Short answer: put your company in one folder of plain files, and make every Claude session read a short index file in that folder before it does anything else. Claude's built-in memory helps with you as a person. It doesn't hold a checked, dated record of your company, and it doesn't follow you across every way you use Claude.
What Claude's built-in memory covers
Claude can search your past chats and keep a memory of you, built from your conversations [2]. That's useful for your preferences and the work in flight. It has three limits for a business:
- It's a summary the model writes, not a record you wrote and can check.
- It lives in separate places. A Project keeps its own instructions and files. Claude Code keeps its own memory files [3]. A teammate's Claude knows nothing about yours.
- It isn't always read. In Anthropic's tests, Claude Sonnet 4.6 read its memory before answering less than half the time, and prompting a model to read its memory first nearly doubled its benchmark score [1].
What works: one folder, read first
Anthropic's own numbers show why it's worth the effort: on their benchmark, Claude Opus 5.5 scored 0.66 without memory and 0.84 with it [1].
- A short index file Claude reads first. Who the company is, how you write, and where everything else lives. Keep it short: Claude Code loads only the first 200 lines of its memory index [3]. Ours is 39 lines.
- One file per subject. A customer, a product, a price, a tool. Each fact carries the date it was checked.
- Decisions in their own files. When you change a price or a rule, write the decision down with the date, and strike the old line instead of deleting it.
- Procedures as files. How you publish, how you onboard a customer, how you write a proposal.
- The same folder everywhere. Point Projects, Cowork and Claude Code at it, so every session starts from the same facts.
How to keep it from going stale
Memory that's never cleaned goes wrong quietly. In an early version of our setup, one agent's task file grew to 140 KB with 174 open items, and the agent only worked on what sat near the top. Two habits fix most of it: date every fact, and review the folder on a schedule, so changed facts get struck and replaced.
Do it yourself, or have it done
You can build this by hand in an afternoon and keep it up yourself. Or Chet can do it for you: Chet sets up the folder (we call it Start) from your own documents, grades your Claude setup, and fixes what breaks. $99 a month, cancel anytime.
Maximize Claude with Chet: https://www.coworkers.global/get-chet?utm_source=blog&utm_medium=internal&utm_campaign=chet&utm_content=how-to-make-claude-remember-your-company-cta
FAQ
Does Claude remember previous conversations? It can search past chats and keep a memory of you [2]. That memory is a model-written summary, scoped to where you use Claude.
Do Claude Projects remember my company? A Project holds instructions and files for the chats inside it. Other Projects, Claude Code and teammates don't see them.
How long should the index file be? As short as you can make it. Claude Code reads only the first 200 lines of its memory index [3]; we keep ours under 40.
Sources
- Anthropic, components of agent memory: https://anthropic.com/research/components-of-agent-memory
- Claude Help Center, chat search and memory: https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context
- Claude Code docs, memory: https://code.claude.com/docs/en/memory
Regards,
Charles Stack
Founder, Coworkers.Global