captain-meridian-stack

The meta layer of one person's daily-driver personal AI infrastructure: roadmap, architecture, decisions, and the competitive landscape that explains why this exists.

This is a published artifact, not a maintained project. Last verified: 2026-04-25.

---

What this is

The narrative half of the system. The actual workspace primitives live in atrium; this repo holds:

• ROADMAP.md — phases shipped, queued, parked
• ARCHITECTURE.md — how the layers connect (verified 2026-04-25)
• LANDSCAPE.md — where this sits in the 2026 personal-AI-infrastructure ecosystem (~12 frameworks, ~10 personal config repos, 3 philosophy attempts surveyed)
• CHANGELOG.md — what shipped when
• docs/ — phase documentation, archived items, blog posts

If you want to understand why anyone would build this rather than use Letta or Mem0 or LangGraph, read LANDSCAPE.md first.

If you want to understand the architecture without reading all the docs, ARCHITECTURE.md is the entry point.

If you want the long-form essay, see Building a home for an AI.

What this isn't

• A framework — see the LANDSCAPE doc for who's building those, well
• Multi-provider, multi-user, or generalizable
• A product, a startup, a community, a movement
• The full system — the operator config layer (~/.claude with hooks, agents, channels, people memory) stays private; it's too personal to share

How to read this repo

1. README.md (this file) — orientation
2. LANDSCAPE.md — where this sits in 2026
3. ARCHITECTURE.md — how it's wired (3 lenses: layers, repos, memory surfaces)
4. docs/posts/2026-04-25-building-a-home-for-an-ai.md — the philosophy in long form
5. Then atrium for the actual code

What "artifact, not project" means

• No issues accepted. Not a place for bug reports or feature requests.
• No PRs accepted. Adapt the patterns into your own version.
• No support promised. Built and used daily, but no time owed to questions.
• No promises about updates. Dated snapshot. Re-verify against your own use, not against expectations of upkeep.

If this inspired you, build your own version and link it back. A network of personal-AI-infrastructure attempts beats one canonical fork — different operators will solve different parts well.

License

Writing (all .md files in this repo) — Creative Commons BY-NC-SA 4.0. Adapt freely for non-commercial use, attribute, share-alike. Commercial republication (paid courses, books, packaged products) requires permission.

Code excerpts embedded within markdown files — MIT.

See LICENSE for full terms and NOTICE for the per-file-type breakdown and the (non-binding) preference statement on AI training data.

The writing has guardrails against commercial repackaging; the code in code blocks is permissive. The philosophy transfers by living with it, not by copying.

Companion repo

atrium — workspace primitives. The cognitive verbs as shell scripts, the philosophy docs (AGENTS.md, PRINCIPLES.md, PATTERNS.md), the brain corpus.

Author

Upua (Upua on GitHub). Built collaboratively with Claude. The collaboration is named in the writing because the framing was earned through it, not stipulated.