building this from scratch typically faces twelve to eighteen months of engineering work and upwards of half a million euros before a safe AI companion ever reaches its first user. We compress that to four to six weeks. We handle the safety protocols, the compliance layer, and the integration. The client configures their own program, uploads their content, and launches under their own brand.
It means the AI cannot go where it should not. Every conversation passes through input and output monitoring before the user sees a response. Guardrails are set per deployment, and the system flags anything outside the defined clinical scope. On the compliance side, we run on GDPR compliant, ISO 27001 certified infrastructure, and we’re designing with the EU AI Act’s requirements in mind from the start. Nothing goes live without a real clinical foundation behind it.
Here is an example of how this works in practice. When a user signals distress, a dedicated crisis agent takes over the conversation. It does not immediately escalate, it first asks deeper questions to assess the situation. If the signal is confirmed, it follows a predefined protocol, signposting to a GP, sharing relevant crisis helplines, and requiring the user to confirm they have seen the information before the conversation continues. The whole protocol is customisable per client and per clinical context, so a meditation app and a depression support programme can each define exactly what the right response looks like for their users.
We’re building toward EU expansion, with our first conversations with potential platforms underway and genuine early interest so far. Warm introductions to digital mental health platforms in Germany open to a pilot would be a real help, and any ideas, suggestions, or feedback are very welcome too.
There is a version of AI in mental health that moves fast and hopes for the best. Levama is deliberately building the other one, where the guardrails come first and nothing ships without a clinical foundation underneath it. That patience is harder to market than a launch date, but it is exactly what this space needs. If you run a digital mental health platform in Germany and a pilot sounds interesting, this is a good moment to get in touch.
Follow their journey here: https://www.levama.ai/de