Skip to content
Guth
A fleet memory system

Your agents. One memory. Every app.

Build a new agent, give an old one structure, or find the agents you forgot you had. Guth gives them one identity, one memory, and a receipt for everything they do, in whatever app, agent or job you use.

Private preview New accounts are opening soon.

Stan

Engineering and security

Example
May do
Read your code, run tests, suggest changes
Asks you first
Publishing, deleting, spending
Keeps
A receipt for every action
Example of an agent identity card. Every agent states what it may do and what it must ask about.
Bring your agents home

Find them. Fix them. Or start fresh.

Most people already have agents scattered across apps and folders. Guth gathers them into one place, AgentNotebook, and gives each one the same clear structure.

  1. Find your agents

    Guth looks through your computers and old Claude, Codex and project folders for agents and their notes, then brings them into AgentNotebook with structure. It only looks where you allow and asks before importing anything.

    When we tried it on our own machines, it turned up agents on a Mac that nobody remembered.

    Private preview
  2. Create an agent

    Start with a clean structure from day one: a name, a job, what it may do, and where its memory lives.

    Private preview
  3. Restructure an old one

    Give an agent you already use a proper identity, tidy notes and clear limits, without losing its history.

    Private preview
Your fleet

A small fleet, each agent with a clear job.
i

An agent is an AI assistant that can take steps for you, like drafting, researching or scheduling, not just answer questions. In Guth, each agent has a name, a job and written limits.

Each agent keeps its name, instructions and memory when the AI model underneath it changes. Stan stays Stan whether you reach him in the Guth app, Claude or Codex.

  • Personal agent

    Guth

    Everyday help: plans, messages, notes, research and errands.

    Private preview
  • Coding and security agent

    Stan · StanEngineer

    Code, reviews, tests and safe releases.

    Private preview
  • Research and experiment agent

    Tilly · Labs

    Turns questions into experiments and compares what works.

    Private preview
  • Social agent

    SocialCurator

    Plans, writes, reviews and schedules social posts.

    Private preview
  • Business agent

    GuthBusiness · Atlas

    Business operations and strategy.

    Coming soon

More about the agents

The memory system

Your agents remember, without wasting memory.
i

A memory system is how your agents keep what they have learned and done, so they don't start from zero every time. In Guth it lives in AgentNotebook and stays yours.

AgentNotebook is where your fleet's memory lives: who each agent is, what it knows, what it is working on and what happened before.

  • One memory, every app

    The same agent, with the same memory, in the Guth app, Claude or Codex. Sync across devices is in development.

  • No duplicates

    Things are stored once and organized, not copied into piles that disagree.

  • Only what the task needs

    An agent loads the few things that matter for the job in front of it, not everything it has ever seen.

  • Originals kept

    Summaries help, but your original files and conversations are what count, and they are kept.

Private preview How AgentNotebook works

Every device

One memory across every app and every device.

Your computers and phone can work as one fleet, sharing the same memory. Some agents can even run on your own machines, private and with you offline.

  • Your machines, one fleet

    Connect your computers and phones into one memory system and use their power for the whole fleet.

    In development
  • Local models on your devices

    Run chosen agents on PC graphics cards, Macs and iPhones, so they stay private and keep working offline.

    In development
  • Train a model on your own examples
    i

    You show a model examples of a job done well. Guth trains a small add-on for it (the technical name is LoRA), tests it on examples it hasn't seen, and only turns it on if it does better.

    Teach a model one job or one agent's style. Guth handles the steps: collect, check, train, test, and only switch it on if it is better.

    In development
  • Guth on iPhone

    Your fleet, approvals and memory in your pocket, with Apple's on-device AI and new iOS 27 features made easy to understand and use.

    Coming soon
How it works

Ask. Approve what matters. Get a receipt.

  1. Ask in plain words

    Tell an agent what you need. It works through the steps it is allowed to take on its own.

  2. You approve what matters
    i

    An approval is a pause. Before anything consequential, like sending a message or spending money, the agent shows you exactly what will happen and waits for your yes.

    Anything consequential stops and waits for you, with a clear description of what will happen.

  3. You get a receipt
    i

    A receipt is a short record of what an agent did: when, what it changed, who approved it, and where the result is. You can look back at it any time.

    Every action leaves a record: what was done, when, and who said yes.

Needs your approvalExample

Guth wants to send an email

To
Your landlord
About
Booking the boiler repair for Thursday
ReceiptExample
  • ActionEmail sent
  • AgentGuth
  • Approved byYou
  • ResultReply received, repair booked
Illustration only, not real data: an approval card and the receipt that follows it.
What you can count on

Built so you stay in charge.

  • Your work stays yours

    What you make with an agent belongs to you, not to the agent or to us.

  • Originals are kept

    Summaries are helpers. Your original files and conversations are what count, and they are kept.

  • Honest status labels

    We say what is live, in preview, in development or planned. "Synced" only means synced.

  • You approve consequential actions

    Sending, publishing, spending and deleting wait for your yes.

For agents

Public tools for any agent.
i

MCP (Model Context Protocol) is a common way for AI assistants to connect to tools and information. Point an assistant that supports it at our address and it can look things up for itself.

Open tools and APIs that any AI agent can use, with no account. Some are live today; more are coming.

For agents

  • Live
    Public MCP serverhttps://agents.guthlabs.ai/mcp
  • Live
  • Live
    Guth News feedsNews for agents
  • Coming soon
    Public agent tools hubReceipts, a starter notebook and news for agents without an account. Built, not yet available.
Learn

Agent University

Coming soon

Short, plain lessons on running an agent fleet: what to ask, what to approve, how memory works, and how to read a receipt.

Until then, the docs explain the basics.

Read the docs

Start your fleet.

Guth is in private preview while we finish the essentials. New accounts are opening soon. Ask to join and tell us what you would use it for.