In the ongoing quest to ground AI agents in verifiable truth, a new open-source project called aafp-commons has emerged on GitHub. Developed by user davidnichols-ops, the project provides a local, signed notebook specifically designed for AI agents. The repository, which surfaced on Hacker News on September 6, 2026, introduces a dual-interface approach, offering both Command Line Interface (CLI) tools and Model Context Protocol (MCP) servers. This allows agents to interact with a cryptographically secure, local data store without relying on external cloud services.

The Shift to Local Verification

The core innovation here is the concept of a "signed notebook." Unlike traditional vector databases or ephemeral memory buffers that AI agents often use, this system emphasizes integrity and provenance. By signing the notebook locally, the project ensures that the data an agent reads or writes can be verified for tampering. This is a critical step for autonomous agents that need to trust their own memory and context windows, especially when operating in environments where data sovereignty and audit trails are paramount. The local nature of the storage also mitigates latency issues and privacy concerns associated with sending agent context to remote APIs.

CLI and MCP Integration

The project exposes its functionality through two distinct interfaces. The CLI allows developers and power users to interact with the notebook directly from the terminal, enabling manual inspection, signing, and management of the agent's data. More importantly, the MCP integration allows modern AI agentsβ€”particularly those built on frameworks that support the Model Context Protocolβ€”to connect to the notebook as a standard tool. This means agents can autonomously retrieve context, append new notes, and verify signatures as part of their reasoning loop, effectively giving them a tamper-evident memory bank.

Key Takeaways

  • Local First: The notebook resides on the user's machine, ensuring data privacy and reducing reliance on cloud infrastructure.
  • Cryptographic Integrity: Entries are signed, providing a mechanism to verify that the agent's context has not been altered maliciously or accidentally.
  • Dual Access: The project supports both manual interaction via CLI and automated access via MCP, catering to both human developers and autonomous agents.
  • Open Source: Hosted on GitHub, the project invites community contribution and transparency in how agent memory is structured and secured.

The Bottom Line

As AI agents become more autonomous, the trustworthiness of their memory is becoming as critical as their reasoning capabilities. aafp-commons addresses a glaring gap in the current agent stack: verifiable, local state. While still early-stage, this approach could become the standard for agents that need to maintain a long-term, auditable context without leaking data to the cloud.