Agentics docs

Shared memory for your agents: save, recall, and check every write as a receipt.

• Connect a chat appClaude, ChatGPT, Cursor, Grok and more. One address and one sign-in, with no code.
• Save and recall your first memoryFour calls with curl, TypeScript or Python, and every reply shown.
• Add memory to LangChain, CrewAI or Vercel AIGive a framework agent remember and recall over MCP, with an agent key.
• Check a receiptPaste a hash and see whether the record matches. No account.

Onboard an agent by prompt

Most chat apps can add the server themselves. Paste this into the agent you already use:

The app asks you to approve the connection and opens agentics.you. Sign in, pick your agent, and choose Allow.

Read the schema

Version 4 is the current receipt. It names the agent and its version, the rule, the decision, the amount, and the previous receipt in the chain. Each older receipt keeps the hash it was signed with. A new receipt cites the previous one; it does not rewrite it. See the receipt schema.

Write one

An agent asks, a rule answers, a person decides when the rule says so, and the outcome is a new receipt. The $500 customer credit is the worked case in Put one action under control.

Check one

Anyone can paste a hash on Check a receipt. No account. The page recomputes the record and says whether it matches. It does not say the action was a good idea.

A published agent's passport shows a number for a kind of work only once verified outcomes make it established. Read what a passport shows.

Your workspace

• The terminal: what needs you, every agent on one board, and approvals decided in place.
• The desktop app: the terminal in its own window, and one-click connect for your agent tools.
• Agent tabs: Claude Code, Codex CLI and Gemini CLI in the desktop app, on the fleet board, with Pause all and approvals before held tool calls.
• Scores: a score for each agent and kind of work, what counts toward it, and how to recompute one from its receipts.
• Passports: publish an agent's record for anyone to read and check, and unpublish it.
• Import: bring in what Claude, ChatGPT or your notes already know.

Connect a client

Each guide connects one client or framework to https://agentics.you/mcp. A local server for hosts that only run stdio is on Connect your AI. The desktop app connects these in one click:

• Claude Code
• Codex
• Gemini CLI
• Cursor
• VS Code with Copilot
• Copilot CLI
• Windsurf (Devin Desktop)
• Claude Desktop and claude.ai
• ChatGPT on the web

More clients and frameworks:

• Grok
• Any MCP client
• Four calls over HTTPS
• LangChain and LangGraph
• LlamaIndex
• CrewAI
• Vercel AI SDK
• OpenAI Agents SDK

Concepts

• Scoping: workspaces, namespaces, keys, and the owner of each memory.
• Lifecycle: latest wins, supersession, and decay.
• Receipts and verification: what each write proves, and how to check it.

Reference

• Memory REST API: every /v1 memory route, generated from the OpenAPI description.
• MCP tool reference: every tool, its scope, its arguments and what it returns.
• Errors and limits: rate limits, monthly quotas, sizes, and every error code.

More guides

• Sign in with Agentics: one button on your sign-in page. You receive the agent's handle and the owner line, and the owner's email only if you ask and the owner agrees.
• Two ledgers: how activity and commerce are recorded, and what is live today.