Humalike

The Turn-taking API gives an agent human-like timing in a live chat. You feed it inbound messages, and it returns a speak or stay-silent decision, then paces the agent's reply out as a sequence of messages instead of one wall of text. It also accepts non-message activity like typing and edits as signals. Timing is the whole game here.

Point the Social Learning API at a chat transcript and it produces a profile of how that group communicates, plus a ready-to-use prompt block. That's how an agent picks up norms, tone and who's who without you writing rules by hand. No hand-tuning. It just learns from the transcript you feed it.

Social Memory is person-centric. It ingests a group conversation and lets you recall facts and ask questions about specific people later, so the agent remembers what matters to each one across separate sessions. Not sessions. People.

The Theory of Mind API predicts how a reply will land and hands back a refined version. It's aimed at the awkward cases, the message that's technically correct but lands wrong in a particular group. Ever sent a joke that died in a group chat? That's the failure mode it targets.

The Social Observability API turns a transcript into a reception report for one of your agents, with evidence cited from the conversation. There's a separate Audit flow for getting a full review of how an agent behaved in a raw transcript. It's evidence, not vibes.

The Persona API builds a grounded, lifelike population from a single prompt, and you can deepen an existing persona or validate a batch against schema and consistency gates. Useful when you need many distinct characters rather than one generic assistant. That's the whole point.

Every API shares a base URL, bearer-token auth, a consistent error envelope and a single credit balance. That single humanlike AI agent API surface keeps integration work small. Because the layer sits above your model, you can swap the underlying model without rebuilding the social behavior.