How to Set Up DeepSeek on Janitor AI (Step by Step)

To use DeepSeek on Janitor AI, create a DeepSeek API key, then add a new configuration in Janitor AI's API Settings with the proxy URL https://api.deepseek.com/chat/completions and the model name deepseek-flash. Save, refresh the page and send a test message. The whole setup takes about five minutes, and chatting costs a fraction of a cent per message. This guide covers the technical setup only: keys, URLs, model names, settings, costs and error fixes. It follows the official Janitor AI proxy quickstart and DeepSeek's API documentation, checked on October 7, 2026.

What does "proxy" mean on Janitor AI?

On Janitor AI, a proxy is a connection to an outside language model that you pay for or provide yourself, instead of the site's built-in model. Janitor AI is a character chat platform. Its built-in model is called JLLM. A "proxy" setting lets you send chat requests to another provider's API, such as DeepSeek or OpenRouter, using your own API key.

The word is misleading. It is not a network proxy that hides your traffic. It is an OpenAI-compatible endpoint address. Janitor AI sends each chat to that address and shows the reply. DeepSeek's API speaks the OpenAI request format, so it works as a drop-in endpoint, per DeepSeek's first API call guide.

New to the term? What a proxy means on Janitor AI explains it in plain English before you start.

What do you need before you start?

You need three things: a Janitor AI account, a DeepSeek platform account with some balance, and an API key.

1. A Janitor AI account, signed in on the website.
2. A DeepSeek platform account with a topped-up or granted balance. DeepSeek bills per token from that balance and returns a 402 error when it runs out, per its error code list.
3. An API key from the DeepSeek platform. Treat it like a password.

Do not share the key, paste it into public bot descriptions, or post screenshots that show it. Anyone with the key can spend your balance.

How to set up DeepSeek on Janitor AI

These steps use DeepSeek's own API, which is the most direct route. Janitor AI's menus change from time to time, so if a label differs, look for the same field names. We could not test inside a Janitor AI account for this article, so the field layout comes from the official quickstart, and the values come from DeepSeek's documentation.

1. Create a DeepSeek API key. Sign in to the DeepSeek platform, top up a small balance, open the API keys page and create a key. Copy it once and store it somewhere safe.
2. Open a chat on Janitor AI. Start any chat with a character.
3. Open API Settings. Click the gear icon at the top right of the chat, choose "API Settings", then choose "Add Configuration", as the official quickstart describes.
4. Fill in the fields using the values below.
5. Save. Click "Add", then "Save Settings".
6. Refresh the page. Janitor AI's own error guide says to save and then refresh the whole page, even if you changed nothing.
7. Send a test message. If it replies, the setup works.

Use these exact values:

Field	Value
Config name	Any label, for example DeepSeek Flash
Model name	deepseek-flash
Proxy URL	https://api.deepseek.com/chat/completions
API key	Your DeepSeek key, with no extra spaces
Custom prompt	Leave blank at first

Every URL variant, and which ones to avoid, is listed in DeepSeek proxy URL for Janitor AI. Type the proxy URL exactly. Janitor AI's error guide says a wrong URL, with extra slashes or repeated segments, produces a generic network error.

Which DeepSeek model name should you use?

Use deepseek-flash for most chats and deepseek-v4-pro only if you want the larger model and accept a higher bill. As of October 7, 2026, DeepSeek lists two models on its pricing page.

Model name	What it is	Output price per 1M tokens	Notes
deepseek-flash	DeepSeek-V4.1-Flash	$0.60 off-peak, $1.20 peak	Cheapest, handles images, 1M context
deepseek-v4-pro	DeepSeek-V4-Pro-0813	$1.98 off-peak, $3.96 peak	Larger text-only model

Older guides tell you to use names such as deepseek-chat, deepseek-v3 or deepseek/deepseek-chat-v3-0324:free. Those are out of date. DeepSeek says the old deepseek-v4-flash name only works as a temporary alias that routes to V4.1-Flash. A wrong or retired model name is a common cause of "No endpoints found" errors, as Janitor AI's error guide explains.

How much does DeepSeek cost on Janitor AI?

DeepSeek costs roughly 0.03 to 0.28 cents per message on deepseek-flash, depending on caching and the time of day. That is $0.32 to $2.76 for 1,000 messages. The table uses an assumed prompt of 8,000 tokens, which includes the character definition and chat history, and a reply of 300 tokens. Your real prompt size depends on your Janitor AI settings, so treat these as a model and not a promise.

Scenario (per 1,000 messages)	deepseek-flash	deepseek-v4-pro
Off-peak, 90% of the prompt cached	$0.32	$1.28
Off-peak, no caching	$1.38	$5.87
Peak, 90% of the prompt cached	$0.64	$2.56
Peak, no caching	$2.76	$11.75

The maths uses the DeepSeek price table: Flash input is $0.15 per million tokens on a cache miss and $0.003 on a cache hit off-peak, and double at peak. DeepSeek's context caching is on by default and reuses the start of your prompt. A long chat grows by appending new messages, so most of each prompt matches what DeepSeek stored before. Editing or deleting an early message changes the start of the prompt and loses the cache for everything after it.

