AltarOS

A sovereign intelligence system for people holding complex, multi-threaded work.

What It Is

An instrument, not a task manager

AltarOS runs entirely on your machine. It observes the threads of work you're holding — how they relate, where attention is gathering, what's ready to move — and reflects that pattern back to you.

You speak your thoughts. The system transcribes, routes, and accumulates them as raw material. Over time, a topology of relationships emerges between your projects, people, and commitments. Local AI models synthesize patterns without your data ever leaving your device.

It's built on the understanding that complex work isn't a set of tasks to manage — it's a living field of attention to observe.

What You Get

Nine slots. One surface.

Your altar holds up to nine active threads — the domains that occupy your attention right now. Everything else rests in the field, ripening until it's ready.

Voice Capture

Speak your thoughts. Local transcription routes them to the right project nodes, people, and commitments.

Discover

A conversational interface over your entire field. Ask questions, surface patterns, trace connections across projects.

Topology

A living coupling graph that tracks how your projects, people, and entities relate — built from weeks of observation, not manual entry.

Morning Practice

Daily coherence check-in. Commitments surfaced with actionable follow-through. Start the day oriented.

Synthesis

Three-layer intelligence — rules, local LLM, Claude API — finds momentum, gaps, and suggested moves across your entire field.

Sangha

Relational topology for the people in your work. Dunbar layers, active threads, shared documents — connection as infrastructure.

Temporal Intelligence

Commitments, deadlines, and coordination timelines extracted from your voice input. Complex multi-person schedules tracked across projects.

Extensible

Build custom views, skills, and workflows through Claude Code. Adapt the system to your domain.

Architecture

Sovereign by design

Nothing syncs. Nothing phones home. Your vault is a set of markdown files on your machine, processed by local AI models through Ollama. The system observes and suggests — but the consent membrane ensures nothing changes in your vault without your explicit approval.

Your Machine

  AltarOS (Streamlit)
    ├── Altar Grid · 9 active nodes
    ├── Voice → Transcription → Routing
    ├── Sangha · Relational topology
    ├── Topology · Coupling graph
    ├── Local LLMs (Ollama · qwen3.5)
    └── Your vault · never shared

  ↕ Consent membrane

  Sangha OS Commons (opt-in shared layer)
    ├── Shared project state
    ├── Entities & commitments
    └── Push intentional · Pull free

The shared layer — Sangha OS — is where teams coordinate. Push project state, shared entities, relational context, and complex coordination timelines to the commons. Pull whenever you want to see what others have contributed. The consent membrane ensures nothing leaves your vault without your explicit action.

Extensibility

Build on it. Make it yours.

AltarOS is designed to be extended, not just used. Through Claude Code, you can build custom views, skills, and workflows that adapt the system to your specific domain — regenerative development, production coordination, research, governance, whatever your work demands.

Custom Views

Build new interfaces over your vault data. Timeline views, coordination dashboards, domain-specific displays.

Skills

Reusable configurations — prompt templates, entity schemas, voice tuning, connectors. Shared with the community, attributed to you.

Workflows

LLM-powered pipelines that run locally. Weekly digests, coordination summaries, pattern detection tuned to your context.

IP Contributions

New analytical methods and synthesis architectures that make the engine smarter for everyone. Recognized through fractional ownership in the Contribution Pool.

Principles

What it's built on

Sovereign

Your data lives on your machine. AI runs locally. Nothing leaves without your intent.

Observational

The system watches and reflects. It doesn't direct, assign, or optimize. It helps you see what's already there.

Consensual

Every vault mutation requires explicit approval. Every commons push is intentional. The consent membrane is the architecture.

Relational

Built on Koronetics — a framework for coherence in living systems. Nodes aren't tasks. They're threads of attention in a relational field.

Emergent

Topology builds over time. Patterns surface from observation, not configuration. Give it a week.

Local-first

Obsidian vault. Markdown files. No vendor lock-in. If AltarOS disappeared tomorrow, your data would still be exactly where you left it.

Getting Started

What you need

A Mac with about 10GB of free disk space and 30 minutes for setup. AltarOS runs as a local application — Streamlit serves the interface, Ollama runs the models, and your Obsidian vault provides the data layer.

You'll start with 3–5 nodes: the threads of work that occupy your mind right now. Make your first voice transmission. The system begins learning from your first session — and the topology deepens with every interaction.

Early access is a
collaborative relationship

You get a working instrument that evolves with your use. We get insight into how the system meets real needs. Together, we make it better.

Request Access