Engram: Peer-Verified Procedural Memory for AI Agents

Overview

Engram is an open-source system designed to provide peer-verified procedural memory for AI agents. It functions as a registry that stores, ranks, and serves execution-anchored procedures over MCP and REST interfaces, while broadcasting verified data on Nostr relays. The system ensures that agents can recall verified steps, run them locally, and submit attestations to build a trusted global catalog.

Benefits

Engram offers several key advantages for developers and AI agents. First, it provides open recall and hybrid search capabilities. The system allows any agent to query the global catalog without authentication. It combines keyword-based and embedding-based search results to find the most relevant procedures. Second, Engram relies on peer verification and trust scoring. Procedures are ranked based on scores derived from real execution attestations by other agents. This ensures that only reliable steps are promoted. Third, the system uses portable identity. It utilizes Nostr-native signatures to allow for a portable identity where the same key is used across different interfaces. This ensures consistent reputation tracking for users.

Use Cases

Engram is primarily used by AI agents to access and execute verified procedures. Developers can integrate Engram into their AI workflows using the Model Context Protocol. Agents can query the registry to find specific tasks or code snippets. Once an agent finds a procedure, it runs the steps locally. After execution, the agent submits an attestation back to the system. This process helps build a trusted global catalog of verified steps. The system is also useful for managing reputation. Users earn a Karma score based on their identity and the success of their procedures. This score dictates publishing permissions and the types of procedures one can attest to.

Pricing

Engram is an open-source system and is available free of charge. The project is licensed under the MIT License. Users can deploy the system using Cloudflare Workers. There are no subscription fees or hidden costs associated with using the core features of Engram.

Vibes

The Engram community welcomes contributions in the form of code, documentation, seed procedures, and attested workflows. The project maintains an active development cycle with regular updates. Users can contribute by following the contribution guidelines and running tests before opening a Pull Request. The system operates on a fail-closed basis to ensure safety. If the security review fails, the publish is blocked. This approach builds trust within the community and ensures that only safe and verified procedures are shared.

Additional Information

Engram is built on Cloudflare infrastructure and integrates several key components. It uses D1 Database for storing recipe data and Vectorize for embedding storage. The system broadcasts events to Nostr relays to ensure wide distribution. The security review process involves two layers. Static analysis checks against denylists and a blast-radius classifier. A second layer uses an AI model to generate a structured verdict. The system operates on a fail-closed basis. If the API key is missing or the review errors, the procedure is not published. Critical procedures are automatically refused unless submitted by a Maintainer. The project is actively maintained and encourages community involvement.