Honcho is an 🏷️#open-source memory library from Plastic Labs for building stateful agents, used in 📝Hermes Agent as a catalog memory provider that models the user across sessions.
Plastic Labs describes Honcho as memory infrastructure for agents that understand changing people, agents, groups, projects and ideas over time. Applications store messages and events on sessions; Honcho reasons over them in the background and updates a representation of each peer, which agents then query for context, search results or natural-language answers. It runs managed at api.honcho.dev, as a local stack through its CLI, or self-hosted as a FastAPI server, and its repository is licensed under AGPL-3.0.
In Hermes, Honcho is a catalog 📝memory provider plugin maintained by Plastic Labs and installed with hermes plugins install honcho. Homes upgraded from a Hermes release that bundled Honcho get the plugin installed automatically, with config and memory carried over. The plugin adds five tools, covering the user's peer card, 📝semantic search, session context, synthesized reasoning and stored conclusions.
Hermes injects Honcho context in two layers: a base layer of session summary, user representation and peer card, refreshed at most every contextCadence turns (default 1), and a dialectic supplement of LLM reasoning about the user, refreshed at most every dialecticCadence turns (default 2). Each Hermes profile gets its own AI peer while sharing one user peer within a workspace, so each profile builds an independent representation of the same user.