Download EIDARA

Current release

EIDARA v1.0 — first public release

29 files · GitHub-ready · MIT License

Includes: compile.py (1,170 lines), validators.py (322 lines, 16 pure functions), watcher.py (auto-compile on file changes), DARA.md (constitution — W1–W15), INSTALL guide (11-step AI-guided setup), folder structure, 103 pytest tests, .gitignore (pre-configured).

        $ git clone https://github.com/jrotllant/eidara.git
$ cd eidara
$ python compile.py
      

Clone from GitHub → Download ZIP

Try the demo (Sample Vault)

Want to see DARA in action before personalizing? Download the Sample Vault — a pre-populated demo with 17 neurons, 6 enablers, a populated changelog, and INBOX feedback. Drop it on your Desktop, run python compile.py, and explore the system in action in 2 minutes.

Download Sample Vault (.zip) →

AI-guided install

Download the folder

Put it on your Desktop

Open any AI. Give it the INSTALL file.

The AI does everything else.

Works with Claude, GPT, DeepSeek, or any capable model. You don't type a single terminal command.

Get notified of new releases

EIDARA evolves. New validators, new features, better tests. Each update comes with an UPDATE.md file — give it to your AI and it applies the changes automatically.

Only version releases. Typically 1–2 emails per month. Unsubscribe anytime.

Public roadmap

When	What	Why it matters
v1.0	GitHub release + eidara.dev + newsletter	The launch
Jul 2026	Semantic search (SQLite)	Find info without remembering exact words
Sep 2026	MCP Server	Connect DARA to Cursor, Cline, Claude Desktop
Q4 2026	Web dashboard + audit trail	Who wrote what, when. Full traceability.
2027	Agent marketplace + optional cloud sync	Community agents. Cloud never required.

Want to know when these ship? Get notified ↑

FAQ

Do I need to know Python?

No. You need Python installed (the INSTALL guide helps with that), but you never write Python code. The compiler is a black box — you feed it markdown, it produces BRAIN.md.

Does it work with my AI model?

If the model can read a markdown file and follow written instructions, it works with DARA. We've tested Claude, GPT, DeepSeek, and Opus 4.7. The protocol is model-agnostic by design.

How much context does BRAIN.md use?

Starts at ~900 tokens with the starter vault. Scales efficiently — a typical setup with 20+ entries fits in ~3,500 tokens. Budget warnings at 50K/100K/150K. Designed to fit any modern context window.

What if I break something?

Git tracks every change. RECOVERY.md has 6 emergency scenarios. The compiler creates timestamped backups automatically. Worst case: git checkout the previous commit.

Is my data private?

100%. Your files never leave your machine. No telemetry. No cloud sync. No analytics on your content. Local-first, always.

Can I try it before personalizing?

Yes. Download the Sample Vault — a pre-loaded demo with example neurons, enablers, and a pre-compiled BRAIN.md. Run compile.py and see it rebuild. Takes 2 minutes.