Sovereign Control Over Memory and Identity
Akita exists because memory is identity. The system ensures that the person using it — not a platform, not a corporation — decides what is remembered, how it is organized, and who gets access.
Every piece of data in Akita (highlights, journal entries, health metrics, research dossiers) is stored locally. There is no vendor lock-in by design: data lives in files, SQLite databases, and a self-hosted vector store. If any external service disappears, the local copy remains.
Choosing What to Communicate to LLMs
Working with AI is a conversation, and conversations require trust. Akita gives its user fine-grained control over what context is shared with language models through the MCP protocol — tools expose specific slices of knowledge, not the entire corpus.
This is a deliberate architectural choice: the LLM sees what you choose to show it, when you choose to show it. Private notes can be encrypted. Draft thoughts can stay in the inbox. The boundary between self and machine is drawn intentionally.
Building the System Yourself
Akita is not a product. It is a personal system built by and for its user. Every component exists because it solves a real need, not because it was specified in a product roadmap. This means:
- No unnecessary abstractions: if three lines of code work, they stay as three lines
- No premature generalization: features are built for one person’s workflow
- Full understanding: the user knows every line, every decision, every trade-off
This is the opposite of consumer software. There is no onboarding flow, no settings page for features you’ll never use. There is only a system that does exactly what its builder needs.
Owning the Data Pipeline
From ingestion to publication, the entire pipeline is self-hosted and self-controlled:
Sources (RSS, Readwise, books, health, journal)
→ Ingestion (MCP gateway, Python)
→ Storage (files, SQLite, Qdrant)
→ Intelligence (deep research, briefings, summaries)
→ Publication (Astro → Cloudflare Pages)Each stage is inspectable, modifiable, and replaceable. The system doesn’t depend on any single service remaining available or unchanged. When Readwise changes its API, only one adapter changes. When a better vector store appears, only the storage layer changes. The knowledge itself — the notes, the highlights, the connections — persists regardless.
The Garden Metaphor
The digital garden is not a blog. It is a living, growing collection of ideas at various stages of maturity. Some notes are seeds (quick captures in the inbox). Some are growing (draft notes, fiches accumulating observations). Some are mature (published essays, book reviews with ratings).
This metaphor shapes the architecture: content moves through stages, from private thought to public expression, at its own pace. There is no publish-or-perish pressure. A fiche can sit for months, accumulating highlights and reactions, before being promoted to a full entry — or not.