No key found for some Agents on 0.7.7

[brentp913]

What happened?

After upgrading LibreChat from 0.7.6 to 0.7.7, all of my Agents that used non-OpenAI models give me the error message:
"No key found. Please provide a key and try again."

The console log in docker shows the following:

librechat-api  | 2025-03-26 19:22:41 error: [handleAbortError] AI response error; aborting request: {"type":"no_user_key"}

Additionally, when I try the Deepseek R1 model, I get these errors:

librechat-api  | 2025-03-26 20:07:30 error: [api/server/controllers/agents/client.js #sendCompletion] Operation aborted Deepseek API key not found. Please set the DEEPSEEK_API_KEY environment variable or pass the key into "apiKey" field.
librechat-api  | 2025-03-26 20:07:30 error: [api/server/controllers/agents/client.js #sendCompletion] Unhandled error type Deepseek API key not found. Please set the DEEPSEEK_API_KEY environment variable or pass the key into "apiKey" field.
librechat-api  | 2025-03-26 20:07:30 error: [handleAbortError] AI response error; aborting request: Deepseek API key not found. Please set the DEEPSEEK_API_KEY environment variable or pass the key into "apiKey" field.

In the previous version of LibreChat, it used the apiKey's that I set in librechat.yaml were used. I suspect that because some of the model names I'm using under custom endpoints align with actual services, that 0.7.7 is trying to use the API keys from the environment variables, which I don't set because I use LiteLLM as a proxy.

Version Information

ghcr.io/danny-avila/librechat-rag-api-dev   v0.4.0                        7fd465954a4f   2 weeks ago     7.79GB
ghcr.io/danny-avila/librechat               v0.7.7                        92d57359fc25   2 weeks ago     888MB

Steps to Reproduce

1. Run 0.7.7 through Docker. Using the following librechat.yaml:

version: 1.2.1
interface:
  endpointsMenu: true
  modelSelect: true
  parameters: true
  sidePanel: true
  presets: true
  prompts: true
  bookmarks: true
  multiConvo: true
  agents: true
endpoints:
  agents:
    disableBuilder: false
    capabilities: []
  custom:
    - name: 'Anthropic'
      apiKey: 'removed-for-privacy'
      baseURL: 'http://our-litellm-server:4000'
      models:
        default:
          - 'anthropic-claude-3-5-sonnet'
          - 'anthropic-claude-3-7-sonnet'
          - 'anthropic-claude-3-opus'
          - 'anthropic-claude-3-5-haiku'
        fetch: false
      titleConvo: true
      titleModel: 'meta-llama3-1-8b-instruct'
      titleMessageRole: 'user'
      summarize: false
      summaryModel: 'meta-llama3-1-8b-instruct'
      forcePrompt: false
      modelDisplayLabel: 'Anthropic'
      dropParams:
        - presence_penalty
        - frequency_penalty
      iconURL: anthropic

Note: Our LiteLLM server accesses Claude through AWS Bedrock if that matters at all.

2. Create a new agent
3. In the model settings, select "Provider": "Anthropic", "Model": "anthropic-claude-3-5-sonnet". Alternatively, "Provider": "Deepseek", with "Model": "deepseek-r1" will also fail and give additional error messages.
4. Send anything to the model.

What browsers are you seeing the problem on?

Microsoft Edge

Relevant log output

This error shows up a bunch:

{"level":"error","message":"[handleAbortError] AI response error; aborting request: {\"type\":\"no_user_key\"}","stack":"Error: {\"type\":\"no_user_key\"}\n    at getUserKey (/app/api/server/services/UserService.js:47:11)\n    at process.processTicksAndRejections (node:internal/process/task_queues:95:5)\n    at async initializeClient (/app/api/server/services/Endpoints/anthropic/initialize.js:12:7)\n    at async initializeAgentOptions (/app/api/server/services/Endpoints/agents/initialize.js:128:19)\n    at async initializeClient (/app/api/server/services/Endpoints/agents/initialize.js:208:25)\n    at async AgentController (/app/api/server/controllers/agents/request.js:47:24)\n    at async /app/api/server/routes/agents/chat.js:37:5"}


This error shows up for Deepseek:

