AI Trends

How to Continue Using DeepSeek via the API, Harness, and Chat Demo After Free-Use Limits Were Introduced

36Kr · AIUpdated Nov 20 21:00

AI summary

DeepSeek’s free chat service has introduced limits on long conversations, per-request input, and daily usage. This article explains how to continue using it through the official API, DeepSeek Harness, and the open-source Chat Demo, as well as how to migrate chat history. In testing, a single deep-research session consumed 55.64 million Tokens and cost 4.92 yuan, while an ordinary Q&A session cost 0.17 yuan. Chat Demo stores the API Key and settings in localStorage, and stores conversations and original attachments in IndexedDB. Clearing browser data may result in the loss of these records.

Why read this

AI score 80

This article outlines migration paths using the official API, Harness, and open-source clients, and compares the costs of deep research and ordinary Q&A based on actual bills.

Topics

Tags

Body · AI translation

DeepSeek has begun limiting usage.

Over the past two days, more and more people have found that the free DeepSeek service they had used for nearly two years suddenly began displaying a message saying, “Usage limit reached.” Some users could no longer continue lengthy conversations, while others triggered the limit after chatting for less than two hours.

Users were furious. Some even flooded app stores with negative reviews of DeepSeek, saying they had accumulated hundreds of thousands of words in their DeepSeek chat windows, and stories they had spent weeks developing continuously with the AI could no longer be continued.

At present, the main restrictions on free DeepSeek conversations include:

Long conversations cannot be continued: Lengthy conversations that have been inactive for more than seven days are no longer supported, and the system prompts users to export their historical data.

Input length limit: The maximum amount of text allowed in a single input has been reduced to around 40,000 Chinese characters.

Daily usage cap: The total number of conversations per day is limited. Once the cap is reached, a “Server busy” message appears, and access is restored the following day.

Many users in the comments have even begun pleading with DeepSeek to charge them, hoping the DeepSeek app will introduce monthly, quarterly, or annual plans so that frequent users can pay for access. The bad news is that the DeepSeek app currently offers no membership plan at all.

The good news is that if you really are willing to pay to continue using DeepSeek, there is actually an “unlimited” way to do so—and it is not some unorthodox workaround, but a fully legitimate official channel.

Today, APPSO will walk everyone through the process step by step. Even ordinary users who have never encountered the concept of an API can learn how to pay for access, move their previous chat history over, and effectively “buy a membership” for DeepSeek.

Using the API is not difficult

Most people may think an API sounds complicated. Its full name is Application Programming Interface. Put simply, we can think of it as being like the login credentials we enter to use the DeepSeek app: a phone number and password/verification code.

The phone number previously used to log in to the app can be understood as DeepSeek allocating a certain amount of free usage to each phone number, whereas API usage requires us to add funds ourselves.

Adding funds is also straightforward. Find “API Open Platform” on DeepSeek’s official website, click it, and log in using a phone number or direct WeChat authorization.

On the DeepSeek Open Platform homepage, the left sidebar contains, in order, “Usage,” for checking daily usage, “API Keys,” “Top Up,” and “Billing,” along with API documentation, product pricing, and other information at the bottom.

If you have not created an API before, your first step can be to visit the top-up page and add the desired amount using WeChat Pay or Alipay.

Under the current pricing for the DeepSeek Flash and Pro models, the Flash model version is DeepSeek-V4.1-Flash-1017, while the Pro model version is DeepSeek-V4-Pro-0813.

Taking the DeepSeek Flash model during peak hours as an example, we set the reasoning level to the highest Max setting on DeepSeek’s official Agent platform and sent the prompt: “Help me conduct in-depth research on laptops currently on the market. Which ones would you recommend at around 6,000 yuan? Give me a web report.”

