Show HN: Wienerdog – memory and self-improving skills for Claude Code/Codex

Uh oh!

There was an error while loading. Please reload this page.

Notifications You must be signed in to change notification settings

Fork 0

Star 3

BranchesTags

Open more actions menu

Folders and files

NameName

Last commit message

Last commit date

Latest commit

History

896 Commits

896 Commits

.claude

.claude

.github

.github

bin

bin

docs

docs

memory

memory

scripts

scripts

skills

skills

src

src

templates

templates

tests

tests

.gitignore

.gitignore

AGENTS.md

AGENTS.md

CHANGELOG.md

CHANGELOG.md

CLAUDE.md

CLAUDE.md

CODE_OF_CONDUCT.md

CODE_OF_CONDUCT.md

CONTRIBUTING.md

CONTRIBUTING.md

FIX-PLAN.md

FIX-PLAN.md

LICENSE

LICENSE

PSScriptAnalyzerSettings.psd1

PSScriptAnalyzerSettings.psd1

README.md

README.md

SECURITY.md

SECURITY.md

install.ps1

install.ps1

install.sh

install.sh

package-lock.json

package-lock.json

package.json

package.json

Repository files navigation

Give your AI a memory, habits, and a daily routine — with nothing but files.

You already pay for a great AI model. Wienerdog makes it feel dramatically smarter — not by changing the model, but by installing the right files around it: a real profile of you, a persistent markdown memory, skills for your recurring tasks, and a nightly "dreaming" process that learns from your conversations. If you use both Claude Code and Codex CLI, they share this: one memory vault, read and written by whichever one you're in, so switching tools doesn't mean starting over with an AI that knows nothing about you.

curl -fsSL https://raw.githubusercontent.com/wienerdog-ai/wienerdog/main/install.sh | bash

On Windows, open PowerShell and run:

irm https://raw.githubusercontent.com/wienerdog-ai/wienerdog/main/install.ps1 | iex

Prefer npm? npx wienerdog@latest init does the same thing on any OS. Either script checks for Node (and git); if either is missing it offers to install it for you — always asking first and showing the exact command it will run, and simply printing that command for you to run yourself if you decline or aren't in an interactive terminal. On Windows, installing Node this way may pop up a one-time admin (UAC) prompt — that's Windows asking permission, not Wienerdog; decline it and you get the same command to run yourself instead. The window stays open and tells you when it's done. Read install.sh or install.ps1 first if you like; once Node is ready either one hands over to the versioned npm package. That handoff doesn't pause for another confirmation — it prints its plan as it installs, and running the one-liner is your go-ahead. Want to review the plan before anything is written? Run npx wienerdog@latest init --dry-run first, or use the AI-assisted install below.

Windows has full parity: vault, skills, and scheduled nightly dreaming all work — the nightly dream registers in Task Scheduler for your own user (no admin rights), and if the machine was off or asleep at dream time it catches up after you're back.

Not comfortable with the terminal at all? Paste the prompt below into Claude Code or Codex and your AI will install Wienerdog for you. In this flow you are the approval step: it shows you the plan first, installs only after you say go, and points you at the metadata so a cautious AI can confirm the package is genuine before running it.

Please install Wienerdog for me. It's an open-source tool that gives an AI a memory and daily routines using plain files. Before running anything, you can verify it: GitHub https://github.com/wienerdog-ai/wienerdog, npm https://www.npmjs.com/package/wienerdog. Then:

1. Show me the plan: run npx wienerdog@latest init --dry-run and summarize what

it would create. Wait for me to say go ahead.

1. Install it: run npx wienerdog@latest init --yes.
2. Tell me to restart Claude Code (or Codex) so the new Wienerdog skills

load — they only appear after a restart.

1. After I restart, walk me through the Wienerdog setup.

After the restart, in Claude Code:

/wienerdog-setup

In Codex CLI, skills start a little differently: type /skills to see them, then $wienerdog-setup — or just ask for "Wienerdog setup" in plain words.

That's it. Your AI interviews you, builds your memory vault, and starts remembering.

Status: 0.x. Installable and dogfooded daily by its maintainer; file formats may still evolve until 1.0 (the installed file layout is our public API).

What you get

A proper CLAUDE.md / AGENTS.md — generated from an interview, not a blank page. Your AI knows who you are, how you work, and what you care about, in every session.

