matriosha 2.0.4

Matriosha is a Python CLI for an encrypted, auditable AI context engine.

Matriosha

Encrypted, auditable, model-agnostic, local-first AI memory for agents.

Hold AI accountable. Own your data.

Install

Matriosha supports Python 3.11, 3.12, 3.13, and 3.14.

python3 -m pip install matriosha

If your environment already points pip to Python 3.11+, this also works:

pip install matriosha

Quickstart

Initialize an encrypted local vault:

matriosha --mode local vault init

Remember text:

matriosha --mode local memory remember "Important launch context" --tag launch

Remember a file:

matriosha --mode local memory remember --file ~/Documents/agent-notes/launch-context.md --tag launch

Search semantically:

matriosha --mode local memory search "What is the launch motto?"

Verify the audit trail and vault integrity:

matriosha audit verify
matriosha vault verify
matriosha vault verify --deep

Connect an agent locally

Generate a local-only token:

matriosha token generate my-agent --local --scope write --expires 30d

Connect a local desktop agent:

matriosha agent connect --local --name my-agent --kind desktop --token <token>

List local agents:

matriosha agent list --local

Treat generated tokens like passwords. Tokens are shown once.

Optional custom vault location

Set MATRIOSHA_HOME before initializing a vault if you want a predictable local memory directory:

export MATRIOSHA_HOME=./memory
matriosha --mode local vault init

Local and managed modes

Local mode is offline-first and does not require authentication.

Managed mode is optional and is designed for cloud-backed operational workflows such as login, workspace, sync, policy, quota, token workflows, and agent workflows.

matriosha auth login
matriosha --mode managed status

For local-only token and agent setup, use --local on token and agent commands.

Why Matriosha?

AI agents are gaining longer memories, but many memory systems are opaque, vendor-bound, or difficult to verify.

Matriosha keeps AI context outside the model provider and makes it:

• Encrypted: vault data is protected with modern cryptography.
• Auditable: local audit events and vault integrity can be verified.
• Model-agnostic: memory is not tied to one AI vendor.
• Local-first: start offline, then opt into managed workflows when needed.
• Agent-ready: local, managed, desktop, server, and CI agent workflows are supported.

Security model

• Argon2id passphrase hardening
• AES-256-GCM authenticated encryption
• SHA-256 and Merkle-root integrity checks
• Local audit verification
• Ed25519 signature-ready workflows
• Local-only tokens for local agent access

Requirements

• Python >=3.11,<3.15
• A Unix-like shell for the examples, such as Terminal on macOS, Linux shells, or WSL/Git Bash on Windows

Agent setup guide

For interactive VM or local-machine setup, use the public JSON guide:

• https://matriosha.in/assets/matriosha_agent_setup_guide.json

This guide is designed so AI agents can help users install, initialize, verify, and connect Matriosha safely.

Project links

• Homepage: https://matriosha.in
• GitHub: https://github.com/drizzoai-afk/matriosha
• Issues: https://github.com/drizzoai-afk/matriosha/issues
• PyPI: https://pypi.org/project/matriosha/
• License: BSD 3-Clause