DeepSeek ultimately took 19 minutes and consumed a total of 55.64 million Tokens. Of these, 99.3%, or a cumulative 55.04 million Tokens, were cache hits; uncached input accounted for 381,000 Tokens, while output accounted for 209,000 Tokens. Based on the price per million Tokens, the calculated cost was therefore (input 38.1 * 2 + output 20.9 * 8 + cache 5500 * 0.04) / 100 (in millions) = 4.634 yuan.

The Open Platform showed that the amount consumed was 4.92 yuan, which also included fees for tool-search and other services. Currently, online searches for text and images, as well as webpage retrieval, are billed per use at 0.005 yuan each. We used tool-search 38 times in this test, costing 0.19 yuan.

This is how most people feel the first time they use an API: “Why did the backend show that I was charged several tenths of a yuan when all I did was say hello?”

If you add 50 yuan, you can complete only about 10 similar in-depth research tasks. The model’s input includes not only the prompt we send but also context such as the conversation history sent with the request, and Tokens do not correspond directly to the number of Chinese characters. In addition, cache hits reduce input costs, while fees for tools such as online search are charged separately.

However, if we only want to use it for conversations without involving workflows such as Agents, Token consumption drops significantly. We asked the same question in DeepSeek’s Q&A mode, and it generated an answer in just 6 seconds at a cost of 0.17 yuan. Of that, 16 tool-search calls cost 0.08 yuan, while 67,038 Tokens cost 0.09 yuan.

The Token charge was calculated as (31,700 cache-hit input Tokens × peak rate of 0.04 + 31,500 cache-miss input Tokens × peak rate of 2 + 3,100 output Tokens × peak rate of 8) / 100 = 0.089 yuan.

Therefore, if we simply want to replicate the previous experience of using the DeepSeek App or its web version, 50 yuan would support nearly 300 such queries, even with “Deep Thinking” and “Smart Search” enabled.

Once we understand the API’s pricing rules, we can safely add funds. As long as the DeepSeek account has a balance, any API Key we create can be used normally. An API Key is like your login password—a string beginning with sk-.

If an API Key is leaked and used by someone with malicious intent, they could potentially use up the entire amount we just added.

An API Key is displayed only once, but if you forget it, you can simply create a new one and delete the old one. The DeepSeek Open Platform currently allows users to retain up to 300 API Keys, which is more than enough for most people.

The Official Desktop App

Once we have an API Key, the easiest approach is to download DeepSeek’s official Agent platform, DeepSeek Harness.

We have previously shared several tutorials about DeepSeek Harness. It is a product similar to Codex or the recently popular Doubao Work, Qwen Office, and WorkBuddy. Its biggest advantage over ordinary chat interfaces is its ability to use tools on the computer and operate the computer to complete a variety of more complex tasks.

DeepSeek Harness now also supports signing in directly with a DeepSeek Open Platform account and using the balance on that platform. After signing in, we can see the amount added on DeepSeek’s official Open Platform under “Account & Balance” in the settings.

The API Keys management interface will also show an “App-managed” API Key managed by Harness.

By creating workspaces for different projects, starting a “New Session” within a workspace, selecting a model and reasoning level, and choosing whether to enable Agent team collaboration, we can chat with DeepSeek just as we would on the DeepSeek website, without restrictions as long as there is sufficient API balance.

Whether we are co-creating interactive fiction or text-based games, or asking DeepSeek to role-play and provide emotional companionship, we will no longer be affected by the usage limits imposed in the official App.

The latest version of DeepSeek Harness also provides a plugin for importing DeepSeek conversations with one click.

There Is Also an Official Open-Source Alternative for Mobile Devices

What if I do not have a computer and simply want to use the DeepSeek App on my phone as before?

Just a few days before the controversy over the limits erupted, DeepSeek announced that it was opening up APIs for its full suite of app features. This update made features such as online search, file input, webpage reading, and voice available to developers. DeepSeek even stated that developers could use the APIs to replicate the complete DeepSeek App.

At the same time, DeepSeek also released an open-source chat client that supports importing historical chat records.

