DeepSeek Imposes Limits on Old Long Conversations; Summary of Chat History Migration Tools
Since November 15, the DeepSeek App and web version have gradually imposed limits on long conversations that have been inactive for an extended period. Some users have received a "insufficient computing power" notice, and long conversations inactive for more than 72 hours are temporarily unavailable, while others see a limit of 7 days. Old chat windows can still be opened and past Q&A reviewed, but messages can no longer be sent, and a pop-up prompts users that they can export their history for use in other apps. The article summarizes migration and backup solutions including the official Chat Demo, DeepSeek Harness (DSH), the community plugin dsh-chat-migration, the browser extension DeepShare, and the open-source project deepseek-keeper, and recommends first downloading the original records from the data management page as a backup.
Why read this
AI score 60
The original article outlines the conditions that trigger the limits and the differences among several migration tools, helping affected users decide how to preserve the context of their old conversations.
Topics
Body · AI translation
Two days later, a group of users discovered that their old DeepSeek conversations had suddenly become impossible to continue.
Starting November 15, usage-limit notices began appearing on the app and web versions. Some users received a "insufficient compute" message, with long conversations inactive for more than 72 hours temporarily unavailable, while others saw a limit of 7 days.
The pop-up also included a line: "You can export your history to use in other apps."
By today, reports had also clustered around limits such as roughly 40,000 characters per single input and suspension of use once the daily quota is reached.
The chat history is still there, but continuing the conversation is a hassle
A Phoenix Tech search of online posts found that most users currently report that old DeepSeek chat windows can still be opened and previous questions and answers can still be browsed, but messages can no longer be sent.
In Reddit's DeepSeek community, one user said that multiple chat windows they had accumulated were restricted, with only a few still usable.
Another user began looking for ways to transfer their history. He first exported a ZIP from DeepSeek, then had ChatGPT help split it into JSON files, attempting to migrate the original chat content over.
But exporting files is not the same as restoring a conversation.
The original chat history was too long, and the new platform could not accept all of it at once. That user could only rely on other models to repeatedly organize it, compressing the history in segments, extracting key information, and then gradually placing it into new chat windows.
Other users ran into an even more awkward situation:
They wanted DeepSeek to help summarize what had been discussed before as background material for a new conversation, only to find that the old window could no longer reply, making even this last step impossible.
A Phoenix Tech check of its own DeepSeek found that, as of publication, conversation records from January 2025 could still send messages normally and receive corresponding contextual replies.
For example, when using DeepSeek to revise an article continuously, a user may have already spent several days telling it which expressions cannot be used and which paragraphs have been finalized. Starting over in a different window means these requirements have to be reorganized.
Although chat history can be exported, getting another model to fully absorb the context and continue the previous work is not so easy.
Why has DeepSeek started imposing limits on long conversations?
DeepSeek has had multiple service anomalies in the past two months.
There is a detail in this limit: some of those affected are long conversations that have not been used for some time.
From a technical perspective, this may be related to the computational cost of long contexts.
A chat window used continuously for several months may have accumulated dozens or even hundreds of rounds of questions and answers. If a user reopens the window and continues asking questions, how much historical content the model needs to process depends on how the system manages context.
DeepSeek has previously adopted context caching technology, which can reduce the computational overhead of repeated content. But the cache does not persist forever. If a long conversation has not been used for a long time, after the original cache expires, processing a large amount of historical messages again may incur higher costs.
This also makes "inactive for a long time" a noteworthy restriction condition.
However, DeepSeek has not yet announced the specific technical reason for this limit.
Why some users see 72 hours while others see 7 days, and whether this is also related to conversation length or account usage, has also not been clearly explained.
For chats that can no longer continue, users can first export their history, and both official and community migration tools are available to try.
The official side has just launched two migration paths
On November 13, DeepSeek opened a full set of application APIs and also open-sourced the chat client Chat Demo.
Chat Demo provides a ready-made chat interface. After configuring a DeepSeek API Key, it can call the model as well as search, file reading, voice, and other functions. Developers can deploy it themselves or modify it on that basis.
The official documentation also includes a feature: support for importing DeepSeek Chat history and continuing the conversation in the new client.
However, note that according to the official instructions, the import operation will replace all existing local conversations and attachments in the client. Therefore, if other chat records have already been accumulated in Chat Demo, it is best to back them up first before performing the import.
Another tool is DeepSeek Harness (DSH).
Also on November 13, DSH released the v0.2.2-rc.3 preview version, which added an experimental DeepSeek session import plugin.
It can import history from the DeepSeek app or web version into the workspace.
Compared with Chat Demo, DSH is more oriented toward an Agent work environment, and imported chats can be used together with files and tasks.
However, it is still an experimental feature and requires users to enable it themselves.
Both tools support migrating history, but the barriers to use differ. Chat Demo requires configuring an API Key and bearing call costs, while DSH requires installing and configuring the corresponding plugin and model service.
Beyond official tools, there are several more convenient migration methods
In addition to DeepSeek's officially launched Chat Demo and DSH, community developers have also made some tools specifically for handling historical chat records.
For example, dsh-chat-migration, released by developer Xiaozhuang, is a community plugin that adds DeepSeek historical conversation migration capability to DSH.
It actually consists of two plugins: "Import Conversations" and "Chat Mode."
The former can read the JSON or ZIP exported officially by DeepSeek, preview the original chat records, and then select the conversations to migrate. Titles, chronological order, question-and-answer content, and the thinking records saved in the export file can all be converted into chat records in DSH.
Conversations that have already been imported will also be marked, so they can be skipped when importing again.
"Chat Mode" provides an entry point that is closer to ordinary chat software, so you do not need to bind a workspace every time.
Note that this community plugin and the DSH official experimental import plugin mentioned earlier are different projects, and their installation methods are also different.
If you do not usually use DSH, you can also take a look at the browser extension DeepShare.
Its operation is closer to the chat page familiar to ordinary users.
Users first download their history from DeepSeek, and then DeepShare can build a local history library in the browser. After that, you can search past chats, select the passages you need, and organize them into Markdown files.
For example, if you previously revised an article back and forth with DeepSeek and want to continue writing it with another AI, you can first organize the discussions related to that article, and then upload the Markdown file to a new chat window.
DeepShare also supports directly placing selected history records as attachments into a new DeepSeek conversation.
However, this method only lets the model re-read the old content; it does not restore the original chat thread.
According to the developer documentation, the parsing, searching, and Markdown generation of history records are all completed locally in the browser, and no additional API Key configuration is required.
If you do not even need to continue chatting and only want to save past records, there is also an open-source project called deepseek-keeper.
It is mainly used for offline browsing of DeepSeek chat records. You can view historical content in a terminal or browser, making it suitable for making local backups of old chats.
These tools can all start from the history files exported by DeepSeek, but their processing methods differ.
If you want to preserve the conversation format and continue using DSH, you can consider the conversation import plugin; if you want to organize old content for other AIs to use, DeepShare provides a lighter-weight approach; if you only want to archive, there is no need to specially deploy a chat client.
Whichever tool you choose, it is best to first download the original history records from DeepSeek's "Data Management" page and keep a backup.
Export files may contain private chats, work documents, and other sensitive information. Before using third-party tools, it is recommended to confirm whether the files need to be uploaded to external servers, and attachments such as PDFs and images used in the original chats should also preferably be saved separately.
Finally, I still recommend that everyone "prepare for a rainy day": while DeepSeek can still be used, first have it do a summary review: what has been discussed before, which requirements have been confirmed, and what things remain unfinished.
After all, when you really run into quota limits, no matter what you say, it will ignore you.
AI Trends