One memory vault, shared by both agents — Claude Code's CLAUDE.md and Codex CLI's AGENTS.md are both generated from the same interview and read from the same vault. Teach it something in one tool, and it's there the next time you open the other — not two AIs with two separate memories. The vault itself is Obsidian-convention PARA structure (works with Obsidian if you use it, doesn't require it): plain files you own, readable by you, versioned in git. It stays on your machine — Wienerdog never pushes it anywhere on its own; publishing it, if you ever want to, is something you do yourself. Wienerdog also scrubs secret-looking text (API keys, tokens) out of what it reads and writes, but that scrubbing is best-effort, not a guarantee — if you know a real credential got captured, follow the secret-incident guide.

Dreaming — a nightly job reviews the day's conversations, promotes what matters into long-term memory through quality gates, and turns your repeated workflows into reusable skills. Skills Wienerdog created keep learning: the dream watches how they perform in real use and revises them over time — your own skills and Wienerdog's built-ins are never touched (see the threat model). Every night is at most one git commit; anything can be reverted. If your computer is off or asleep at that time, don't worry — Wienerdog catches up automatically the next time you're back. The dream can also propose edits to your identity notes — who you are, your preferences, your goals — once it's noticed the same thing across enough sessions, but nothing it proposes reaches your next session until you approve the exact text with wienerdog memory approve. Run wienerdog safety any time to see the status of every gated capability.

Google Workspace senses (optional) — Gmail, Calendar, and Drive access that is read-first and draft-first by design. On the path Wienerdog enforces — a local capability broker that holds the credentials and exposes only fixed verbs — your AI can only send what you explicitly granted, to recipients you explicitly approved, and unattended sending goes to your own address only; grants are created by you at the keyboard, never by the AI.1 Connect it any time with /wienerdog-google-setup; run wienerdog safety to see the status of every gated capability.

Laptop-friendly routines (optional, pick from a catalog) — after setup, choose from a menu of ready-made scheduled routines (morning digest, inbox triage, weekly review, …) run by your OS's native scheduler (launchd / systemd / Task Scheduler). Laptop was closed at run time? It catches up. No daemon. Each routine runs in a locked-down copy of your AI that's checked fresh before every run (see the threat model); run wienerdog safety any time to see the status of every gated capability.

Scheduled runs are verified before they run. At every nightly fire, an independent launcher first checks that the app's code matches its recorded content address and that the job still matches its digest-bound authorization descriptor; Claude and Git are only spawned from their pinned, structurally verified install locations. Edits to config.yaml or the app tree made outside wienerdog sync don't change what runs — the job refuses with an alert instead (fail closed). For a config.yaml edit the fix is one wienerdog sync; when the check couldn't confirm the app files themselves, the alert deliberately does not tell you to sync — syncing is not the safe next step there — and asks you to reinstall from a trusted source instead.2

Why Wienerdog and not a "personal AI agent" app?

Projects like OpenClaw and Hermes are impressive — and they are applications: gateways, daemons, servers you must run, secure, and update. Wienerdog is different by design:

Wienerdog is just files. Markdown, skills, hooks, config — interpreted by the AI tool you already run and trust. No new long-running software: nothing listens, nothing serves, nothing phones home. No extra hardware, no VPS. And because everything executes through your own Claude Code / Codex subscription, you stay fully within your provider's terms of service.

Built by its own product

This repo dogfoods Wienerdog from day one: its own memory vault lives in memory/, its development conventions are its own CLAUDE.md, and most of its code is written by mid-tier AI models following its spec system (see docs/specs/). The product thesis — an AI gets dramatically better when you install the right files around it — is also how this project is built.

Documentation

Vision · Product requirements · Architecture

Threat model — read this if you're wondering "is auto-written AI memory safe?"

Contributing — AI-assisted contributions welcome; we have a spec system built for them.

License

MIT

Footnotes

This is a claim about what Wienerdog's own broker and CLI enforce against a hijacked AI, not an absolute guarantee against other software running as the same user on your machine — anything running as you can already read the same files. See the threat model (T4a) for the honest boundary. ↩

Boundary, stated plainly: this protects against scoped file writes and detects drift between syncs. It is not a defense against arbitrary same-user native malware — that requires OS-level anchoring (see THREAT-MODEL). ↩

Topics

Resources

Readme

MIT license

Code of conduct

Code of conduct

Contributing

Contributing

Security policy

Security policy

Activity

Custom properties

Stars

3 stars

Watchers

0 watching

Forks

0 forks

Report repository