Skip to content
agentics.you
Start free

Give your agents a shared memory.

Your AI agents remember. They know their limits. There’s a record of everything they did. And you can verify it yourself.

Start freeConnect an agent
https://agentics.you/mcp

Paste into Claude, ChatGPT, Cursor or Grok.

Shared memory is free forever. No card to start.

memory recallSample

CursorShared memory

184 ms

recall("how do we retry failed payments?")

  1. Scoped

    Cursor’s memory plus the team’s. Three private memories skipped.

    6 ms

  2. Searched

    By keyword and linked memories. 48 found.

    71 ms

  3. Matched

    5 kept and packed into 1,240 tokens.

    64 ms

  4. Kept current

    A fact changed on Oct 12. The newer version wins.

    9 ms

  5. Cited

    Each result carries the receipt it was saved with: 9f3c…2a1b.

    34 ms

Top result

Failed payments retry 3 times with exponential backoff, then page on-call.

Saved by Claude Code · Oct 12Replaces retry 5 times · Sep 30

Verified 9f3c…2a1b
Every memory is saved with a receipt anyone can check. See a receipt checked

Shared memory for any MCP client, and any model through MCP or the REST API.

And any other MCP client. How each one connects · What each one has passed

Counting receipts on the public record…Check any of them

Agent memory

Agent memory that’s scoped, verifiable, and auditable.

One memory for every agent you run, with four qualities built in.

  • Scoped

    Each agent reads only the memory its scope allows, so context stays with the agents meant to have it.

  • Verifiable

    Anyone can check that a memory hasn’t changed since it was written. The check is public and needs no account.

  • Auditable

    Every memory write and rule decision is on the record, so you can see what an agent saved and when.

  • Budget-aware

    Your rules set how much an agent may spend before it needs your approval. Every action is recorded with the agent that took it and the person who approved it.

Verifiable

Every memory carries a receipt.

Every memory write is a receipt, anchored to a public record every 30 minutes. Run the check on a sample receipt. No account needed.

  • Unchanged. The receipt matches the copy written when the agent acted.
  • In sequence. The receipt links to the receipt before it, so none was removed or slipped in.
  • On the public record. The receipt’s batch, sealed every 30 minutes, matches a fingerprint on a public record that agentics.you can’t rewrite.

Counting receipts on the public record…Check a receipt

Sample receipt

Allowed

$42

Accounts agent

Paid the vendor invoice

Invoices at or under $50 are allowed

Four checks against the public record. No account needed.

Sample receipt

Not checked

harbor-invoice-2026-10-03

a3b5a7a7…814d

Run the check to see whether this receipt is genuine.

Agent
Accounts agent
Action
Paid the vendor invoice
Rule that allowed it
Invoices at or under $50 are allowed
Outcome
Invoice paid. The vendor marked it settled.
Workspace
Harbor Studio
Time
  1. Matches its fingerprintWaitingThe contents are the same ones that were written.
  2. Links to the previous receiptWaitingIt names the receipt just before it, and that link holds.
  3. Included in its batchWaitingIt sits in the sealed batch from that window.
  4. On the public recordWaitingThat batch is on a public record we cannot rewrite.
Raw receipt
{
  "receipt_hash": "a3b5a7a7fd61fea0ee05fb832dc94db42eccd544f7c8d358e335a9ce3422814d",
  "receipt": {
    "receipt_id": "harbor-invoice-2026-10-03",
    "batch_id": "harbor-2026-10-03",
    "action": "Paid the vendor invoice",
    "agent": "Accounts agent",
    "agent_version": "v1",
    "agent_version_id": "accounts-agent.v1",
    "amount_usd": 42,
    "channel": "invoice",
    "currency": "USD",
    "decision": "allowed",
    "outcome": "Invoice paid. The vendor marked it settled.",
    "owner": "Harbor Studio",
    "request": "Pay the October paper invoice for $42",
    "rule": "Invoices at or under $50 are allowed",
    "rule_version": "v3",
    "started_at": "2026-10-03T16:04:00.000Z",
    "prev_hash": "50eb099abb8b4c34a72f9c5bb42b4ef13ed8c7b527f0be97cc4ca51c15a90467"
  }
}

Scoped

Memory follows your rules.

Decide what each agent can remember, recall and do on its own, and what waits for your approval. Each decision is written on the receipt, with the rule that made it.

Read the quickstart

Rule: pay invoices

Sample

  • When $50 or less

    Allowed

  • When Over $50

    Needs you

  • When Vendor not on the list

    Denied

How it works

Connect once. Share one memory.

Add one address to Claude, ChatGPT, Cursor or Grok. Every agent draws on the same memory, your rules decide what each agent can do, and every memory leaves a receipt anyone can check.

  1. 01

    Connect each agent once

    Add https://agentics.you/mcp to Claude, ChatGPT, Cursor, Grok or your own client. Sign in once and choose the agent.
  2. 02

    Share memory, hand off work

    What one agent learns, every agent in scope can recall. When work passes between agents, your rules decide what goes ahead and what waits for your approval.
  3. 03

    Check any action

    Every action and every memory is saved as a receipt. Open a receipt in your account, or share its link so anyone can check it without an account.

Setup

Connect in 2 minutes

Pick the app your agent runs in. Paste one line, approve, and sign in. No code.

Steps and tools for each app
  1. 01Open Customize, then Connectors.
  2. 02Choose +, then Add custom connector.
  3. 03Paste the address, choose Add, and sign in when Claude asks.
https://agentics.you/mcp
  1. 01Open Workspace settings, then Apps, then Create.
  2. 02Name the app Agentics and paste the address.
  3. 03Choose sign-in and let ChatGPT scan the tools.
https://agentics.you/mcp
  1. 01Open Cursor Settings, then MCP.
  2. 02Add this to your mcp.json and save.
  3. 03Sign in when Cursor asks.
mcp.json
{  "mcpServers": {    "agentics": { "url": "https://agentics.you/mcp" }  }}
  1. 01Paste this sentence into the chat.
  2. 02Approve the connection when Grok asks.
  3. 03Sign in, pick your agent, and choose Allow.
Add a custom MCP server called agentics at https://agentics.you/mcp

Questions

Common questions

Anything else? Talk to us.

What is agentics.you?
Agent memory that’s scoped, verifiable, and auditable. Your AI agents remember, they know their limits, and there’s a record of everything they did that you can verify yourself.
Which agents can connect?
Claude, ChatGPT, Cursor, Grok, Hermes, OpenClaw, Claude Code, Codex, Gemini CLI, VS Code, Windsurf, or any other MCP client. Each one adds the same address, https://agentics.you/mcp, and you sign in once to pick the agent. LangChain and other server-side frameworks connect through their MCP adapters with an agent key instead of the sign-in.
Do I need to write code?
No. Connecting an agent takes about two minutes, with no code. The docs have the steps for each app.
What happens when an agent wants to do something a rule doesn’t allow?
Your rules decide what goes ahead and what waits for your yes. Anything that waits shows up under Needs you, and the decision is written on the receipt.
Can someone check a receipt without an account?
Yes. Paste a receipt hash on the check page. The check runs against the public record, and no account is needed.
What does it cost?
Free to start, with no card: 10,000 actions a month. Pro is $29 a month and Team is $199 a month. Priced on actions, never per agent. See pricing.

Shared memory is free forever. Paid plans are priced on actions, never per agent.See pricing

Bring us one real workflow.

We work with a small group of teams that run multi-agent workflows. Bring the workflow where your agents lose context, and we build around it.

Become a design partnerStart free