What Ojin Means by "The Human AI Company"

Direct answer: The Human AI Company is Ojin's description of what it builds: real-time AI agents with a face and a voice rather than a chat window. Its flagship, Human Agents, bundles speech recognition, reasoning, a face model and speech synthesis into one deployable agent. The bet is that presence, as much as correctness, decides whether an interaction lands.

Most AI companies describe what their models can compute. We describe what the result should feel like: human. Calling ourselves the Human AI Company rests on that single claim, which is worth explaining rather than leaving as a tagline.

What is a Human AI company?

A human AI company is a positioning choice about where a company puts its effort. No standard or certification defines it. Most people typing "what is a human AI company" into a search bar want to know what separates it from a conversational AI vendor or a generic model provider, and the difference is architectural, not linguistic. We put a face model and a voice at the centre of our pipeline from the start, instead of treating them as a skin applied to a text-based chatbot after the fact. That choice is not cosmetic. In a 2023 study in Frontiers in Robotics and AI, participants asked to choose between a robot with a face and gaze and a voice-only smart speaker picked the embodied agent as the one they would trust in an emergency, 54% to 23%.

We frame this as building next generation AI, but deliberately mean something narrower than "smarter models." Next-gen AI, in the abstract, usually points at benchmark scores and parameter counts. What we're betting on is closer to interface and interaction: an agent judged by whether someone wants to keep talking to it, not only by whether its answers check out. Both matter. Our argument is that only one of them has been getting the attention it deserves.

Why does presence matter more than correctness?

Our founding bet is that presence, more than correctness, is what makes an AI interaction feel trustworthy. An exchange lands less because the model is right than because it feels like a conversation with someone present rather than a form with a typing indicator. A correct answer delivered in a chat window and the same correct answer delivered by a face that holds your gaze and answers without a delay are not the same experience, even though the words are identical.

"Human AI" makes a narrower claim than "generative AI" or "conversational AI" do. Generative AI describes a capability. Human AI describes an experience: a system built to be present, which is a higher bar than being responsive.

A 2026 study in Scientific Reports, which manipulated both the warmth and the competence of human and AI counterparts in a trust game with 400 participants, found that AI agents were trusted less than people were, and that the gap was widest in the low-warmth conditions. Competence moved trust too, for humans and AI alike. What the study isolates is a warmth penalty that lands specifically on AI. Correctness sits on one axis. Warmth, the sense that something present is responding to you, sits on another, and the second one is where AI has the most ground to make up. A model can be state of the art and still lose the room if nothing about the interaction feels present.

Embodiment is one of the levers that moves warmth. A 2025 meta-analysis in Humanities and Social Sciences Communications, pooling 800 effect sizes across 142 papers, found that human-like social cues added to text-based agents lifted user perceptions only modestly and behaviour barely at all. Text has a low ceiling, which is why we did not start from text.

Why "Human AI Company" instead of just "AI company"?

The label was not the easy path. "AI company" is legible to almost anyone on first read; it says roughly what a business does without requiring an explanation. "Human AI Company" requires one, and we decided that cost was worth paying.

The reasoning traces back to our product thesis rather than to branding. If our product were a text-based assistant, "AI company" would describe it fully, because there would be nothing present in the interaction beyond text arriving on a screen. Our product is different in kind, not only in degree. Human Agents puts a face and a voice into the interaction, and that changes what our own name needs to communicate. A generic "AI company" label would flatten a face model and a voice model into the same category as a text generator, when our entire argument is that they are not the same category. Naming "human" at the level of the company description is a way of refusing that flattening in the name itself, rather than leaving it to a product page to make the distinction.

There is a second, more practical reason. "AI company" has thinned out through sheer repetition. Stanford HAI's 2026 AI Index puts organisational adoption at 88%, so a label built on using AI now describes most of the economy rather than anything specific to us. "Human AI Company" narrows the claim to something checkable: either our agents have a face and voice that behave like presence, or they do not. That specificity is deliberate. It gives anyone evaluating us something concrete to test rather than a category to take on faith.

