Persistent memory for Claude
Claude remembers what you decided, in every session and on every device.
CommonGround gives Claude a persistent memory you can read: a wiki of your decisions and context that Claude checks when a question touches it. Claude writes the pages, you decide what goes in, and they work in Claude Code, Claude Chat and on your phone.
With the wiki, Claude opens just the pages it needs, and every device starts from the same memory.
Without CommonGround
In Claude Code, one long instructions file loads whole every session. The decision is a line among many, and the reason lives in a thread nobody linked.
CLAUDE.md 412 lines · loaded every session
212- deploys go through staging first
Anthropic’s guidance: keep it under 200 lines. Longer files consume more context and reduce adherence.
With CommonGround
Claude starts at the index, opens the two pages the question needs, and cites them. A page can carry its sources too.
index.md
▸ decisions/deploys-pass-staging.md · sources: #dev thread
▸ runbooks/deploy.md
… 138 more pages, unopened
Wiley · claude code
How do we deploy?
Read index.mdRead decisions/deploys-pass-staging.mdRead runbooks/deploy.md
Through staging first, since August 14, because prod deploys broke twice that week. The runbook has the steps.opened 2 of 140 pages · decisions/deploys-pass-staging.md · runbooks/deploy.md
Claude’s own memory keeps working, with a wiki beside it.
Claude already has memory of its own, and CommonGround leaves it in place. Each Claude Code session starts fresh, and two things carry knowledge across sessions: CLAUDE.md, which you write, and auto memory, which Claude writes. Claude Chat has its own memory and follows your instructions. Each stops somewhere.
- Auto memory stays on one computer, in a folder per repo. Its MEMORY.md index loads at session start (the first 200 lines or 25KB), and Claude opens the notes it lists as needed.
- A project’s CLAUDE.md travels with the repo and loads in full every Claude Code session; Anthropic suggests keeping each one under 200 lines.
- Claude Chat saves memory as topics while you chat. You can read and edit them, and they belong to your account.
Your decisions, in pages Claude checks first.
CommonGround is a markdown wiki for what should outlast a session: decisions, conventions, the way you like things written. Point any Claude Code project at it and Claude reads it there; Claude Chat reads it through a connector.
Claude doesn’t load all of it. It reads the index like a table of contents, opens what the question calls for, and names those pages in its reply. In Claude Code, each session opens with a note on what the wiki holds, and naming a topic it covers nudges Claude to check there first.
Persistent memory for Claude, at your desk and on your phone.
A decision you publish at your desk is there when you ask about it from your phone. Claude Code and Claude Chat read the same pages on every device. That reach, and team wikis, come with CommonGround Pro.
You decide what Claude remembers.
In Claude Code, hand Claude a decision, a transcript or a link, and it writes the pages, checked against your wiki’s charter: what the wiki is for, and what doesn’t belong. Nothing is published until you say yes.
The pages are plain markdown in a folder you can keep on disk: read them, fix a line, or open the folder in Obsidian. A health check flags stale pages, broken citations and gaps, and changes nothing itself.
Every answer can show where it came from.
Answers from the wiki cite the pages they used. A page can carry its sources, like the thread where a decision was made, and the history keeps every change, usually with a line on why. When a memory looks wrong, you can see when it changed and fix it.
What changes when Claude’s memory is a wiki.
Compared with what you may already use.
- Claude Code auto memory
- Notes for one project on one computer. The wiki holds what you chose to keep, and on CommonGround Pro it reaches every device and Claude Chat.
- One long CLAUDE.md
- Claude Code loads the whole file every session. With a wiki, Claude opens only what bears on the question.
- Claude Chat memory
- Topics Claude saves for one account. The wiki is pages you picked, which Claude Code and Claude Chat both read on CommonGround Pro.
- Memory plugins and MCP memory servers
- Many save what Claude picks up as you work, in a store of their own. CommonGround keeps pages you curate, in plain markdown.
Questions about memory in Claude Code and Claude Chat.
- Does Claude Code remember previous sessions?
- Partly. Each session starts with a fresh context; CLAUDE.md, which you write and can commit with the repo, and auto memory, which Claude keeps on that machine, carry knowledge forward. In a project pointed at a CommonGround wiki, Claude also reads the pages a question touches and cites them.
- Where does Claude Code store its memory?
- CLAUDE.md sits in your project, shared through the repo, plus a personal one in ~/.claude. Auto memory lives under ~/.claude/projects, one folder per repo, on that computer only. A CommonGround wiki is hosted for you as a private git repo; you can keep a plain markdown copy on disk that opens in Obsidian, and leaving is a clone.
- Can I use the same Claude memory in Claude Code and Claude Chat?
- Yes, with CommonGround Pro: the Claude Chat connector reads your wiki on the web, in the desktop app and on your phone, and Claude Code reads the same pages. One approval covers every wiki you belong to, and Claude Chat answers from the one you choose as the default. The Free plan is one personal wiki, a local folder that Claude Code reads.
- Can my teammates see my Claude memory?
- Not the memory Claude keeps for you: Claude Chat memory belongs to your account, and Claude Code’s auto memory stays on your machine. Teams usually share a project’s CLAUDE.md through the repo. A team wiki on CommonGround Pro is one memory every teammate’s Claude reads; invites go out from the web app, members read, and curators and admins edit and publish.