Story How it works Tests Whitepaper Download Author Contact

How it works

A two-gate system: write structured files, compile them into one, read anywhere.

Two-gate architecture

DARA.md (Constitution) 15 rules · Entry point for all AIs VAULT/ (Gate 1: WRITE) NEURONS/ (projects) ENABLERS/ (tools) INBOX/ (feedback) Any AI writes here compile.py LIBRARY/ (Gate 2: READ) BRAIN.md Single compiled file Writers deal with individual, well-structured source files. Readers get a single file that fits in a context window.

What's a Neuron?

A Neuron is one topic. One file. One concern.

# Examples
project-alpha.md — everything about your main project
dara-system.md — the memory system's own documentation
profile-claude.md — how Claude should interact with Javier

Each starts with →brain: — a compressed summary line that goes directly into BRAIN.md. The rest is detail for when an AI needs depth.

What's an Enabler?

An Enabler is a tool, agent, or credential.

agent-architect.md — instructions for the system architect role
cred-cloudflare.md — how to access Cloudflare (no secrets in plaintext)
tool-typingmind.md — how TypingMind is configured

Same encoding, same rules. Just a different category.

The compiler (compile.py v3.1)

1,170 lines. Python stdlib only. No dependencies.

10-step pipeline:

1. Backup current BRAIN.md
2. Prune old backups (max 60, max age 180 days)
3. Read all VAULT files
4. Auto-purge archived entries (60-day soft-delete)
5. Deduplicate (name overlap + Jaccard similarity)
6. Validate + auto-fix (16 validator functions)
7. Check INBOX (count pending feedback)
8. Compile BRAIN.md (summaries + SHA256 checksum)
9. Delta report (per-entry size comparison)
10. Git auto-commit

Run it manually: python compile.py
Or let the watcher run it automatically every time a file changes.

The writing protocol (W1–W15)

RuleDescription
W1Check first — look up in BRAIN.md INDEX before creating a new neuron
W1(b)External-session writes — read DARA.md fully before writing from any external context
W2Fix errors — self-healing. Spot it, fix it, log it.
W3(a)Never delete files without explicit owner instruction
W3(b)5 protected files (Architect-only)
W3(c)Other files: any AI may fix errors; structural changes need permission
W4Update existing when: new info on same topic, data changed, error detected
W5Create new ONLY when: topic doesn't fit anywhere existing
W6After any write: regenerate BRAIN.md (watcher does it in ~8s)
W7Log everything in changelog.md — append only
W8File naming: kebab-case, always
W9Session closing trigger — ask "anything DARA should remember?"
W10Live memory detection — flag significant info as it arises
W11Write lean, review everything — consensus flags for ambiguous removals
W12Verify writes via shell after edits
W13Structure for BRAIN efficiency — large neurons start with summary
W14Enabler summaries — every agent file starts with summary:
W15Platform-aware reading — Standard vs Light mode

Fix or flag, never ask

The cardinal rule for AI behavior inside DARA:

If certain → fix it.

If uncertain → flag it (INBOX).

Never ask the human "should I update this?"

This is what makes it self-maintaining. Every AI leaves the system slightly better than it found it.

Protection mechanisms

Five layers ensure nothing gets lost, corrupted, or silently deleted.

Integrity locks

SHA256 checksums on 5 critical files. Any unexpected modification is detected immediately at compile time.

Machine-readable warnings

BRAIN.md starts with a header that tells every AI: "this is compiled output — don't edit directly." Prevents accidental overwrites.

Consensus for deletion

Removing content requires 3 independent AI confirmations across separate sessions. No single AI can unilaterally delete knowledge.

Git auto-commit

Every compile triggers an automatic commit. Full history of every change. Rollback to any point in time with one command.

Timestamped backups

Automatic snapshots before every compile. Auto-rotated: max 60 backups, max 180 days. Even if git fails, backups are there.

Governance: from writer to architect

Every AI interacts with DARA at one of three levels. No configuration needed — the level is determined by what the AI does, not who it is.

W

Writer

Any AI starts here. Read BRAIN.md, write neurons, fix errors (W2), log in changelog. This is 95% of all interactions. No activation needed.

L

Librarian

Any AI can execute agent-librarian.md and become the Librarian. 7-phase maintenance: 11 health checks, INBOX triage, corrections, reporting. Self-triggers every 3 days.

A

Architect

Any AI that detects a structural problem can request Golden Door activation. Modifies the Constitution (DARA.md), compiler, and protected files. Dual-gated: Golden Door + SHA256 checksums.

The system maintains itself without the owner's involvement — until something structural needs to change.

Self-healing in three layers

Errors don't accumulate. They get caught and fixed at three different moments — automatically.

Layer 1: In-flight

Any AI that reads VAULT data and spots an error fixes it directly. No permission needed, no escalation, no waiting. The error disappears before the next session even starts. This is rule W2 — self-healing is expected behavior.

Layer 2: At compile time

16 validator functions catch what AIs miss: broken references, header mismatches, encoding issues, null bytes. The compiler auto-fixes what it can and flags what it can't. Runs every time, automatically.

Layer 3: Cross-session consensus

Ambiguous content gets flagged for review. Other AIs in future sessions vote independently. At 3/3 confirmations, it's removed. Disagreement resets the flag. After 30 days without consensus, the flag expires — content stays.

The system gets better with every interaction. No human cleanup needed.

Works with every AI, even without tool calls

Standard Mode

AI reads full BRAIN.md, opens VAULT files as needed for depth. For Claude (Cowork, Desktop), GPT with plugins, Gemini with filesystem, TypingMind, and any model with reliable file access.

Light Mode

AI reads only BRAIN.md (INDEX + summaries). For DeepSeek, ChatGPT web, and models without filesystem access. Same core data, smaller context footprint.

The protocol adapts to the model's capabilities — not the other way around.

The foundation is ready. Now it evolves.

Each feature builds on a system that already works — tested across 4 models, proven stable. Cloud sync will come, but it will never be required. Local-first, always.

🔍

Semantic search

Find neurons by meaning, not filename. Ask DARA "what do I know about pricing strategy?" and get ranked results across the entire vault.

MCP server

Native integration for Cursor, Cline, and any MCP-compatible IDE. Your editor reads BRAIN.md automatically — no manual file loading.

Conflict detection

When two AIs write contradictory information about the same topic, the compiler flags it. No silent overwrites. No lost context.

📈

Web dashboard

Visual overview of your vault: neuron map, health metrics, changelog timeline, INBOX queue. Read-only — your data stays local.

🤖

Agent marketplace

Community-built agents as downloadable enabler files. Install a "research assistant" or "code reviewer" agent by dropping a file in your vault.

Optional cloud sync

Encrypted sync between devices for users who want it. End-to-end encrypted. Self-hostable. Never required — local-first, always.

Read the whitepaper →