NEW YORK, NY — September 2, 2026
Plastic Labs today released @honcho-ai/dsh-honcho, a plugin that gives agents running in the DeepSeek Harness a persistent model of the person they work with. The plugin is MIT-licensed, installs with one command, and connects to a hosted Honcho account or a self-hosted Honcho instance. The DeepSeek Harness is DeepSeek's open-source agent harness, built on an everything-is-a-plugin architecture and currently in developer preview; it has drawn more than 220,000 GitHub stars since its August debut (GitHub, September 2026).
At session start, the plugin injects what Honcho already knows: a profile of the user, a summary of the current project, and the conclusions relevant to the task at hand. As the session runs it captures user and assistant messages in the background, redacting secrets before anything leaves the machine, and exposes three tools the agent can call directly—honcho_search, honcho_chat, and honcho_remember. Typing /honcho shows workspace status and a link to the dashboard. Memory scope is configurable per directory, per repo, per git branch, per session, or globally, and the injection budget defaults to 1,500 tokens.
The point isn't that the harness remembers the last session. Most memory plugins do that, and it's table stakes. Instead, Honcho builds a model of the user that allows it to better infer the user's intent and more efficiently complete tasks. It finds the 1,000 tokens that matter, rather than the millions available in the context. This means that the same account that already personalizes Claude Code, Codex, OpenClaw, OpenCode, or Hermes Agent can personalize a DeepSeek agent from its first message, and feed back what it learns in DeepSeek to the others.
Now developers switching to the DeepSeek Harness don't need to start from 0 and can make sure they have the best context possible to drive their agents. – Vineeth Voruganti, CTO, Plastic Labs
The plugin is live on npm and documented at honcho.dev/docs. It requires Node 22.19 or later and a DeepSeek Harness build compatible with 0.1.2-alpha.3; the harness is a developer preview, so plugin interfaces can still move. Pricing follows the Honcho account behind the plugin; the plugin itself adds nothing.
dsh plugin --profile <name> add @honcho-ai/dsh-honcho
About Plastic Labs
Plastic Labs is the New York team behind Honcho, a memory and continual-learning layer for stateful agents. Honcho builds a model of the people an agent works with—their preferences, patterns, and how they like things done—and carries it across every harness they run.
Learn more about Plastic Labs: https://plasticlabs.ai Learn more about Honcho: https://honcho.dev