How to use DeepSeek on Janitor AI?
Get a DeepSeek API key at platform.deepseek.com, then in Janitor AI open API Settings, pick Proxy, click Add Configuration, and enter proxy URL https://api.deepseek.com/chat/completions, model name deepseek-v4-flash, and your key. Save, then refresh the page. Skipping the refresh is why most setups fail.
Why — the first-principles explanation
Janitor AI is a front end, not a model. It stores your characters and your chat history, but when a reply needs to be written, that text has to come from a language model running on someone's expensive GPUs. Janitor gives you a limited free model, and for anything else it makes you bring your own brain — you point it at an outside model and pay that provider directly. In Janitor's vocabulary this connection is called a proxy.
The reason DeepSeek works here at all is a quiet technical convention. DeepSeek's API is deliberately OpenAI-compatible: it accepts requests in the same shape OpenAI's API uses, at the same `/chat/completions` path. Janitor knows how to speak that shape. So Janitor can talk to DeepSeek without either company doing anything special for the other — you are just redirecting the same envelope to a different address.
That framing explains every error you will hit. A proxy setup is three matching parts: a URL (where to send it), a key (proof you may), and a model name (what to run when it arrives). All three must come from the same provider. A DeepSeek key sent to an OpenRouter URL is a valid password handed to the wrong building — it fails, and Janitor reports it as a vague "network error" that tells you nothing. Nearly every setup problem people post about is this mismatch, not a broken site.
One trap is specific to right now. Most DeepSeek-on-Janitor tutorials online tell you to type `deepseek-chat` or `deepseek-reasoner` as the model name. Those names are being retired on 2026-07-24 at 15:59 UTC. Until then they still work, because DeepSeek maps them onto the non-thinking and thinking modes of `deepseek-v4-flash`. After that date, a guide written in 2025 will hand you a model name that no longer exists, and your chats will simply stop. Use `deepseek-v4-flash` or `deepseek-v4-pro` and you skip the whole problem.
An example that makes it click
Think of Janitor AI as a phone with no service. The handset is yours — contacts, message history, the whole interface — but it cannot place a call until you put in a SIM card from a carrier you pay. DeepSeek is the carrier. The API key is your SIM.
And the three fields are just a phone number typed in full. The proxy URL is the country code, the model name is the extension, and the key is the PIN that proves the line is yours. Get one digit from a different carrier's number and the call does not ring wrong — it does not connect at all, and the phone just says "failed." That is your network error.
How to do it
- Create an account at platform.deepseek.com and add credit. DeepSeek's API is prepaid and separate from the free deepseek.com chat site — a chat account alone will not work.
- In the DeepSeek platform, open API keys and create a new key. Copy it immediately — the full key is shown only once.
- Add funds. As of 2026-07, deepseek-v4-flash costs $0.14 per million input tokens and $0.28 per million output tokens, so a few dollars covers a great deal of roleplay.
- Open Janitor AI, start any chat, and click API Settings at the top.
- Choose Proxy, then click +Add Configuration.
- Config Name: anything you like, e.g. DeepSeek.
- Model Name: deepseek-v4-flash (cheaper and faster) or deepseek-v4-pro (stronger, roughly 3x the price). Do not use deepseek-chat or deepseek-reasoner — they are retired on 2026-07-24 15:59 UTC.
- Proxy URL: https://api.deepseek.com/chat/completions — copy and paste it exactly, and do not click inside the field to edit it by hand.
- API Key: paste your DeepSeek key with no leading or trailing spaces.
- Click Save Settings, then do a full refresh of the browser page. Janitor's own guide calls this step crucial; settings often do not take effect until you reload.
- Send a test message. If you get a network error, check that the URL, key, and model name all come from DeepSeek and not a mix of providers.
Key facts
- DeepSeek's API base URL is https://api.deepseek.com and is OpenAI-compatible, which is why Janitor AI's proxy feature can talk to it unmodified (DeepSeek API docs, as of 2026-07).
- The model names deepseek-chat and deepseek-reasoner are deprecated on 2026-07-24 at 15:59 UTC; until then they map to the non-thinking and thinking modes of deepseek-v4-flash.
- Current model IDs as of 2026-07 are deepseek-v4-flash and deepseek-v4-pro, both with a 1M-token context window and up to 384K tokens of output.
- Pricing as of 2026-07 per million tokens: deepseek-v4-flash $0.14 input (cache miss) / $0.28 output; deepseek-v4-pro $0.435 input (cache miss) / $0.87 output. Cache hits cost about $0.0028 and $0.003625 respectively.
- Janitor's official proxy guide requires four fields — Config Name, Model Name, Proxy URL, and API Key — and states that refreshing the page after saving is essential.
- Janitor's official beginner guide uses OpenRouter as its worked example, which offers 50 free messages per day; DeepSeek direct is an alternative route using the same proxy mechanism.
An open-weight Chinese model family that matched frontier quality at low cost.
▶ The 60-second explainer (script)
To use DeepSeek on Janitor AI, you need three things that all come from DeepSeek: a URL, a key, and a model name. Here's the whole setup. First, go to platform dot deepseek dot com — that's the developer platform, not the free chat site — make an account, and add a few dollars of credit. Create an API key and copy it right away, because you only see it once. Now open Janitor AI, start any chat, and click API Settings. Choose Proxy, then Add Configuration. For Proxy URL, paste https colon slash slash api dot deepseek dot com slash chat slash completions. For Model Name, type deepseek-v4-flash. And here's the part that trips people up: almost every tutorial online still tells you to type deepseek-chat or deepseek-reasoner. Those names are being retired on July 24th, 2026. Use deepseek-v4-flash or deepseek-v4-pro instead. Paste your key, click Save Settings, and then refresh the page — a full browser refresh. Janitor's own documentation says this step is essential, and skipping it is the single most common reason people think the setup is broken. If you still get a network error, it's almost always because you mixed providers — a DeepSeek key with an OpenRouter URL, for example. The URL, the key, and the model name have to match. Cost-wise, flash runs about fourteen cents per million input tokens, so a few dollars goes a long way.
What authoritative sources say
People also ask
Is DeepSeek free on Janitor AI?
No. DeepSeek's API is prepaid and billed per token, separate from the free chat at deepseek.com. It is very cheap though — as of 2026-07, deepseek-v4-flash is $0.14 per million input tokens, so a few dollars lasts most users a long time.
Why do I keep getting a network error?
Almost always because your URL, API key, and model name are not all from the same provider — a DeepSeek key sent to an OpenRouter URL, for example. The other common cause is not refreshing the page after clicking Save Settings.
Should I use deepseek-v4-flash or deepseek-v4-pro?
Start with flash. It is roughly three times cheaper and fast enough for roleplay. Switch to pro only if you find the writing quality lacking, since pro costs $0.435 per million input tokens versus $0.14.
My old guide says to use deepseek-chat. Does that still work?
Only until 2026-07-24 at 15:59 UTC. DeepSeek is retiring both deepseek-chat and deepseek-reasoner on that date. Use deepseek-v4-flash or deepseek-v4-pro so your setup does not break.
Can other people see or steal my API key?
You are pasting the key into a third-party site, so treat it as exposed. Fund the account with a small amount, do not reuse the key elsewhere, and delete it from the DeepSeek platform if you stop using it or notice unexpected spend.