To share memory between AI agents, keep it in a place all of them can read. Delegate does that for you. Tell one agent who you sell to, and Claude Code, ChatGPT, Codex, Cursor and your chats in Delegate Agents know it the next time you open them. You explain your company once.
| What you get | Agents that already know your business when you open them |
| Works in | Delegate Agents, Claude Code, ChatGPT, Codex and Cursor |
| What you write down | Nothing. Agents save what you tell them |
| Where you see it | Toolbox, under Memory and Skills |
Until now it went like this. You tell a chat on Monday that you sell to US fintechs with 50 to 500 employees, and it builds you a decent list. On Tuesday you open Claude Code, ask for the same kind of list, and the first thing it wants to know is who you sell to. Each tool keeps its own notes, if it keeps any, so the more agents you use, the more of your day goes to repeating yourself.
Before
- Tell a chat who you sell to
- Open Claude Code
- you explain it again
Now
- Tell a chat who you sell to
- Open Claude Code
- it already knows
What people keep re-explaining to their agents
We researched what people say in public forums about AI agents that forget between sessions and between tools. The research covered 40 search results, with 8 discussions from the last 90 days read in depth. We ran it in Delegate Agents on November 13, for about $0.12.
- search results read
- 40
- discussions read in depth
- 8
- services used
- 3
- spent in total
- $0.12
Describing the project again turned out to be the small part. What costs people time is rebuilding why a decision was made and where the work stopped. And when they switch tools, each tool holds its own version of the story.
The workarounds they described all take effort. People write a handoff note before they close a chat, keep a CLAUDE.md or AGENTS.md file up to date, or run their own database so that several tools can read the same notes. A few had written wrap-up and pick-up skills to do the handoff for them.
The five steps below take that work off your hands.
Step 1: Connect the agents you use
Connect a tool to Delegate once and it can read your memory from then on. The quickstart has the steps for Claude Code, ChatGPT, Codex and Cursor. Chats in Delegate Agents need no setup at all.
Step 2: Tell one agent about your work
You don't fill in a profile. Mention what matters while you ask for something:
We sell to US fintechs with 50 to 500 employees. Find me ten that raised this quarter.
The agent keeps the first sentence and gets on with the list. A Delegate chat shows "Saved to your memory" under its answer, with a link to what it kept.
Step 3: Open a different agent and ask again
The second agent already knows who you sell to, so it starts on the work. A chat you start in Delegate Agents gets your memory with every run, and a connected agent like Claude Code reads it at the start of a task.
If you change your mind later, say so in any of them. Ask an agent to forget something and every line about it is gone.
Step 4: Save a job that went well as a skill
Sometimes an agent gets a job exactly right. The call brief has the length you wanted and uses the sources you trust. Keep that: ask the agent to save what it just did as a skill, and the next brief follows the same steps without you describing them again.
You choose what becomes a skill, and your agents take it from there. Claude Code and the other agents you connect pick up the skill that fits the task. In a Delegate chat you pick the skills an assistant follows.
Step 5: See what your agents remember
Everything they keep is in Toolbox in plain sentences, each with who saved it and on which day, for example "Claude · 6 Nov". You can edit or remove any line.
5 things people keep re-explaining, and what to say once
These are the five things the research found people repeating, each with a line you might say to your own agent.
- Decisions, and why you made them. "We never email companies with fewer than ten employees, because they have no budget for us."
- Where the work stands. "The Q4 list is checked up to row 120. Enrichment comes next."
- Company facts. "We sell to US fintechs with 50 to 500 employees."
- How you like things done. "Give me lists as a spreadsheet, never as a table in the chat."
- What has changed. "We raised prices on October 1, so quote the new ones." The new line replaces the old one.
A procedure is different. If you catch yourself explaining the same steps again, save them as a skill.
Questions people ask
Do Claude Code and ChatGPT share memory?
Not by themselves, because each keeps its own notes. Once both are connected to Delegate, both read the same memory, so what you told one is there in the other.
What is the difference between memory and a skill?
Memory holds facts about you and your work. A skill holds the steps for a job you want done the same way every time.
Is my memory private?
Yes. Your memory is yours alone, and each teammate has their own.
Which AI agents does it work with?
Chats in Delegate Agents, and Claude Code, ChatGPT, Codex and Cursor once you connect them. They all read the same memory.
Can I take my skills to another tool?
Yes. Each skill downloads as a SKILL.md file in the open Agent Skills format, so you can use it in other tools that read that format.
Toolbox has been open to every Delegate account since November 12. Open Delegate Agents and mention what you sell while you ask for something.
What do you keep having to tell your agent?