None of this makes "human" a metaphor. It works more like a spec. We describe ourselves this way because the product is built this way, not the other way around.

How this differs from how most AI companies describe themselves

Look at how AI companies typically introduce themselves and a pattern shows up quickly. The opening line is usually about the model: parameter count, benchmark placement, reasoning capability, context window, response speed. These are real, measurable properties, and they matter for plenty of use cases. They describe the engine, though, not the experience of sitting across from it.

Our opening line is about none of them. It is about what happens on the other side of the interaction: whether the person talking to a Human Agent feels like they are talking with someone present, rather than typing into a system that eventually replies. Ours is a different question from "how capable is the model," and it is the one we chose to lead with.

This is not a claim that model quality stops mattering to us. Speech recognition still has to be accurate. Reasoning still has to hold up. Speed is a real property too, and we publish ours: Oris Portrait renders under 200ms of face-model latency, our own figure for the face model alone rather than for the full exchange from speech in to speech out. What differs is where we place our headline claim. A company that leads with benchmarks is telling a prospective user the model is good. A company that leads with presence is telling them the experience will feel different, and betting that the difference is what a person notices first, before they get far enough in to judge the reasoning underneath.

That bet only makes sense if the underlying claim is testable rather than decorative. It is why we point people towards talking to a live agent rather than reading a spec sheet. A face-and-voice claim either holds up in the room or it does not, and no benchmark score substitutes for that test.

Who founded Ojin?

Our founder and CEO, Christian Mio Loclair, is an artist and computer scientist trained in both code and dance. We are a Berlin company. Loclair trained in computer science, specialising in human-computer interaction, alongside a parallel career as a choreographer studying how people read each other's movement and expression. That combination led him to build AI systems meant to be read the same way. His work on AI and human identity has been exhibited at the Centre Pompidou, ZKM, and Ars Electronica, including ZKM Karlsruhe's BioMedia exhibition, which ran from 18 December 2021 to 28 August 2022.

We have spent years building this technology and this team. The current chapter builds directly on that work rather than starting over, and it is not a different company wearing someone else's history.

What is Ojin's flagship product?

Our core deployable product is Human Agents: a bundled system combining speech recognition, a reasoning layer, a face model, and speech synthesis into a single agent a business can embed on its own site. We build and run every stage of that pipeline, from speech in to speech out. For teams that already have a conversational pipeline they trust and only want the visual layer, the underlying face models are also available separately through the real-time agent API, dropped into an existing stack rather than replacing it. Both routes are listed on our pricing page.

Under that flagship sit two developer-tier face models: Oris Portrait, built for speed and scale, and Oris Presence, built for the most sophisticated rendering our platform offers, aimed at being the first to clear the uncanny valley rather than skirt around it. The valley is Masahiro Mori's 1970 observation. In the translation published by IEEE Spectrum in 2012, Mori writes: "I have noticed that, in climbing toward the goal of making robots appear human, our affinity for them increases until we come to a valley, which I call the uncanny valley." Affinity climbs until a figure looks almost but not quite human, then drops into unease. Still images have largely climbed out: a 2022 PNAS study by Sophie Nightingale and Hany Farid found synthesised faces indistinguishable from real photographs and rated slightly more trustworthy. The harder case, and Oris Presence's, is a face that moves, speaks and responds in real time. Both are face models; neither is the orchestration layer. Human Agents is the product that ties speech, reasoning, and face together into something you deploy.

What does "human" not mean at Ojin?

We are not asserting our agents are secretly human, and we are not built around deception. Every visitor is told plainly that they are talking to an AI agent, and that disclosure is a design default rather than a compliance afterthought. Article 50 of the EU AI Act requires systems that interact directly with people to tell them they are dealing with an AI. The text requires that people are "informed that they are interacting with an AI system, unless this is obvious from the point of view of a natural person who is reasonably well-informed, observant and circumspect." The European Commission puts the application date at 2 August 2026. We treat that, and GDPR, as baseline architecture decisions, consistent with being a German company operating under German and EU law from day one.

