Plastic Labs

Plastic Labs is an AI research company focused on building Honcho, an AI-native memory layer that models personal identity for AI applications. Honcho functions as a plug-and-play platform that gives LLM-based apps persistent, evolving user profiles, enabling them to understand individual communication styles, preferences, and behavioral patterns over time.

The core market problem Honcho addresses is that LLM applications lack true user understanding. Developers typically resort to ad-hoc solutions like storing chat logs in vector databases and doing retrieval-augmented generation, which can only recall past conversations but cannot grasp deeper user traits such as learning styles or tone sensitivity. This forces every team to reinvent user-state management infrastructure. Honcho abstracts away this complexity, offering rich, queryable user models that any LLM app can integrate easily.

Its competitive edge lies in applying cognitive science techniques to build higher-order, transferable user representations. Unlike app-specific data silos, Honcho captures universal traits that can be shared across different applications. This interoperability creates a foundation for a shared user data layer, where insights from one app (e.g., a tutoring app discovering a user learns best via analogies) can benefit another (e.g., a therapy assistant communicating more effectively).

In the past six months (March-August 2026), Plastic Labs has made significant progress. In March, three papers with Plastic Labs authors were accepted at ICLR 2026, covering adaptive decoding, uncertainty extraction, and self-play reasoning. In April, the company launched the Honcho CLI and a refreshed MCP server, enabling developers to manage peers, sessions, and memory from the terminal or directly within AI assistants like Claude. In May, Honcho's /chat endpoint added structured output support with response_format flags. In August, Honcho 3.1.0 was released, introducing Scopes (session grouping for context-specific queries) and workspace-level chat, along with performance improvements including a smaller production image and optimized Redis Cluster support.