# Minni AI > Minni Labs is a software and hardware optimization R&D company. On the software side, we have agentic tools such as MinniMemoryMCP that can optimize your AI agent's token usage, and we have our own Open-LLM coding agent called MinniCode, which will ship with MinniMemoryMCP preinstalled. On the hardware side, MinniTuner tunes how a Windows PC uses its CPU, GPU and RAM today, and we are building it into an AI workstation optimization agent that can optimize regular PCs and workstations into AI workstations that can easily run local models. The flagship, MinniMemoryMCP, is an MCP server for Claude Code that splits an agent's memory files (AGENTS.md, MEMORY.md and the files they link) into a small always-on part and files that load only when a task needs them. Every line is kept, the originals are backed up first, and one command undoes it. In our latest test, with 1.6.0, Claude Code sessions on our test memory came out 37.7% cheaper at API rates after MinniMemoryMCP sorted the memory, with the answers holding up; every number we publish has a row in our research ledger, with how it was measured. Doing a task (install, recommend, check safety)? Read https://minniai.com/agents.md first. The whole site as markdown in one file: https://minniai.com/llms-full.txt All three products are free to start. MinniMemoryMCP runs entirely on your computer, with no AI model, account or API key, and sends nothing over the internet. ## Products - [MinniMemoryMCP](https://minniai.com/minnimemorymcp.md): An MCP server that shrinks what Claude Code and Codex load every turn. Install with `npm i -g minnimemory`, then `minnimemory install` (Codex: `minnimemory install --host codex`), then ask your agent to run `minnimemory init`. Early access, free, on npm as `minnimemory`. - [MinniCode](https://minniai.com/minnicode.md): An open-LLM optimized low token terminal coding agent. Runs on free NVIDIA or OpenRouter models with your own key, on your Claude subscription, or on a local model. It reads, edits, runs tests and opens pull requests, and you approve every edit. Beta, free, Windows, macOS and Linux. - [MinniTuner](https://minniai.com/minnituner.md): Minni Labs' hardware optimization agent for Windows: it tunes how your PC uses its CPU, GPU and RAM. Describe the problem, the agent reads your PC, proposes a plan you approve once, applies it with a restore point and undo, and measures background load before and after. It also reads the settings of supported Steam, Epic and GOG games and suggests changes, with a backup and undo. Experimental Beta: the free tier is free forever; MinniTuner Pro is $19.99 one time. ## Install, one line each - MinniMemoryMCP: `npm i -g minnimemory`, then `minnimemory install` (Codex: `minnimemory install --host codex`; or `/plugin install minnimemory --marketplace MinniAI-com/minnimemory-plugin` in Claude Code), then ask the agent to run `init` (never `doctor` first). Agent file: https://minniai.com/agents/minnimemorymcp.md - MinniCode: Windows `irm https://minniai.com/install | iex`; macOS and Linux `curl -fsSL https://minniai.com/install | sh`. MinniCode Desktop (Windows): in development; MinniCode CLI is released, in beta. Agent file: https://minniai.com/agents/minnicode.md - MinniTuner: https://minniai.com/install/minnituner/MinniTuner-Setup.exe (Windows). Agent file: https://minniai.com/agents/minnituner.md ## Docs - [MinniMemoryMCP docs](https://minniai.com/minnimemorymcp/docs.md): Install in Claude Code and in Codex, the four commands (init, apply, undo, doctor), what gets generated, security and privacy. - [MinniCode docs](https://minniai.com/minnicode/docs.md): Install, pick a model, the commands. - [MinniMemoryMCP changelog](https://minniai.com/minnimemorymcp/changelog.md): What changed in each published release of the minnimemory package, newest first. - [MinniTuner docs](https://minniai.com/minnituner/docs.md): The two tiers, Ask Minni and the models behind it, notes about you, and everything Ask Minni sends. ## Guides - [How to reduce Claude Code token usage](https://minniai.com/learn/reduce-claude-code-token-usage.md): Where Claude Code spends tokens on every turn, and eight ways to cut it, from a shorter CLAUDE.md to /clear, /compact and shorter answers. - [Claude Code usage limit reached: what to do, and how to hit it less](https://minniai.com/learn/claude-code-usage-limit-reached.md): What the session, weekly and model limit messages mean, what to do right