Two other operating principles follow the same pattern. Modularity means any component in the pipeline can be swapped rather than locked to one vendor. Speed to production means shipping instead of running long proof-of-concept cycles.

Where does the Human AI Company framing show up in the product?

The Human AI Company framing runs through everything we build and every audience we serve, rather than a single product page. Developers working through our insights on real-time AI agents meet the same face-and-voice presence layer behind every entry point. Enterprises deploying a human AI agent for support, sales, or onboarding are deploying the customer-facing expression of it. Teams weighing cloud hosting or the self-serve route are choosing how they access the same underlying bet: that presence is the product itself. Every one of those deployments makes the same spoken AI disclosure by default, ahead of the 2 August 2026 application date the European Commission gives for Article 50 of the EU AI Act.

Human Agents is where that reasoning becomes something a business deploys directly: speech recognition, reasoning, a face model and speech synthesis in one agent, with either Oris Portrait or Oris Presence behind the face, rather than separate parts a customer's engineering team has to assemble.

It also shows up somewhere less obvious: Ojin AixHaus, our community space in Berlin, exists because a company claiming to build for human presence should be reachable in a room, not only through documentation.

We also describe ourselves, depending on the audience, as a conversational AI company, and the same underlying agent gets deployed as an AI sales agent, an AI virtual agent, or a human AI avatar depending on what a business needs it to do. The name changes with the job. The system underneath does not.

Why there are two face models instead of one

Inside our product line, the same thesis explains a decision that could otherwise look like ordinary product segmentation. Splitting Oris Portrait from Oris Presence goes beyond the speed-versus-quality trade-off most software companies make with a lite tier and a pro tier. It reflects a belief that different deployment contexts have different presence requirements, and that flattening them into a single model would force every customer to either pay for headroom they do not need or accept a ceiling they cannot live with.

Portrait for volume

A support agent fielding the same handful of questions thousands of times a day needs presence that holds up at volume without slowing the pipeline down. Oris Portrait is built for that case, rendering under 200ms of face-model latency, a figure that covers the face model and nothing else in the pipeline.

Presence for close scrutiny

A flagship brand experience, meant to be the first impression a company makes, needs presence that holds up under close, repeated scrutiny instead, which is what Oris Presence is built for. Treating those as identical problems would have been the simpler engineering decision. Treating them as two separate products is the one consistent with meaning the "human" part of the name, rather than shipping one face model and calling the thesis satisfied.

Frequently asked questions

Is Ojin a Human AI Agent company?

Yes. Ojin builds and deploys Human Agents, real-time AI agents with a face and a voice rather than a chat window, which is what makes it a Human AI Agent company rather than a general conversational AI vendor.

Where is Ojin based?

Berlin, Germany, operating under German and EU law, including GDPR and the EU AI Act's transparency requirements by default.

Is "the Human AI Company" just marketing?

The name describes a specific architecture choice, not a slogan layered on afterwards: a face model, a voice model, and reasoning built together from the start, rather than a chat product with a face added later. Whether that claim holds up is not something to take on faith from a page like this one. The direct way to check it is to talk to a Human Agent and judge for yourself whether presence is there, rather than read a description of it.

Does Ojin build AGI or sentient AI?

No. Ojin does not build artificial general intelligence, and makes no claim that its agents are sentient, conscious, or human in any literal sense. Human Agents are real-time systems that combine speech recognition, reasoning, a face model, and speech synthesis to produce an interaction that feels present. Feeling present is a design goal for how an interaction lands, not a claim about what is happening inside the system. Every visitor is told plainly that they are talking to an AI agent, and that disclosure holds regardless of how convincing the face and voice are.

How to test the presence claim yourself

The fastest way to understand what "Human AI" means in practice is to talk to one of the agents. Try talktolou.ojin.ai, or build your own agent at ojin.ai/signin using docs.ojin.ai to guide you.