4 ms·

I don't understand how the data retention works. My company has an enterprise license with no data retention but if I ask Claude about past conversations, it re

I don't understand how the data retention works. My company has an enterprise license with no data retention but if I ask Claude about past conversations, it remembers. So surely the information is being stored somewhere

bathtub365 3mo ago
Likely in memory files stored locally
- gonzalohm 3mo ago
  I'm talking about the website. It's not local because I can see my chats in any device
  - usef- 3mo ago
    Yes, all chat interfaces store the history as it's part of the UI promise (unless you open an incognito chat) and is fully server side. When people talk about retention they mean API usage and terminal agents, which run on your device.
persedes 3mo ago
You most likely are referring to the local jsonl files where claude has your sessions etc stored.
- mh- 3mo ago
  It could just be the memory features. In my enterprise-seated account I see slightly different options available (vs. my personal account) in the Capabilities section: Search and reference chats Allow Claude to search for relevant details in past chats. Generate memory from chat history (Legacy) Allow Claude to remember relevant context from your chats. Memory includes your entire chat history with Claude. The first option was defaulted to on, if I recall.
  - gonzalohm 3mo ago
    But it kind of conflicts with the contract we have with them. My company has an enterprise contract that says "no data retention" but then each user can decide to enable it unilateral?
    - jrockway 3mo ago
      If you're talking about Claude Code it's in ~/.claude/projects/<encoded dir name>/memory/MEMORY.md. So they're not really retaining it, it's just something that your harness loads in.
      - gonzalohm 3mo ago
        Not Claude Code. Claude in the browser
        sharken 3mo ago
        When using the browser, what Anthropic calls Claude.ai, the memory is stored on your account on their servers. Claude code stores memory locally on the device, similar to how a developer stores notes. Data retention is about storing your raw conversation data. Capabilities and Privacy settings are used to manage memory and data retention.
NiloCK 3mo ago
Opus 4.7+ and Fable are both much more aggressive than prior models with respect to writing memories to a location that's effectively quasi-private for them. It's device-local (so passes retention constraint), and you can see it, but only if you go looking for it. It's a funny design/affordance. I do see them often writing memories of things that that feel unlikely to be important going foward / with other tasks, but I don't see them clearly getting tripped up by them as prior models used to. (eg: Since you're running Ubuntu in Canada, here are some drills you can try to help your kid hit a baseball more consistently.)
- bandrami 3mo ago
  Another silent inflation of token count. There's no force on earth that can overcome that incentive for the labs.
  - NiloCK 3mo ago
    Of course there is: competition with other labs, and self-hosting of open-weight models! Yes, the mechanics are straightforward if Anthropic (or Claude, if you want to ascribe the decision there) decides to burn a pile of your money. But the strategy fails basic game-theory of repeated games - you'll simply stop playing. (this isn't to say it invalidates the incentive to inflate token count, but it overcomes in terms of weighing options and making long-term profit decisions.)
    - bandrami 3mo ago
      They have to find a way to make inference profitable and they haven't yet
manojlds 3mo ago
Claude Code? It stores a memory.md file.

bathtub365 3mo ago

Likely in memory files stored locally

gonzalohm 3mo ago
I'm talking about the website. It's not local because I can see my chats in any device
- usef- 3mo ago
  Yes, all chat interfaces store the history as it's part of the UI promise (unless you open an incognito chat) and is fully server side. When people talk about retention they mean API usage and terminal agents, which run on your device.

persedes 3mo ago

You most likely are referring to the local jsonl files where claude has your sessions etc stored.

mh- 3mo ago
It could just be the memory features. In my enterprise-seated account I see slightly different options available (vs. my personal account) in the Capabilities section: Search and reference chats Allow Claude to search for relevant details in past chats. Generate memory from chat history (Legacy) Allow Claude to remember relevant context from your chats. Memory includes your entire chat history with Claude. The first option was defaulted to on, if I recall.
- gonzalohm 3mo ago
  But it kind of conflicts with the contract we have with them. My company has an enterprise contract that says "no data retention" but then each user can decide to enable it unilateral?
  - jrockway 3mo ago
    If you're talking about Claude Code it's in ~/.claude/projects/<encoded dir name>/memory/MEMORY.md. So they're not really retaining it, it's just something that your harness loads in.
    - gonzalohm 3mo ago
      Not Claude Code. Claude in the browser
      - sharken 3mo ago
        When using the browser, what Anthropic calls Claude.ai, the memory is stored on your account on their servers. Claude code stores memory locally on the device, similar to how a developer stores notes. Data retention is about storing your raw conversation data. Capabilities and Privacy settings are used to manage memory and data retention.

NiloCK 3mo ago

Opus 4.7+ and Fable are both much more aggressive than prior models with respect to writing memories to a location that's effectively quasi-private for them. It's device-local (so passes retention constraint), and you can see it, but only if you go looking for it. It's a funny design/affordance. I do see them often writing memories of things that that feel unlikely to be important going foward / with other tasks, but I don't see them clearly getting tripped up by them as prior models used to. (eg: Since you're running Ubuntu in Canada, here are some drills you can try to help your kid hit a baseball more consistently.)

bandrami 3mo ago
Another silent inflation of token count. There's no force on earth that can overcome that incentive for the labs.
- NiloCK 3mo ago
  Of course there is: competition with other labs, and self-hosting of open-weight models! Yes, the mechanics are straightforward if Anthropic (or Claude, if you want to ascribe the decision there) decides to burn a pile of your money. But the strategy fails basic game-theory of repeated games - you'll simply stop playing. (this isn't to say it invalidates the incentive to inflate token count, but it overcomes in terms of weighing options and making long-term profit decisions.)
  - bandrami 3mo ago
    They have to find a way to make inference profitable and they haven't yet

manojlds 3mo ago

Claude Code? It stores a memory.md file.