mindpalace
aashutosh396/mindpalace
Agente pessoal que aprende, lembra e trabalha para você — terminal + Discord
Ver no GitHubAgente pessoal de IA autônomo e sempre ativo que roda na sua assinatura Claude Max, com interfaces simultâneas em terminal e Discord. Aprende continuamente, mantém uma memória persistente estruturada, cria suas próprias skills e executa tarefas em background.
Instalação
npx convez-ai aashutosh396/mindpalace1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Automatizar tarefas repetitivas (backups, lembretes, agendamentos)
- Manter um vault de conhecimento estruturado (infra, contas, projetos)
- Executar trabalho real em background enquanto você continua conversando
- Criar skills reutilizáveis a partir de procedimentos realizados
Veredito

Uma abordagem ambiciosa de agente autônomo com persistência e auto-aprendizado. A arquitetura é limpa e bem documentada, mas demanda setup significativo e compreensão técnica sólida para aproveitar todo seu potencial.
README
A tradução para o português deste README ainda está na fila.
🧠 mindpalace
A self-learning, always-on personal AI agent that runs on your Claude Max subscription.
Ships generic — sharpens into your world with every conversation. Lives in your terminal and on Discord at the same time. Remembers, learns your conventions, writes its own skills, and runs real work in the background.
Quick start · Features · Docs · Contributing
One command. Talk to it like a person. It replies fast, runs real work in the background — backups, reminders, schedules, code, cron, deploys — files everything it learns into a structured knowledge vault, and gets more precise for you over time. No metered API: it runs on the Claude Max subscription you already have.
✨ Why mindpalace
- 🗣️ Two interfaces, at once — a Claude-Code-style terminal chat and an always-on Discord bot, sharing one brain, memory, and vault.
- 🧩 Self-learning — it writes durable facts to memory, and an Analyst agent turns first-time procedures into reusable skills automatically.
- 🗄️ Knowledge vault — every server, login, project, and decision is filed into a structured, resource-first vault (your second brain), not lost in chat.
- ⚡ Async + alive — long tasks run as background jobs and stream their steps live; the chat never blocks.
- 💓 Proactive — an autonomous heartbeat reviews your world on a timer and reports.
- 🤖 Multi-bot — spin up scoped Discord bots, each with its own persona + a real permission fence (read-only / full / custom tools).
- 🔌 Extensible — a clean
core / agents / gateways / pluginsarchitecture. - 🔐 Yours — self-hosted, admin-locked, secrets stay on your machines. Core ↔ data fully separated, so updates never touch your private data.
🚀 Quick start
git clone https://github.com/aashutosh396/mindpalace
cd mindpalace
./install.sh # add `discord` for the always-on bot: ./install.sh discord
mindpalace # first run: pick a gateway, create its identity, start talking
Requirements: Python 3.9+ · the Claude CLI logged in to a Max plan · (Discord) a bot token with the Message Content intent.
🗣️ Terminal + Discord — both, at once
- Terminal is always there:
mindpalaceopens a polished shell chat (boxed input, live step streaming, markdown replies). - Discord runs as a background service once configured — it never blocks your terminal. You're reachable from both at once; the first person to message the home channel becomes the first admin.
mindpalace gateway discord # token + home channel
sudo mindpalace service install # background systemd/launchd service, survives reboot
🧠 How it learns (the compounding loop)
- You ask it to do something. It does it — streaming each step live.
- The Analyst agent reflects in the background: files facts into the vault
(
infra/,accounts/,projects/,runbooks/,LOG.md) and skillifies any reusable procedure intoskills/. - Next time, it reuses the skill + recalls relevant history (SQLite FTS). It gets sharper.
Global skills (shipped, read-only references) → the agent derives private user skills tailored to your exact setup. Your world compounds into a superb, personal assistant.
🤖 Scoped bots
mindpalace add-bot
> What should this bot do? "summarize deploy-log errors, nothing else"
…Claude drafts a scoped system prompt → you pick a permission tier → paste a token
@mention it anywhere. A readonly bot literally cannot run commands — the fence is the
Claude CLI tool allowlist, not just a polite prompt.
🖥️ Run on a server (headless, always-on)
mindpalace setup
sudo mindpalace service install # systemd (Linux) / launchd (macOS), reboot-persistent
# close SSH — work entirely from Discord. logs: journalctl -u mindpalace -f
🧱 Architecture
mindpalace/
├── cli.py · config.py · setup.py · identity.py · skills.py · bots.py # spine
├── core/ brain · daemon · jobs · heartbeat · service · notify # engine
├── agents/ analyst (+ base, registry) # cognitive agents
├── gateways/ terminal · discord # interfaces
├── memory/ store · session_store (FTS) # recall
├── plugins/ base · loader · contrib # extensibility
└── skills/global/*.md # shipped skills
Your private instance lives outside the repo in ~/.mindpalace/ (config, secrets,
identity, memory, vault, skills, state). Update the engine → your data is untouched.
📚 Documentation
Full docs in docs/:
Getting Started ·
Configuration ·
Gateways ·
Agents & Skills ·
Plugins ·
Commands Reference ·
Architecture
🤝 Contributing
mindpalace is open source and built to grow with its community. Issues, ideas, skills, and PRs are all welcome — start with CONTRIBUTING.md. If it's useful, please ⭐ star the repo — it genuinely helps others find it.
📄 License
MIT — see LICENSE.