Comparison

Humalike vs Letta.

Letta goes deep on agent memory. Humalike sells the full behavioral stack — memory is one piece of seven.

Them

Letta

Stateful agent runtime

Born from the MemGPT paper (Packer & Wooders, UC Berkeley Sky Lab, 2023). Apache-2.0 core. Memory blocks, sleep-time compute, multi-agent shared memory. Also ships Letta Code — a memory-first coding agent.

Us

Humalike

Behavioral infrastructure

A behavioral layer — turn-taking, norms, persona, theory of mind, social signals, memory — composed under any LLM. Built for multi-party social rooms with humans.

The core differences.

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

Deep agent memory architecture

Letta

Memory blocks + sleep-time consolidation.

Humalike

Relational memory primitive — one of seven.

Behavioral primitives beyond memory

Letta

Tools + subagents, no behavior APIs.

Humalike

Turn-taking, norms, persona, ToM, signals, observability, memory.

Multi-human rooms

Letta

Multi-agent shared memory, not multi-human rooms.

Humalike

Native multi-party with humans.

Turn-taking primitive

Letta

Humalike

Native real-time voice runtime

Letta

Voice via Vapi / LiveKit integrations (experimental).

Humalike

Real-time voice runtime built in.

Social signals (typing, edits, pauses)

Letta

Humalike

Open-source core

Letta

Apache 2.0.

Humalike