Most AI agents are tied to a platform. You configure an agent in TypingMind, or n8n, or Claude, and it lives there. Move to a different platform? Rebuild the agent from scratch. Share with a colleague? They need the same platform, the same configuration, the same setup.
This creates lock-in. Your agents become hostages of the tool you used to create them.
What if an agent was a document instead of a program? Write the instructions in a .md file. Any AI that reads it becomes that agent — Claude, GPT, DeepSeek, doesn't matter.
In EIDARA, agents live in the ENABLERS folder. Each file starts with a summary: line and contains the complete protocol:
summary: Quality guardian for the vault. Runs health checks, fixes encoding, processes feedback.
## Protocol
1. Read current state
2. Back up protected files
3. Run 11 health checks
4. Report findingsAny AI that reads this file becomes the librarian. It follows the steps exactly as written. The intelligence lives in the file, not in who executes it.
To share an agent with a colleague, you copy the .md file and drop it in shared Drive, Dropbox, or any shared folder. Your colleague opens it with their AI. That's it.
No installation. No configuration. No platform dependencies. Their AI reads the file and becomes that agent.
No lock-in — your agents aren't tied to a platform. Move from Claude to GPT tomorrow and your agents still work. Version control comes free — an agent is a text file, so Git tracks every change. You can diff, revert, and audit. Full transparency — anyone can read an agent's instructions, no black boxes, no hidden prompts. And composability — combine multiple agent files to create complex workflows where each file handles one concern.
I share my agent-librarian.md with colleagues. They drop it in their EIDARA instance, their AI reads it, and they instantly have a vault maintenance agent. No setup calls. No "works on my machine." Just a file.
EIDARA is the open-source project that makes this possible. GitHub · Website