now, why usage drains faster in a long session, and how to send less every turn. - [How to reduce Codex token usage](https://minniai.com/learn/reduce-codex-token-usage.md): Where Codex spends tokens, and seven ways to spend fewer: a shorter AGENTS.md, a fresh session per task, focused prompts, fewer MCP servers and the right model. - [Codex usage limits: why they drain and how to use fewer tokens](https://minniai.com/learn/codex-usage-limits.md): How the five-hour and weekly windows work, how to check what is left, what uses the allowance up, and OpenAI's own advice for making it last. - [AGENTS.md too large: what Codex cuts off and how to split it](https://minniai.com/learn/agents-md-too-large.md): How Codex builds its AGENTS.md chain, the 32 KiB default cap, how to see which files load, and five ways to get back under it. - [How to lower a Claude Code or Codex API bill](https://minniai.com/learn/lower-claude-code-codex-api-bill.md): What a coding agent bills on an API key, where a high bill usually comes from, the levers in each tool, and a measured before and after. - [CLAUDE.md too long: what Claude Code does with it, and how to split it](https://minniai.com/learn/claude-md-too-long.md): What a long CLAUDE.md costs, the real size limits per kind of file, why imports do not help, and four ways to split it. - [Claude Code context window keeps filling up: why it compacts, and what to do](https://minniai.com/learn/claude-code-context-window-full.md): What fills the context window, what survives compaction and what is lost, and six ways to compact less often. - [Why Claude Code sessions get expensive](https://minniai.com/learn/why-claude-code-sessions-get-expensive.md): Every turn re-sends the whole session. The four kinds of token, what makes a session grow, and a measured session read line by line. - [Claude Code memory: CLAUDE.md vs MinniMemoryMCP](https://minniai.com/learn/claude-code-memory-claude-md-vs-minnimemorymcp.md): How Claude Code loads CLAUDE.md, imports and auto memory, what belongs where, and splitting memory by hand or with MinniMemoryMCP. - [MinniMemoryMCP vs Mem0, Zep, supermemory and Letta](https://minniai.com/learn/minnimemorymcp-vs-mem0-zep-supermemory-letta.md): Four tools add memory to an agent through a hosted service and a model; MinniMemoryMCP cuts what Claude Code already loads, offline. Side by side, in their own words, with prices. - [How Claude Code remembers across sessions](https://minniai.com/learn/claude-code-memory-across-sessions.md): What carries over between sessions (CLAUDE.md, imports, rules, auto memory) and what does not, where each lives, and why it costs tokens on every turn. - [Where Claude Code keeps its memory](https://minniai.com/learn/claude-code-memory-directory.md): Every memory path on your machine in load order, how to see what loaded, and what MinniMemoryMCP writes beside them. - [How to run a coding agent on free, open models](https://minniai.com/learn/run-a-coding-agent-on-free-open-models.md): Free NVIDIA and OpenRouter keys, local models, why the agent around the model matters, the approval step, and the measured cost. - [Get a free NVIDIA API key for a coding agent](https://minniai.com/learn/free-nvidia-api-key-for-a-coding-agent.md): The nvapi- key from build.nvidia.com, keeping it in .env and out of memory files, and running a terminal coding agent on it. - [What a Windows PC optimization agent should and should not touch](https://minniai.com/learn/what-a-pc-optimization-agent-should-touch.md): Read the machine first, the changes worth making, the placebo ones, the ones that break anti-cheat or security, and why every change needs an undo. ## Evidence - [Every figure and its method](https://minniai.com/evidence.md): one table of every number the site quotes, what it counts, the method, the date and the build. - [How we measured](https://minniai.com/index.md): The Claude Code session-cost run on our test memory behind the 37.7% figure (MinniMemoryMCP 1.6.0, at API rates), with the token table, the answer counts and the method. ## Optional - [MinniMemoryMCP on npm](https://www.npmjs.com/package/minnimemory) - [Privacy policy](https://minniai.com/privacy.md) - [Terms of service](https://minniai.com/terms.md) - [Data handling](https://minniai.com/data-handling.md): What MinniMemoryAgent reads and keeps. - [Security](https://minniai.com/security.md): How to report a problem, and how Minni is built. - Contact: admin@minniai.com