Does Claude Keep Context Between Chats? How to Make It Stick
What Claude carries from one chat to the next, how to set up the rest, and how to start a chat with no context at all.

Claude keeps some context between chats, but not the chat itself. Every new conversation starts with an empty context window; what carries over is your memory, your Instructions for Claude, and, inside a Project, that Project's instructions and files.
So if you're re-explaining your role and your project every morning, one of those layers is empty or switched off. Here's what each one does, how to set it up, and how to get the opposite: a chat with no previous context.
Does Claude keep context between chats?
Partly: the messages of one chat never load into another, but four features persist across chats. Here's what each one carries:
| Layer | What carries over | Scope | Plans |
|---|---|---|---|
| Memory | Facts Claude saved as topics from past chats | All chats (Projects get their own too) | All plans; on by default for Free, Pro, Max |
| Chat search | Past chats Claude can look up when you ask | All chats | Paid plans |
| Instructions for Claude | Standing instructions you wrote | Every conversation | All plans |
| Projects | Project instructions and uploaded files | Chats inside that Project | All plans; Free is limited to five Projects |
Sources: Anthropic's guides to memory and chat search, personalization and Projects.
Older guides say Claude forgets everything, and inside a Project too. That was true until memory reached every plan, including free, on March 2, 2026 (Anthropic release notes). The details of how memory works now are in the Claude memory guide.
How do you make Claude remember you across chats?
Set up the four layers in this order, from the ones you write yourself to the ones that run on their own.
1. Write your Instructions for Claude
Instructions for Claude is a settings field whose contents apply to all your conversations. It's the fastest win and the one most people leave empty.
- Click your initials in the lower-left corner and open Settings.
- Find the Instructions for Claude field.
- Write your role, how you want answers, and what Claude should stop doing.
- Save.
Keep it to stable, universal facts. "I'm a backend engineer, Go and Postgres, answer first and explain after" belongs here; this sprint's priorities don't. A full walkthrough with a template is in how to set up Claude preferences.
2. Use a Project for each body of work
A Project gives a set of chats shared instructions, shared files and its own memory. Create one per client, product or long-running job:
- In the sidebar, open Projects and create a new one.
- Add project instructions: the stack, the client, the goal, the constraints.
- Upload the documents Claude should know: briefs, specs, style guides.
Every chat inside the Project starts with that context, layered on top of your Instructions for Claude, so you don't repeat the general stuff. Inside a Project, Claude uses that Project's memory plus your general memory, never another Project's.
3. Leave memory on
Memory saves facts from your chats as topics and reads them in new ones. Check it's on in Settings > Your Claude ("Memory from chats"; Settings > Memory and "Generate memory from chats" in the older layout), and look at the Topics list once in a while to delete what's no longer true. You can also tell Claude "remember this" mid-chat.
4. Point Claude at an earlier chat
On paid plans, Claude can search your past chats. Ask "look at our chat about the pricing page last week and pick up from there" and it pulls the relevant conversation in. On Free, or when a chat has grown long, ask Claude to summarize the decisions so far, then paste that summary into a new chat. A fresh chat with a tight summary usually beats a 200-message thread, because long conversations degrade answers (context rot).
How do you get Claude to respond with no previous context?
Start an incognito chat with the ghost icon in the top right corner. Incognito chats aren't saved to memory or chat history, and Claude doesn't pull from them when it searches past chats.
Two other options:
- One chat without memory: click + before sending your first message and switch off Memory. You can't change it once the chat starts.
- A Project walled off from everything else: turn off "Use account memory" on the Project's page.
Instructions for Claude apply to all your conversations, so if you want an answer free of your usual framing, clear or shorten that field too.
What doesn't work well?
Three habits undo most of the setup above:
- Overloading the instructions field. It loads into every conversation, so a page of rules dilutes the three that matter. Task-specific context belongs in a Project.
- Leaving stale context in place. Instructions describing a project you finished in spring get applied confidently to everything you do now. Stale context is often worse than none.
- Living in one endless chat. It feels like persistence, but the window fills, early details fade, and every reply costs more tokens.
If you work in Claude Code rather than the Claude app, context persists through CLAUDE.md and auto memory instead; see how to give Claude Code context.
Why isn't that enough if you use several AI tools?
Because all four layers live inside your Claude account. Set them up perfectly and you get a Claude that knows you, next to a ChatGPT and a Cursor that know nothing, each with its own instructions field to fill and its own memory to train.
A portable context layer moves the foundation out of any one tool. Unabyss builds your personal context from Gmail, Google Calendar, Notion, Slack and GitHub, keeps it current, and serves it to Claude, ChatGPT and Cursor over MCP. In an August 2026 Unabyss benchmark, that structured context cut cost by 36% and time by 38% compared with connected tools alone (What Structured AI Context Is Worth). Keep Claude's own layers for Claude-specific tuning, and let the shared context cover who you are and what you're working on.
→ See how Unabyss gives every AI tool the same context