DeepSeek’s newly launched Chat Demo also provides a feature for importing historical conversations. We can go to Data Management in the DeepSeek App’s settings, export all historical conversations, and then import the exported conversations into the new DeepSeek Chat Demo.

Here is how it works: in the original DeepSeek App, after tapping Export Conversations, we receive a download link. Copying that link into a browser and opening it will automatically download a ZIP archive.

Next, return to the API-based Chat Demo, find Data Management in the settings, and import the archive downloaded earlier. Note that uploading historical conversations will overwrite all current conversation records.

The drawback of using a browser is that all future DeepSeek conversations are stored locally. If the browser data is cleared, the conversations cannot be recovered.

According to the project documentation, the API Key and settings are saved in the browser’s localStorage, while sessions and original attachment files are stored in IndexedDB. Both localStorage and IndexedDB can be understood as types of local databases built into the browser, representing a common division of labor in local-first storage.

Unlike the DeepSeek App, it does not save all our conversations on DeepSeek’s official cloud servers, where they can be viewed on either the mobile app or the web as long as we are logged in.

The benefit is that migration is convenient. Using the import and export features provided by DeepSeek, we can treat these chat records as our own work products and organize them in folders.

As for the memory features offered by applications such as ChatGPT and Claude, the DeepSeek App currently does not provide cross-conversation retention either. A simple example is asking ChatGPT to evaluate what kind of person we are: ChatGPT will analyze all our conversations to determine what we do and what kind of person we are.

But with DeepSeek, even after I have had 171 conversations with it, when I ask what impression it has of me, it only says, “At present, I can only see what you have voluntarily shared in this conversation. I do not retain your personal information across conversations.”

Therefore, when exporting conversations, we have no need to worry that the new DeepSeek will no longer know us, because the old DeepSeek was never trying to “know” us in the first place.

What if we still want to “pick up” on our computer where we left off in a DeepSeek conversation on our phone? In that case, we need to configure a third-party cloud database ourselves to synchronize the conversation data in real time.

For example, Cloudflare, hailed as the internet’s cyber bodhisattva, provides free static asset requests through Pages. Combined with Supabase’s free tier, which currently offers a 500 MB database and 1 GB of file storage, this can initially be used to validate personal usage requirements.

Alternatively, other solutions can be used. GitHub, for instance, has quite a few projects for synchronizing DeepSeek Harness conversations, as well as many similar projects for DeepSeek chat sessions.

We also casually asked Codex to develop a solution that uses a third-party cloud drive to synchronize conversation data.

By using Nutstore to store DeepSeek conversations, we only need to log back into Nutstore and re-enter the API Key on a different device to continue previous conversations.

If even registering for API access feels troublesome, but we still do not want to be constrained by DeepSeek’s official app, the final option is to use an “API relay service.”

WorkBuddy, Qianwen Office, and Doubao Work have now all officially announced support for third-party models. These third-party models include the latest DeepSeek models, with different Agent platforms offering different rate multipliers.

For example, in WorkBuddy, the DeepSeek V4.1 Flash 1017 model is priced at 0.11x, while DeepSeek V4 Pro is priced at 0.51x. However, on DeepSeek’s official open platform, the prices of the two models do not simply differ by approximately fivefold.

Image | WorkBuddy

In Doubao Work, meanwhile, DeepSeek V4.1 Flash is currently still shown as discounted, with a multiplier of 0.01x.

Image | Doubao Work

Points are also consumed differently across platforms. The same 1,000 points may accomplish different amounts of work in WorkBuddy and Doubao Work, and the number of points consumed behind each stated multiplier may also vary.

Image | Qianwen Office

Now, we only need to add funds to these platforms to likewise gain access to DeepSeek without usage caps.

As with using the API ourselves, the only thing we will probably need to adjust to is paying a little attention to our Token bill every time we chat from now on.

This article is from the WeChat public account “ifanr,” authored by Discover Tomorrow’s Products, and published by 36Kr with authorization.

Source

36Kr · AI