<p align="center" <img src="docs/assets/mascot.svg" alt="Muse — the bluebird mascot" width="120" / </p
<p align="center"<iMeet Muse — a personal AI project built to understand the life you are already living.</i</p
<h1 align="center"Muse</h1
<p align="center" <bA personal AI that learns how you live and work—and gets better at knowing when and how to help.</b<br/ <iLocal-first, provider-neutral, and honest about what is not built yet.</i </p
<p align="center" <a href="LICENSE"<img alt="License: MIT" src="https://img.shields.io/badge/License-MIT-green.svg" /</a <a href="package.json"<img alt="Node ≥ 22.12" src="https://img.shields.io/badge/node-%E2%89%A5%2022.12-43853d.svg" /</a <a href="https://www.typescriptlang.org/"<img alt="TypeScript" src="https://img.shields.io/badge/built%20with-TypeScript-3178c6.svg" /</a <a href="#what-muse-will-not-do-boundaries"<img alt="Local-first" src="https://img.shields.io/badge/privacy-local--first-6f42c1.svg" /</a <a href="https://ollama.com"<img alt="Runs on Ollama" src="https://img.shields.io/badge/runs%20on-Ollama-000000.svg" /</a · <bEnglish</b · <a href="README.ko.md"한국어</a · <a href="README.ja.md"日本語</a · <a href="README.zh-CN.md"简体中文</a </p
Muse is a continuing personal agent for one person's life and work, not only a work assistant. Its north star is Attunement: learning when help fits, when quiet is better, and whether the last suggestion actually helped.
The first proof point is Personal Continuity. You choose a life or work thread and link its exact local tasks and notes; Muse can then help you resume it without reconstructing everything. Automatic thread detection, observation, and timing remain roadmap work.
What works today: personal memory, grounded recall, local personal stores, guarded tools and browser actions, traces, checkpoints, and the first explicit Personal Continuity path. See the product contract and implementation plan.
The Muse moment we are building
You stopped while comparing three places to stay. Since then, the flight changed, one cancellation deadline moved close, and an 18-minute gap opened in your day. Muse prepares only the changes, explains why now, cites every source, shows exactly how far an action would go, and asks whether to hold one option.
If you answer “this evening instead,” Muse proposes a visible rule scoped to that trip—not a hidden global preference. That experience has three parts:
- Shadow Muse learns when to help or stay quiet before it interrupts.
- Continuity Capsule restores the stopping point, changes, evidence, next step,
prepared work, and expected time.
- Policy Card shows what Muse proposes to learn about collaborating with you, with
evidence and trial/edit/reject/rollback controls.
Muse does not remember apps; it remembers the state you intended to continue.
The full signature experience remains a roadmap, not a shipped claim. Its first library-level substrates now live in the partially implemented, lightweight Attunement Graph Engine: an agent-native temporal/provenance graph and personal context compiler, not merely a third-party graph DB. Its intended advantage is a small set of bounded, verified personal-temporal operators: the model asks what changed, what evidence supports a policy, or what forgetting would invalidate; Muse computes the exact path, completeness, and authority boundary in code. The Agent-Native Graph Core blueprint defines the proposed next architecture: scope-safe snapshots, proof-closed Working Graphs, typed completeness, an immutable logical journal, and a lightweight local storage Adapter with no external Graph DB requirement. The first changesSince-style operator is shipped as an I/O-free library contract; no Capsule UI, automatic timing, durable graph, or action authority is implied. Muse can now also seal one exact caller-declared projected observation as a bounded, content-addressed Observation Receipt without copying its personal source text. This preserves the previous observation's recorded next step; it does not prove the user's exact stopping point. The same library subpath can purely capture one caller-supplied raw Continuity snapshot through the shared projector and return that receipt without source I/O or persistence. The existing state-to-state query now delegates to one reusable internal prepared-observation comparison core. The observation subpath can also verify that receipt, derive its exact boundary, project one caller-supplied current snapshot, and return the same explained-change result as the raw state-to-state query. That receipt is caller-declared integrity input for the future Capsule—not proof of an external observation, automatic stop-point detection, persistence, or the Capsule itself. Muse now also has a bounded trusted-host Provider that reads one configured local Attunement file and mints a process-local, content-addressed snapshot capture. Its serializable receipt proves integrity only; exact Provider provenance belongs to the in-process capture, freshness remains unassessed, and missing data never becomes an absence claim. This closes the first real-source boundary for the Agent Graph without making the graph read files or depend on an external Graph DB. The private Agent Graph seam can now verify that exact process-local mint before reading state, independently recompute its bytes and digest, project it into a verified Continuity Observation Receipt, and compile receipt-bound graph evidence with truthful Provider snapshot provenance. It never fabricates a graph commit or generation: a single read is explicitly unassessed, which forces downstream settlement to abstain while still preserving exact evidence links and bounded nomination accounting. This is process-local engine substrate, not a public API, durable graph, current-freshness proof, Capsule/runtime wiring, action authority, or user-visible behavior. See the separate wow + graph roadmap.
<p align="center"<img src="docs/images/web-home.png" alt="Muse console home — model chip, integrations, and what Muse has learned" width="860" /</p