Peak hours are 01:00-04:00 and 06:00-10:00 UTC on weekdays, and weekends are off-peak all day. We explain the schedule and how it lands in your time zone in DeepSeek V4 pricing now depends on what time you run it.

What settings work best?

Start with the defaults, then change one thing at a time. The temperature slider may not do anything on DeepSeek. DeepSeek's thinking mode guide says thinking mode is on by default, and that in thinking mode the temperature, presence_penalty and frequency_penalty parameters "will not trigger an error but will also have no effect." We could not confirm whether Janitor AI's proxy form switches thinking off, so test it: send the same message three times at 0.2 and three times at 1.0, and compare how much the replies vary. If the slider works on your setup, Janitor AI's temperature guide describes 0.5 to 0.7 as a medium range suited to natural conversation. Our guide to the best DeepSeek model and prompts for Janitor AI covers the test in more detail.

Thinking mode also affects cost. The reasoning DeepSeek writes before its answer is billed as output, so the per-message costs above, which assume a 300-token reply, are a floor. See DeepSeek API pricing: what 1,000 messages really cost for how much thinking can add.

Keep the custom prompt short. Every token in it is sent with every message, so it adds to the cost. Janitor AI's Advanced Prompting guide explains what goes there. Its token guide explains why long chats eventually push old messages out of the model's memory, which also applies when you manage cost.

Is there a free way to use DeepSeek on Janitor AI?

Not reliably. Many older guides point to a free DeepSeek model on OpenRouter, but OpenRouter's public model list on October 7, 2026 contains no DeepSeek model with the :free suffix. The official Janitor AI quickstart still shows deepseek/deepseek-chat-v3-0324:free, and its page was last updated in 2025, so that model ID is no longer a safe bet.

You have two realistic options.

• DeepSeek directly. You pay per token from a small balance. This guide's cost table shows a few cents covers a lot of chatting. New accounts may receive a granted balance, so check your account page.
• OpenRouter. OpenRouter lists deepseek/deepseek-v4.1-flash at roughly $0.05 per million input tokens and $1.20 per million output tokens. For Janitor AI, set the proxy URL to https://openrouter.ai/api/v1/chat/completions and the model name to deepseek/deepseek-v4.1-flash. Check the model page for current rates, because routers change them. If you are weighing routers, read whether to keep routing through OpenRouter after the Stripe acquisition.

Be careful with "free DeepSeek proxy" sites. A third party that offers a free endpoint sees your prompts and, if you give it one, your key. Use only DeepSeek's own API or a router you trust, and never paste a key into an unknown site. For the wider question of what providers do with your text, see whether the free AI API tier trains on your data.

How do you fix DeepSeek errors on Janitor AI?

Most failures come from a wrong URL, a wrong model name, an empty balance or a missing refresh. Work through the basics first. Janitor AI's error guide recommends this order: check the configuration says Active, click Save Settings, refresh the whole page, then click Test.

Symptom	Likely cause	Fix
"A network error occurred"	Wrong proxy URL, or you did not refresh	Use exactly https://api.deepseek.com/chat/completions, save, refresh
401 Authentication Fails	Wrong or mistyped API key	Paste the key again with no spaces, or create a new key
402 Insufficient Balance	DeepSeek balance is empty	Top up on the DeepSeek platform
404 or "No endpoints found"	Retired or misspelled model name	Use deepseek-flash or deepseek-v4-pro
400 about maximum context length	Prompt too large for the request	Shorten the memory, custom prompt or character definition
429 Rate Limit Reached	Requests arriving too fast	Wait and retry, per DeepSeek's error codes
500 or 503	DeepSeek server problem or overload	Retry after a short wait, and check the DeepSeek status page
"Network Error" after clicking Check Key/Model	That button is reported as buggy	Ignore it, save settings and refresh instead

If DeepSeek's own endpoint is overloaded, switching the configuration to a router is a quick workaround. Keep both configurations saved and activate whichever works.

Want to compare DeepSeek with other models?

Our comparisons cover DeepSeek vs OpenAI and DeepSeek vs Claude, including cost per message at the same prompt size.

When something goes wrong

If the connection fails, start with Is Janitor AI down? How to check and fix errors, which separates site errors from proxy errors. If you also use Gemini and it stops working, see Janitor AI suspended from Gemini for what to check and why a DeepSeek backup helps. To stop a bot from writing your character's actions, read how to stop Janitor AI talking for you.

Security checklist

1. Create a dedicated API key for Janitor AI so you can revoke it without breaking anything else.
2. Top up a small balance rather than a large one, so a leaked key has limited cost.
3. Never share the key in bot descriptions, comments, screenshots or support chats.
4. Use only DeepSeek's endpoint or a router you have checked.
5. Review your usage on the DeepSeek platform weekly.

Sources

Janitor AI: TL;DR Quickstart Proxy Instructions, Janitor AI: OpenRouter Error Guide, Janitor AI: Understanding Temperature, Janitor AI: Advanced Prompting 101, DeepSeek: Your First API Call, DeepSeek: Models and Pricing, DeepSeek: Context Caching, DeepSeek: Error Codes, DeepSeek-V4.1-Flash model card, OpenRouter: DeepSeek V4.1 Flash.