nanoMuse
nanoMuse is an open-source Meta Muse agent that runs on a phone, keeps working while the app is closed, asks before performing actions you cannot undo, and operates apps through their screens. Supports any OpenAI-compatible model.
Ficha actualizada 23 sept 2026, 8:06 · upstream 23 sept 2026, 7:17 · release v0.7.0
Qué es nanoMuse
Descripción editorial construida a partir del repositorio, su documentación y sus señales de mantenimiento.
OpenMuse is a personal agent you talk to from your phone. It can research, write pages and documents, run shell commands and Python scripts, send mail, read your calendar and contacts, browse the web, stop before performing irreversible actions with an approval mechanism, work on goals over weeks while the app is closed, check in on a schedule, and start work when new mail arrives or events occur. It also writes you a feed of short posts about what it knows about you and what you asked it to follow. The agent remembers you through memory that can be read, edited, or forgotten, recalled by meaning rather than keywords. It follows skills in the Agent Skills format and uses any MCP server. It runs on any OpenAI-compatible model including DeepSeek, OpenAI, OpenRouter, Ollama, vLLM, and gateways with their own headers.
- Research and write comparison articles between two products (e.g., Sony WH-1000XM6 vs Bose QuietComfort Ultra) for long flights
- Check available free disk space on a machine without opening the terminal
- Set up conversational Japanese practice before a trip to Kyoto in December with 30 minutes daily
- Create and maintain a weekly newsletter following topics you've asked about previously
- Automate routine tasks like checking weather, sending reminders, or updating calendar events
- Write blog posts based on your goals and past conversations without manual drafting
- Manage personal finance by reading bank statements and generating summaries
- The agent never directly touches tools; a Sentinel layer decides allow/ask/deny per call, keeping secrets out of the model context
- Runs on any OpenAI-compatible endpoint including local models via Ollama or vLLM with streaming support
- Provides a feed of your own thoughts and what you've asked it to follow over time
- Supports MCP servers for accessing external tools like databases, APIs, and file systems
- Runs entirely on the phone without requiring cloud API keys or server infrastructure
- Every tool call is audited via audit.jsonl with configurable risk levels per tool
- The agent cannot perform actions that require physical access to your device (e.g., installing software, modifying system files)
- All data flows through the model context window; very large documents may cause truncation
- No native support for Windows desktop applications or mobile OS-level permissions beyond web browsing
- The agent cannot execute commands that require elevated privileges on Linux systems (e.g., sudo without root access)
- Memory is limited by token budget and can be overwritten if not managed carefully
- No built-in multi-agent coordination; all tasks run sequentially within a single session
Perfil de infraestructura
Estimación inicial derivada del repositorio y de los servicios detectados.
Despliegue detectado
Qué señales reales de instalación hemos encontrado.
Run with `docker compose up -d app` to start the agent in detached mode. The image supports linux/amd64 and linux/arm64 architectures. Use `-e DEEPSEEK_API_KEY=sk-...` for API keys, or set environment variables via Docker Compose secrets. Mount volumes (`-v openmuse-data:/data`) persist your data between runs. For production use, consider running behind a reverse proxy with TLS (e.g., nginx) to protect the agent from direct access. The web app can be served alongside on the same host for local testing.
Stack y dependencias
Componentes inferidos a partir de manifests, configuración y documentación.
Puede sustituir a
Relaciones detectadas por el análisis del repositorio y su categoría.
Experiencia real de la comunidad
Ayuda a contrastar la estimación automática con instalaciones reales.
Todavía no hay datos comunitarios para este proyecto.