{"level":"error","message":"[api/server/controllers/agents/client.js #sendCompletion] Operation aborted Deepseek API key not found. Please set the DEEPSEEK_API_KEY environment variable or pass the key into \"apiKey\" field.","stack":"Error: Deepseek API key not found. Please set the DEEPSEEK_API_KEY environment variable or pass the key into \"apiKey\" field.\n    at new ChatDeepSeek (/app/node_modules/@langchain/deepseek/dist/chat_models.cjs:371:19)\n    at StandardGraph.initializeModel (/app/node_modules/@librechat/agents/dist/cjs/graphs/Graph.cjs:202:23)\n    at new StandardGraph (/app/node_modules/@librechat/agents/dist/cjs/graphs/Graph.cjs:65:32)\n    at Run.createStandardGraph (/app/node_modules/@librechat/agents/dist/cjs/run.cjs:50:31)\n    at new Run (/app/node_modules/@librechat/agents/dist/cjs/run.cjs:40:39)\n    at Run.create (/app/node_modules/@librechat/agents/dist/cjs/run.cjs:61:16)\n    at createRun (/app/api/server/controllers/agents/run.js:75:14)\n    at runAgent (/app/api/server/controllers/agents/client.js:680:21)\n    at AgentClient.chatCompletion (/app/api/server/controllers/agents/client.js:714:13)\n    at AgentClient.sendCompletion (/app/api/server/controllers/agents/client.js:373:16)"}
{"level":"error","message":"[api/server/controllers/agents/client.js #sendCompletion] Unhandled error type Deepseek API key not found. Please set the DEEPSEEK_API_KEY environment variable or pass the key into \"apiKey\" field.","stack":"Error: Deepseek API key not found. Please set the DEEPSEEK_API_KEY environment variable or pass the key into \"apiKey\" field.\n    at new ChatDeepSeek (/app/node_modules/@langchain/deepseek/dist/chat_models.cjs:371:19)\n    at StandardGraph.initializeModel (/app/node_modules/@librechat/agents/dist/cjs/graphs/Graph.cjs:202:23)\n    at new StandardGraph (/app/node_modules/@librechat/agents/dist/cjs/graphs/Graph.cjs:65:32)\n    at Run.createStandardGraph (/app/node_modules/@librechat/agents/dist/cjs/run.cjs:50:31)\n    at new Run (/app/node_modules/@librechat/agents/dist/cjs/run.cjs:40:39)\n    at Run.create (/app/node_modules/@librechat/agents/dist/cjs/run.cjs:61:16)\n    at createRun (/app/api/server/controllers/agents/run.js:75:14)\n    at runAgent (/app/api/server/controllers/agents/client.js:680:21)\n    at AgentClient.chatCompletion (/app/api/server/controllers/agents/client.js:714:13)\n    at AgentClient.sendCompletion (/app/api/server/controllers/agents/client.js:373:16)"}
{"level":"error","message":"[handleAbortError] AI response error; aborting request: Deepseek API key not found. Please set the DEEPSEEK_API_KEY environment variable or pass the key into \"apiKey\" field.","stack":"Error: Deepseek API key not found. Please set the DEEPSEEK_API_KEY environment variable or pass the key into \"apiKey\" field.\n    at new ChatDeepSeek (/app/node_modules/@langchain/deepseek/dist/chat_models.cjs:371:19)\n    at StandardGraph.initializeModel (/app/node_modules/@librechat/agents/dist/cjs/graphs/Graph.cjs:202:23)\n    at new StandardGraph (/app/node_modules/@librechat/agents/dist/cjs/graphs/Graph.cjs:65:32)\n    at Run.createStandardGraph (/app/node_modules/@librechat/agents/dist/cjs/run.cjs:50:31)\n    at new Run (/app/node_modules/@librechat/agents/dist/cjs/run.cjs:40:39)\n    at Run.create (/app/node_modules/@librechat/agents/dist/cjs/run.cjs:61:16)\n    at createRun (/app/api/server/controllers/agents/run.js:75:14)\n    at runAgent (/app/api/server/controllers/agents/client.js:680:21)\n    at AgentClient.chatCompletion (/app/api/server/controllers/agents/client.js:714:13)\n    at AgentClient.sendCompletion (/app/api/server/controllers/agents/client.js:373:16)"}

Screenshots

No response

Code of Conduct

• I agree to follow this project's Code of Conduct

[danny-avila]

Seems to be an issue with LiteLLM and not LibreChat, double check your config there.

[danny-avila]

In the previous version of LibreChat, it used the apiKey's that I set in librechat.yaml were used. I suspect that because some of the model names I'm using under custom endpoints align with actual services, that 0.7.7 is trying to use the API keys from the environment variables, which I don't set because I use LiteLLM as a proxy.

This could be the issue, try naming the custom endpoints differently than the main provider/endpoint values i.e. not "Anthropic" as that is reserved (normally anthropic), same with "OpenAI", etc. The custom endpoint names may be getting normalized as all lowercased.

Be sure to also set an apiKey value, even if the value is "none." There is no such LC behavior that does what you're describing on the "model" level.

[brentp913]

This could be the issue, try naming the custom endpoints differently than the main provider/endpoint values i.e. not "Anthropic" as that is reserved (normally anthropic), same with "OpenAI", etc. The custom endpoint names may be getting normalized as all lowercased.

Okay, I changed the names in my librechat.yaml:

-    - name: 'Anthropic'
+    - name: 'Bedrock Anthropic'

This fixes the issue on 0.7.7. When I roll back to 0.7.6, both Anthropic and Bedrock Anthropic work just fine. I can definitely change this on my end, and it resolves the issue, so thank you!

Be sure to also set an apiKey value, even if the value is "none." There is no such LC behavior that does what you're describing on the "model" level.

I am setting the apiKey for every model under endpoints: -> custom:, but I don't set the API keys anywhere else.

Seems to be an issue with LiteLLM and not LibreChat, double check your config there.

Everything seems to be working fine from the LiteLLM side. I tried a few different things with 3 other AI tools (e.g., continue.dev, another chat app, and an internal company tool), all with the same exact access keys, and the other tools don't seem to be having any issues with API keys. Even inside of LibreChat the models are all working fine. The only time the bad API key message appears is when I try to use those models inside "Agents". In the regular context, no issue at all. I realize that this is NOT definitive proof that it's an issue with LibreChat, and that it COULD still be an issue with my configuration or LiteLLM, but I wanted to mention it in case it helps narrow down the cause at all.

Like I mentioned above though, just renaming the providers to non-built-in values seems to fix it, so I'm good with that! Thank you again for your assistance!