DeepSeek Proxy URL for Janitor AI: Which to Use

Short answer: the official DeepSeek proxy URL for Janitor AI is https://api.deepseek.com/chat/completions. Pair it with the model deepseek-flash and a key from DeepSeek's own platform. Every other address belongs to a third party that sits between you and the model.

That one decision, official endpoint or reseller, drives your cost, your privacy and how often the connection breaks. This guide lists every URL variant, shows how to test your key before you touch Janitor AI, and explains which third-party options are still worth using. All values were checked on October 7, 2026.

The official DeepSeek proxy URL

DeepSeek's first API call guide lists two base URLs:

• OpenAI format: https://api.deepseek.com
• Anthropic format: https://api.deepseek.com/anthropic

Janitor AI speaks the OpenAI chat format, so use the first one and add the path. In the proxy URL field, paste:

https://api.deepseek.com/chat/completions

Do not stop at the base address. DeepSeek's own example calls https://api.deepseek.com/chat/completions, and Janitor AI's error guide says a URL with a wrong or missing path produces "a network error occurred".

Every URL variant and what it does

Some guides give a different address, so here is each one. We confirmed on October 7, 2026 that both chat paths reach DeepSeek's server and ask for a key. Without a key, each returned HTTP 401 and not 404, which means the path exists.

URL	Use it in Janitor AI?	Notes
https://api.deepseek.com/chat/completions	Yes	The documented path. Start here
https://api.deepseek.com/v1/chat/completions	Yes, as a fallback	Also reaches the server. Community guides use it
https://api.deepseek.com	No	A base address. Janitor AI does not add the path for you
https://api.deepseek.com/anthropic	No	For Anthropic-format tools such as Claude Code
https://api.deepseek.com/beta	No	For beta features like prefix completion, not normal chat

Test your key before you open Janitor AI

Testing the key and model in a terminal separates a DeepSeek problem from a Janitor AI problem. If this command works, the key, balance and model name are fine, and any remaining error is in the Janitor AI form.

curl https://api.deepseek.com/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $DEEPSEEK_API_KEY" \
  -d '{
    "model": "deepseek-flash",
    "messages": [{"role": "user", "content": "Reply with the word ok."}],
    "max_tokens": 20,
    "thinking": {"type": "disabled"}
  }'

A working call returns JSON with a choices array. A failure returns an HTTP code that DeepSeek documents in its error code list.

Code	Meaning	Fix
401	Authentication fails	Wrong or mistyped key. Create a new one
402	Insufficient balance	Top up on the DeepSeek platform
400 or 422	Invalid format or parameters	Check the model name and body
429	Rate limit reached	Slow down and retry
500 or 503	Server error or overload	Wait and retry

Matching model names

Model	API name	Use when
DeepSeek-V4.1-Flash	deepseek-flash	Default. Cheap and fast
DeepSeek-V4-Pro	deepseek-v4-pro	Flash feels flat and you accept about 4x the cost

Old names such as deepseek-v4-flash still work and bill at the Flash price, but DeepSeek says the matching models are retired, so use the new name. Model names are exact strings. Use lowercase with hyphens and no spaces, because a display name will fail. Older guides tell you to use deepseek-chat, a name that DeepSeek's current pricing page does not list.

Third-party proxy options

A third-party host resells access to DeepSeek models through its own endpoint, with its own key and its own model names. The best known is OpenRouter, which Janitor AI's official quickstart uses as its example.

	Official DeepSeek	OpenRouter
Proxy URL	https://api.deepseek.com/chat/completions	https://openrouter.ai/api/v1/chat/completions
Key from	DeepSeek platform	OpenRouter account
Flash model name	deepseek-flash	deepseek/deepseek-v4.1-flash
Pro model name	deepseek-v4-pro	deepseek/deepseek-v4-pro-0813
Listed Flash price	$0.15-$0.30 in, $0.60-$1.20 out	about $0.05 in, $1.20 out
Free DeepSeek model	No	None listed

The OpenRouter figures come from its public models list on October 7, 2026. That list contained no DeepSeek model with the :free suffix, so the free model ID in older guides, deepseek/deepseek-chat-v3-0324:free, is stale. Janitor AI's quickstart still shows it, and its page was last updated in 2025. If you route through OpenRouter, read whether to keep routing through it after the Stripe acquisition.

Why people choose a third party: one key for many models, a different payment method, or a fallback when DeepSeek's endpoint is busy. Why to be careful: the host sees every prompt, prices change, and a host may serve a different build than the one you expect.

Is a free DeepSeek proxy safe?

It can be, but treat "free" as a trade. Check four things before you paste a key:

1. Who runs it? A named company with documentation and a status page beats a URL in a forum post.
2. What do they log? Your chats are in the request. Read the privacy policy, and see whether the free AI API tier trains on your data.
3. What does the key control? A key tied to a small prepaid balance limits the damage if it leaks.
4. Can you leave? Avoid anything that asks you to install software or to hand over your official DeepSeek key.

Never give a third party your official DeepSeek key. Their endpoint takes their own key, not yours. Anyone holding a working key can spend the balance behind it.

Why does my DeepSeek proxy URL give an error?

Most failures come from the URL, the model name or a missing refresh. Janitor AI's error guide recommends the same first steps for every proxy error: confirm the configuration says Active, click Save Settings, refresh the whole page and then click Test.

Symptom	Likely cause	Fix
"A network error occurred"	Wrong URL, or no refresh after saving	Use the exact URL, save, refresh
404 or "No endpoints found"	Retired or misspelled model name	Use deepseek-flash
401	Key from a different provider	Match key, URL and model to one provider
402	Empty balance	Top up
Works in curl, fails in Janitor AI	Trailing space or the wrong field	Re-paste the key and URL, save, refresh

If the curl test works and Janitor AI still fails, the cause sits in the form, not in DeepSeek.

Checklist

1. Key, URL and model all come from the same provider.
2. The URL ends in /chat/completions.
3. The model name is lowercase and exact.
4. The key has no trailing space.
5. The DeepSeek balance is above zero.
6. You saved the configuration and refreshed the page.

The full walkthrough is in how to set up DeepSeek on Janitor AI, the cost per message is in DeepSeek API pricing, and the word "proxy" is explained in what is a proxy on Janitor AI.

Sources

DeepSeek: Your First API Call, DeepSeek: Models and Pricing, DeepSeek: Error Codes, Janitor AI: TL;DR Quickstart Proxy Instructions, Janitor AI: OpenRouter Error Guide, OpenRouter models API.