ai-agent-

Henry is a personal AI agent framework that runs on the user's computer. It gives one agent memory, local knowledge, tools, scheduled work, and safe ways to act. The agent can be used from the terminal, a private web dashboard, or Telegram. The framework is public, while identity, files, credentials, memory, and knowledge stay in the private setup.

The agent is intended to remember decisions, understand the owner's work, use existing tools, and handle repeatable jobs. It can remember durable preferences and decisions through Engram, search a private knowledge base with local embeddings and source references, use connected tools through Claude or Codex, run scheduled workflows and reminders, coordinate bounded specialist agents through Luna, support job research and application tracking, read Gmail and draft replies, review pull requests in separate passes, and work through terminal, dashboard, and Telegram. Modules are optional, and a fresh setup should start with the owner's problem before enabling only the workflows that solve it.

Reading and local work can happen directly. Email, posts, applications, comments, and other external actions are staged first, and approval and execution are separate steps. The model can change, while memory, permissions, workflows, and approval history remain under the owner's control. The public repository does not contain the owner's memory, knowledge corpus, resume, credentials, browser profile, or runtime databases. The project is licensed under MIT.