Summary

Icarus Memory Protocol provides a universal, database-less shared memory system for AI agents, enabling seamless collaboration and knowledge sharing across any framework or platform. It addresses the challenge of inter-agent communication and persistent memory by storing agent interactions as markdown files in a central directory (~/fabric/). This allows agents to write, read, and search shared context, facilitating complex workflows and enabling self-training capabilities for improved performance and cost efficiency.

How It Works

The core mechanism relies on simple bash scripts (fabric-adapter.sh) to manage memory stored in markdown files within ~/fabric/. Each entry includes YAML frontmatter detailing agent, timestamp, type, and references. Memory is tiered by age: 'hot' (<24h), 'warm' (1-7 days), and 'cold' (>7 days). A curator.py daemon manages re-tiering, compaction (using Claude), and indexing. This approach offers a lightweight, highly accessible, and framework-agnostic solution for persistent agent memory.

Quick Start & Requirements

• Primary Install: Clone the repository and run bash setup.sh. For Hermes agents, copy the plugins/icarus/ directory and associated scripts to the agent's plugin folder.
• Prerequisites: Bash, Python, Git. For Hermes integration: a Hermes agent. For self-training: a Together AI API key. Node.js is required for the Claude Code CLI.
• Links:
  • Repository: https://github.com/esaradev/icarus-daedalus
  • Protocol Spec: PROTOCOL.md
  • Schema: SCHEMA.md
  • Demo: examples/hermes-demo/

Highlighted Details

• Hermes Plugin: Extends Hermes agents with 7 tools (e.g., fabric_write, fabric_recall, fabric_search) and 4 automatic hooks for context injection, memory retrieval, decision capture, and session summarization.
• Cross-Platform Demo: examples/hermes-demo/ showcases two agents (Slack/Telegram) collaborating and recalling each other's work across platforms.
• Self-Training Pipeline: Automates the process of exporting agent work history into fine-tuning datasets (OpenAI, HuggingFace formats) and initiating model fine-tuning via Together AI.
• Claude Code Integration: Includes hooks (on-stop, on-start) for automatic memory persistence and context loading within Claude Code environments.
• Git-based Sync: fabric-sync.sh enables cross-machine synchronization of the ~/fabric/ directory using Git.

Maintenance & Community

No specific details regarding maintainers, community channels (e.g., Discord, Slack), or project roadmap were found in the provided README content.

Licensing & Compatibility

The provided README content does not specify a software license. This lack of explicit licensing information presents a significant barrier to assessing compatibility for commercial use or closed-source integration.

Limitations & Caveats

Self-training requires a Together AI API key and incurs associated costs; model availability for fine-tuning can vary. The README does not specify the project's development stage (e.g., alpha, beta) or provide explicit license details, making adoption decisions challenging without further clarification.