Personal memory · Self-evolving 🔒 Local-first · Your data stays yours

An AI that
keeps learning you

HushClaw remembers your preferences, projects and way of working, reflects on every correction to get better, and runs on your own machine, so what it learns about you stays with you.

Quick Start

Install

$ bash <(curl -fsSL https://raw.githubusercontent.com/CNTWDev/hushclaw/master/install.sh)

Uninstall

Uninstall bash <(curl -fsSL https://raw.githubusercontent.com/CNTWDev/hushclaw/master/uninstall.sh)
Keep data bash <(curl -fsSL https://raw.githubusercontent.com/CNTWDev/hushclaw/master/uninstall.sh) --keep-data
No prompt bash <(curl -fsSL https://raw.githubusercontent.com/CNTWDev/hushclaw/master/uninstall.sh) -y

Our philosophy

Models change. Your context shouldn't.

New models and agents arrive every month, and your work is scattered across them. HushClaw gives your preferences, background and experience one place to accumulate, so every model you use starts from what you've already taught it.

🧠

Knows you, deeply

Long-term personal memory of your preferences, projects, opinions and the way you like work done. One memory carries across every model and agent, and you can see where each recollection came from.

  • Preferences and communication style
  • Projects, facts and decisions
  • Your methods and how they evolve
🔁

Evolves through reflection

After each task HushClaw looks back at what happened: what worked, what you corrected, how you rated the answer. Lessons become memories and skill improvements, every change is shown to you, and memory changes can be undone.

  • Learns from corrections and ratings
  • Refines its own skills over time
  • Personal regression checks keep it honest
🔒

Local-first, private by default

HushClaw runs on your computer and is used through your browser. Chats, memories, tasks and files are stored and encrypted locally. Sensitive actions such as running shell commands ask for your approval first.

  • Encrypted local database (SQLCipher)
  • Your memory is independent of any model vendor
  • Back up and move it whenever you like

How it learns you

A loop that runs quietly with every conversation, and that you can always inspect and correct.

  1. 1

    Remember

    Relevant memories about you are recalled and handed to whichever model is answering.

  2. 2

    Act

    The assistant works with your tools, files and skills, asking before anything sensitive.

  3. 3

    Reflect

    It reviews the outcome, including your corrections and ratings, and extracts what to keep.

  4. 4

    Evolve

    Memories and skills are updated. Each turn shows what changed, so you stay in control.

One workspace for everyday work

Everything runs in a local web app at http://localhost:8765.

💬

Chat

Streaming conversations, document analysis and file handling, with a quiet one-line status while it works.

🔀

Any model

Pick the right model per task through the built-in gateway, or connect Anthropic, OpenAI-compatible, Gemini or Ollama directly.

🗂️

Memories you can audit

Browse preferences, notes, opinions and experience; check their sources and fix anything that's wrong.

🤝

Agents & skills

Give different jobs their own assistant, tools and SKILL.md skills, and let agents collaborate on a task.

📅

Tasks & calendar

Track tasks and schedules, optionally synced with your local or external calendars.

🔌

Connections

Plug in email, the browser and other services when you need them, each with its own authorization.

🚀 Open source on GitHub

Start building a memory that's yours

One command installs HushClaw on your computer. The longer you use it, the better it knows you.