Zed AI: OpenRouter Support

[daun-io]

Hi, I'm from OpenRouter and a fan of Zed and would love to see OpenRouter integration as a new assistant provider.
OpenRouter supports dozens of the best coding models and hundreds of different models with one API key and an OpenAI-style schema.
I saw that some Zed community users have requested this feature 1 2.
Is there anything I can do to help? Thanks! 💙

[deanrie]

It would be nice to have this support.

[mintisan]

It would be nice to have this support.

[minato7namikazi]

it would be awesome : )

[tm17-abcgen]

Would love to have this too. ZedAI currently works with Anthropic only. Since OpenRouter has access to Anthropic, i hope that ZedAI can be used with OpenRouter too, to make use of the prompt caching also!

[dkphhh]

we need this！

[hrefhref]

isn't this already supported : #13276

[drivencompassion]

Awaiting OpenRouter config?

[gliptic]

OpenRouter's API is OpenAI compatible, so just e.g.:

  "language_models": {
    "openai": {
      "version": "1",
      "api_url": "https://openrouter.ai/api/v1",
      "available_models": [
        {
          "display_name": "Qwen2.5-72B",
          "name": "qwen/qwen-2.5-72b-instruct",
          "max_tokens": 131072
        }
      ]
...

[drivencompassion]

Is there an example of configuring multiple models like openai/chatgpt-4o-latest, o1-mini & o1-preview?

[drivencompassion]

[drivencompassion]

[drivencompassion]

In addition to the LLM providers (e.g., the OpenAI API Key option), may there be an option for OpenRouter?

Run assistant (shift+command+p): show configuration

[bfulop]

I just had the same issue:

I had to add my Openrouter API key in the export OPENAI_API_KEY = env var and restart zed

[drivencompassion]

Where else is env var stored? I get an invalid settings error when trying to add "export OPENAI_API_KEY": "" to settings.json.

[bfulop]

In my case it's in .zshrc file in my home directory.

[bfulop]

If you don't have this set in you shell, then you can set it via the assistant: show configuration and replace the OpenAI key with your openrouter key. In my case because it's set via the shell I cannot edit it.

[azdanov]

Does it work for anyone else?

[phil65]

Seems as if right now we cant "mix" multiple OpenAI(-compatible) APIs because they all would have to share the same API key. Would be great if Zed would get rid of these restrictions so that someone use multiple different OpenAi-compatible APIs simultaniously. :)

[netapy]

+1, is the Zed team on this ? It would massively open up the ability to use any models at token level pricing

[TYTheBeast]

Waiting on this as well, I just discovered Zed and this missing integration is a killer, I have an OpenRouter API key but no Anthropic API key.

[pconerly]

+1 for this. I want to throw deepseek-r1 at my problems!

[bfulop]

I'm using this. Not perfect (it overrides openai) but works:

  "language_models": {
    "openai": {
      "api_url": "https://openrouter.ai/api/v1",
      "available_models": [
        {
          "name": "deepseek/deepseek-r1",
          "display_name": "DeepSeek R1",
          "max_tokens": 64000
        },
        {
          "name": "deepseek/deepseek-chat",
          "display_name": "DeepSeek V3",
          "max_tokens": 64000
        }
      ],
      "version": "1"
    }
  },

Then in my .zshrc I replaced the OPENAI_API_KEY with the openrouter key.

[pconerly]

Thanks! That worked. FYI @drivencompassion , i put these settings into ~/.config/zed/settings.json, and then configured the 'openai' env variable via the Zed UI.

screenshot:

[drivencompassion]

After configuring this method in settings, and in .zshrc replacing key listed in OPENAI_API_KEY with the openrouter key, while keeping OPENAI_API_KEY=sk-or-v1- … of openrouter key, and testing a prompt, going to openrouter activity shows OpenAI: o1-mini model was selected, not DeepSeek R1 Anyone else try this and see different activity?

…

On Jan 21, 2025, at 4:47 AM, Bálint Fülöp ***@***.***> wrote: I'm using this. Not perfect (it overrides openai) but works: "language_models": { "openai": { "api_url": "https://openrouter.ai/api/v1", "available_models": [ { "name": "deepseek/deepseek-r1", "display_name": "DeepSeek R1", "max_tokens": 64000 }, { "name": "deepseek/deepseek-chat", "display_name": "DeepSeek V3", "max_tokens": 64000 } ], "version": "1" } }, Then in my .zshrc I replaced the OPENAI_API_KEY with the openrouter key. — Reply to this email directly, view it on GitHub <#16576 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAFZMFDWHAJA6WSVKXHBKXL2LYJUXAVCNFSM6AAAAABM3LHOWOVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTCOJQGAZTCMA>. You are receiving this because you commented.

[drivencompassion]

~/.config/zed/settings.json, updated … how did you configure the 'openai' env variable via the Zed UI, as DeepSeek R1 is not an option under my config options like your screenshot.

…

---- On Jan 21, 2025 at 12:41:25 PM, Peter Conerly wrote: Thanks! That worked. FYI @drivencompassion , i put these settings into ~/.config/zed/settings.json, and then configured the 'openai' env variable via the Zed UI. screenshot: Screenshot.2025-01-21.at.9.30.36.AM.png (view on web) — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: ***@***.***>

[pconerly]

In the context window (or the "llm chat window") there's a 3-dot button to open options, and one of those options is "Configure". I also gave Zed a restart but I don't know if that was necessary

[drivencompassion]

Tried rebooting and restarting app, but 3-dot options does not have an option for DeepSeek R1, although settings does have: "language_models": { "openai": { "api_url": "https://openrouter.ai/api/v1", "available_models": [ { "name": "deepseek/deepseek-r1", "display_name": "DeepSeek R1", "max_tokens": 64000 }, { "name": "deepseek/deepseek-chat", "display_name": "DeepSeek V3", "max_tokens": 64000 } ], "version": "1" } }

…

---- On Jan 21, 2025 at 1:40:15 PM, Peter Conerly wrote: In the context window there's a 3-dot button to open options, and one of those options is "Configure". I also gave Zed a restart but I don't know if that was necessary Screenshot.2025-01-21.at.10.38.43.AM.png (view on web) — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: ***@***.***>

[amin-abouee]

I've tried both OpenRouter and the experimental Gemini model, but neither worked for me.

"language_models": {
"openai": {
"api_url": "https://openrouter.ai/api/v1",
"available_models": [
{
"name": "deepseek/deepseek-r1",
"display_name": "DeepSeek R1",
"max_tokens": 64000
},
{
"name": "deepseek/deepseek-chat",
"display_name": "DeepSeek V3",
"max_tokens": 64000
}
],
"version": "1"
},
"google": {
"available_models": [
{
"name": "gemini-exp-1206",
"display_name": "Gemini 1206",
"max_tokens": 1000000
},
{
"name": "gemini-2.0-flash-exp",
"display_name": "Gemini 2.0 Flash",
"max_tokens": 1000000
}
]
}
},