Comparison

Humalike vs Inworld.

Inworld goes deep on realtime voice. Humalike sits above the voice stack and handles the social behavior — turn-taking, persona, theory of mind, memory.

Talk to usVisit Inworld

Them

Inworld

Realtime voice AI platform

Realtime TTS-2 (May 2026), Realtime STT with prosody profiling, a Router across 200+ LLMs, plus an Agent Runtime. Originally built for game NPCs; in 2026 a voice-first developer platform.

Us

Humalike

Behavioral infrastructure

Behavioral primitives — turn-taking, norms, persona, memory, theory of mind, social signals — composed above any voice stack and any LLM. Inworld is a clean voice layer to plug in underneath.

The core differences.

No feature-checklist arms race. Just the dimensions that decide which one fits — with the honest call on each.

Realtime voice depth (TTS / STT)

Inworld

TTS-2 + STT with speaker profiling.

Humalike

Plug Inworld in as the voice layer.

Built-in LLM router

Inworld

Router across 200+ models.

Humalike

Bring your own model + router.

Composable behavioral primitives

Inworld

Voice/router primitives, no behavior APIs.

Humalike

7 behavioral APIs you compose.

Multi-human rooms

Inworld

Voice agents are 1:1.

Humalike

Multi-party as a first-class case.

Cross-session relational memory

Inworld

No memory primitive exposed.

Humalike

Memory per person, across sessions.

Theory of mind + social signals

Inworld

Speaker profiling on voice only.

Humalike

Beliefs, intent, typing pauses, edits.

Text-channel agents

Inworld

Voice-first stack.

Humalike

Voice and text.