⌬ AIIA — AI Information Architecture

A·I·I·A. A palindrome on purpose: the architecture mirrors the name.

Your memory on one side. Your AI tools on the other. AIIA is the surface in between — where what you decide becomes what they know, and what they notice becomes what you remember.

Without it, every session starts blank. Every decision gets re-explained. Every pattern gets re-derived. With it, one continuous thread runs between your thinking and the tools that act on it. Claude Code, Cursor, Copilot — pick your tool. The mirror works the same.

Open source. Apache 2.0. Local-first. Free model. Your data stays yours.

Quick start

pip install aiia              # PyPI install
aiia                          # first-run check + setup prompt
aiia next                     # surface your top story to work on now
aiia ask "what did I decide about lockfile policy?"

That's the daily loop. aiia next chooses, aiia ask recalls, aiia memory add captures.

If you don't want PyPI, clone and pip install -e .[dev]. AIIA needs Python 3.10+, ~5GB free for a local model, and either Ollama running on localhost:11434 or an ANTHROPIC_API_KEY if you prefer cloud tiers.

What flows through the mirror

From you, into your AI tools' working context:

Decisions you've made, with the why — recoverable, not buried in a chat log
Patterns and conventions you've named — applied without re-explaining
Stories in your backlog — surfaced when relevant, ranked by priority × freshness
Session history — what you were doing when you stopped

From your AI tools, back into your record:

Observations from running code — what changed, what broke, what's pending
Decisions captured during sessions — saved without you stopping to write them down
Lessons from debugging — kept for next time
New stories — when something turns up that deserves its own focused attention

The substrate that makes both directions work:

9 typed memory categories, quality-scored, decay-aware, hand-editable
Story prioritization across Obsidian, GitHub Issues, or a JSON file
Local-first three-tier LLM stack (LOCAL → ANTHROPIC → GOOGLE)
Background jobs that run while you sleep — security scans, daily reports, memory consolidation
MCP server so Claude Code talks to AIIA natively
A CLI (aiia) for direct use without going through any other tool

How it works

   Your memory ⇄ ⌬ AIIA ⇄ Your AI tools
         │
         ├── Decisions, patterns, stories, sessions persist on your Mac
         ├── Speaks to Claude Code / Cursor / etc. via MCP
         ├── Captures AI observations back into your record
         ├── Free local model by default; Anthropic / Google as fallbacks
         └── CLI surface (`aiia next`, `aiia ask`, `aiia memory`)

Two surfaces are running underneath:

Brain API on localhost:8100 — memory, RAG, LLM routing
Command Center on localhost:8200 — dashboard, story queue, voice

The CLI is a thin client over both. Claude Code reaches AIIA over MCP. Your terminal reaches AIIA over aiia. Same brain. Both halves of the mirror in sync.

What AIIA is, that other tools aren't

vs. Claude Code / Cursor / Copilot directly. Those are one-directional — they consume your context, but their observations don't flow back into your memory. Every session ends and the work is gone. AIIA writes both ways: your decisions reach their context, their work re-enters your record.

vs. RAG-only solutions. RAG is one-directional retrieval — find similar chunks, return them. AIIA is bidirectional flow — typed memory + scheduled work + story capture + safety-gated execution. Memory isn't "search this embedding"; it's structured records with categories, decay, and quality scoring.

vs. Custom GPTs / system prompts. Static one-shot context. AIIA is a continuous mirror — context updates as your work progresses, in both directions, without you copying anything anywhere.

vs. Computer-use agents (OpenClaw, NemoClaw, etc.). Those operate your computer. AIIA operates the layer between you and your tools — it maintains memory, schedules background work, and feeds context through MCP. Infrastructure, not a screen driver.

Who it's for

Solo developers and small teams running production SaaS who want their AI tools to have institutional memory
Platform engineers building multi-tenant products who need a local intelligence layer that knows every service, tenant, and deployment
Anyone tired of re-explaining context to Claude / Cursor / Copilot every single session

Built at Aplora AI

AIIA is open-sourced from the runtime layer powering Aplora AI, a multi-tenant AI platform for professional services — law firms, financial analysts, marketing agencies, healthcare operations. The substrate that mirrors your developer memory is the same substrate that mirrors Aplora's tenants' institutional memory: every legal doc reviewed, every financial pattern observed, every compliance decision made.

If AIIA helps your daily work and you want a managed version with team sync, audit logs, and SOC 2 / HIPAA posture, we're building that. The OSS version is the foundation. Everything load-bearing lives here, in public, under Apache 2.0.

Status

v0.5.0 (Q2 2026). Beta. Running in production on the author's machine since February 2026. Interfaces stable but evolving — pin a version if you depend on it.

See CHANGELOG.md for the release history. See SECURITY.md for vulnerability reporting. See CONTRIBUTING.md before opening a PR.

License

Apache 2.0. See LICENSE.

AIIA

README