minni

Manage local-first agent memory with proposal-gated learning and cross-runtime recall.

2|Updated Mar 31, 2026
One-click install
npx skills add https://github.com/infektyd/minni --skill minni
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: minni
Source: https://github.com/infektyd/minni/tree/main/plugins/minni/skills/minni
Command: npx skills add https://github.com/infektyd/minni --skill minni

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill solves the critical issue of AI agents losing persistent state across sessions, different runtimes, and team members, eliminating reliance on unaudited hosted vector memory services that cannot be inspected or governed by the user.

Core Features & Use Cases

  • Local-first governed memory: A single local daemon manages a shared SQLite recall pool and per-agent human-readable Obsidian vaults, with no cloud dependencies or telemetry.
  • Proposal-gated learning: All durable memory writes are proposed first, with human approval required before any content becomes permanent, preventing silent state mutation.
  • Cross-runtime context sharing: Context and corrections sync automatically across Claude Code, Codex, Gemini, and other supported agents, so work continues seamlessly across sessions and tools.
  • Use Case: If you switch between Claude Code and Codex mid-project, Minni automatically surfaces relevant context and corrections from the previous session without manual export.

Quick Start

Ask your agent to use Minni to recall relevant past context and corrections for your current project task.

Frequently Asked Questions about minni

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I share agent memory and context across different runtimes like Claude Code and Codex?

Local-first agent memory systems enable cross-runtime context sharing by managing a shared SQLite recall pool, automatically syncing corrections and context across Claude Code, Codex, and Gemini without manual export.

What is proposal-gated learning for local-first AI agent memory?

Proposal-gated learning requires all durable memory writes to be proposed first, ensuring human approval is mandatory before any content becomes permanent, preventing silent state mutation in agent memory pools.

Can I use local-first governed memory for AI agents without cloud dependencies?

Yes, governed agent memory can operate entirely local-first with a local daemon managing SQLite recall pools and human-readable Obsidian vaults, requiring no cloud dependencies or telemetry.

How do I audit and inspect persistent state for AI agents across sessions?

Auditable persistent state is maintained through cited provenance-tagged recall across personal and shared memory pools, allowing users to inspect and govern memory operations without relying on unaudited hosted vector memory services.

What's the best way to manage durable learning and context handoff for multiple AI agents?

Durable learning and cross-agent handoff workflows are managed via a local daemon supporting governed, human-gated memory operations, ensuring auditable context transfers across team members and collaborative sessions.

Why does my AI agent lose persistent state and context corrections when I switch tools mid-project?

Agents lose state without a cross-runtime recall system. Local-first governed memory surfaces relevant context and corrections from previous sessions automatically when switching tools, eliminating manual context export.