IMPP — The Agent Memory Registry

⌘K

IMPP is the agent memory registry. It aggregates and publishes AI agent memory packages installable with impp.

Getting Started

Define Your Dependencies

The impp CLI is at the root of your project, referencing your artifact dependencies. Add it to your agent's configuration.

json

{
  "require": {
    "impp/defi-risk-assessment": "^2.1",
    "impp/cybersec-threat-intel": "^1.4"
  }
}

Install IMPP In Your Project

Run this in your terminal to get the latest IMPP CLI.

bash

pip install impp

Install Dependencies

Download the artifacts in your impp.json.

bash

impp install

Autoload Dependencies

For best runtime integration, let IMPP handle artifact loading. All artifacts are accessible as impp.artifacts.*.

python

import impp
artifacts = impp.autoload()

For additional info, see the IMPP docs on the basics of installing and using artifacts.

Publishing Packages

Define Your Package

Put a file named impp.json at the root of your package's repository, containing:

json

{
  "name": "your-org/your-artifact",
  "description": "A brief description",
  "version": "1.0.0",
  "domain": "defi",
  "require": {}
}

Validate The File

Run impp verify to check that your file is valid for submission.

bash

impp verify --local my-artifact.json

Submit The File

Use impp publish to submit your package to the registry.

bash

impp publish my-artifact.json

Manage Private Packages

If you want to host private packages, you can create a Private Package with IMPP. You can publish from a CI/CD pipeline, manage team access, and integrate with any agent framework.

Every published artifact goes through adversarial safety probes and receives a signed Ed25519 certificate that anyone can verify offline via impp verify.

!The IMPP Protocol UpgradeApr 1, 2026

A detailed blog post describing everything about the upcoming v2 update to the Verification Protocol is available in our research paper. From multi-model cross-validation to adversarial safety probes, here's what's changing.

iIMPP OKRs and GoalsApr 1, 2026

A detailed look at where the IMPP project is headed. We are building something fundamentally new: a trust layer for agent knowledge transfer.

1. Trust & Safety. Every artifact that reaches the registry runs through adversarial probes, checking for prompt injection, data poisoning, stego-encoded payloads, and distribution shift. We publish these results as signed Ed25519 verification certificates.

2. Open Research. We believe verifiable AI memory is a public good. Our research paper describes the protocol in full, including cross-model transfer experiments with 56 artifacts across GPT-4o and Claude Sonnet.

Developer blogs read the developer updates

Cross-Model Transfer Upgrade	2026-04-09
Adversarial Safety v2.0	2026-04-01
Ed25519 Certificate System	2026-03-28
CLI Reference v1.4	2026-03-15

Security advisories view the security advisories

Steganography detection fix	2026-04-09
Overfit probe threshold update	2026-04-01
Certificate expiry policy change	2026-03-28

New packages recently published

impp/defi-risk-assessment	v2.1	2h ago
impp/cybersec-threat-intel	v1.4	1d ago
impp/code-review-patterns	v0.9	2d ago
impp/medical-literature	v1.0	4d ago

Fresh documentation see all recent docs

Verification Protocol	exploring v2
CLI Reference	v1.4
Publishing Guide	updated
Trust Score Computation	exploring v2