EIDARA is open-source, local-first context engineering for AI agent memory.
One compiled markdown brain that Claude, GPT, and DeepSeek all read and write — through MCP.
Governed by a constitution, maintained by an autonomous curator. Yours forever. No cloud, no lock-in.
Proven across vendors: Claude and Codex now share one local brain over MCP →
Every AI session starts from scratch. You explain the same project twice. You repeat decisions you made three conversations ago. You waste tokens keeping context alive in threads that should have ended hours ago.
Your AI is powerful. But without memory, you're doing the remembering for it.
SUPER DARA is the autonomous version of EIDARA: a local curator that maintains your memory around the clock, plus senses that capture what matters — so you stop feeding it by hand. Still local, still yours.
Your AI saves knowledge to local files. A script organizes them. Next session, any AI reads everything.
Your AI writes what matters —
decisions, context, preferences —
one topic per file.
A local script validates,
cleans, and organizes
everything automatically.
Any AI opens one file
and instantly knows
your full context.
Save once. Available everywhere.
No terminal commands. No file management. Two phrases are all you need.
→ To recall context
"Check EIDARA for context on project X."
Seconds later, the AI knows everything —
even if it's never seen this project before.
→ To save something new
"Save this decision to EIDARA."
The AI writes it. Next session —
on any platform — it's there.
Claude, GPT, DeepSeek — any model that reads text. In practice, clients from different vendors (Anthropic and OpenAI) already share one local brain over MCP. Switch mid-project; your context follows. No lock-in.
Your files stay on your computer. No cloud account, no API keys, no telemetry, no data leaving your machine. Just markdown files and a Python script.
A compiler validates every entry, removes duplicates, fixes formatting errors, and creates a verified snapshot. You never manually clean up your memory.
When an AI spots an error — outdated info, wrong format, broken reference — it fixes it automatically. The system gets better with every interaction.
Your scattered context — notes, decisions, preferences — is compiled into one compact file the AI reads cold, in a single pass. No retrieval, no embeddings, no vector database. In a real case, ~30,000 tokens of context distilled to ~3,500 — the same memory, a fraction of the size.
EIDARA doesn’t just remember what you type. It has three senses in daily use: it listens to what you dictate, reads what you commit as you type, and picks up the text in the window you’re working in. Passive capture of what you’d have written down anyway — intentional, never ambient, with a one-click privacy control.
52 real-world scenarios across 4 AI models — remembering project decisions,
recalling preferences, handling updates, resolving conflicts. No cherry-picking.
Check the ones that apply.
Download EIDARA. Give the INSTALL file to your AI. It handles everything —
personalization, setup, first compile. You don't type a single terminal command.
MIT License · Python 3.10+ and Git required · macOS, Windows, and Linux