emem

emem is designed as the external memory layer for multi-agent systems, enabling disparate agents to cite and verify shared signed facts, even when these agents operate independently without shared models, vendors, or trust. By anchoring facts to real-world observations, emem guarantees verifiable memory that persists beyond individual agents or sessions.

Key Features

• Shared Memory for Agents: Facilitate effective communication between agents that do not share infrastructure or trust boundaries. Each signed observation can be independently verified by any agent, ensuring the authenticity of the data.

• Physical World Foundation: emem leverages Earth observation data as its foundational substrate, allowing various sensors, drones, and observational devices to contribute to a unified memory. Each fact is assigned a permanent address derived from its own content, making it universally accessible across all agent models.

Why emem Matters

Without emem, agents are forced to rely on paraphrased or approximated information after contexts are compacted, which can lead to decisions made on inaccurate data. Here are key scenarios where emem shines:

• Preservation of Facts: When an agent verifies a value, it retains a single line representing that fact, which can later be resolved back to the original observation, eliminating the loss of precision inherent in summarization.
• Efficiency Across Agents: Instead of duplicating efforts, agents can reference existing verified facts, fostering a collaborative environment where precision is key, and no one needs to trust others blindly.
• Auditability: As each fact is recorded with its provenance, it provides a trail for auditing long after the data was first gathered.

Example Usage

To retrieve a verified memory of the altitude of Bengaluru, use the following command:

curl -s -X POST https://emem.dev/v1/recall \
  -H 'content-type: application/json' \
  -d '{"place":"Bengaluru","bands":["copdem30m.elevation_mean"]}'

This command fetches the signed record for elevation, ensuring that all agents receiving this data can verify it independently.

Understand the Token Grammar

The unique addressing system within emem includes various token shapes:

• emem:fact: A signed observation at a specific location.
• emem:bundle: A collection of facts represented as a single handle.
• emem:entity: A canonical identity for co-referencing objects across agents.

When to Utilize emem

• Situations where facts must persist beyond transient contexts or be validated without relying on the original source.
• Data that is critical for decision-making processes needing a high level of trust and verification.
• Applications involving collaboration between different systems or organizations where data integrity is paramount.

In conclusion, emem revolutionizes how agents interact with facts by providing a robust, verifiable memory system that stands the test of time and trust boundaries. For more insights, tools, and integration guidance, visit emem.dev and explore the possibilities.