Skip to main content

No Base URLs options for AI vendors, can't use at work now

Feature Requests

We use a dumb proxy at my job that worked perfectly fine in v1. I could point Anthropic at our base URL, pop in our API key, and it worked as expected. This is not an OpenAI-compatible API and doesn’t work as a “custom” service.

Unless I’m missing something, there is no more base URL option for service vendors in v2 and now I can’t use BoltAI at work. I can’t express how frustrating and disappointing this is, and oddly enough my license expires today. I love this app, but I can’t justify renewing until I can use it at my job again.

Posted by jd503 ·
Status: Completed19 comments

Log in to comment and vote

Comments19

  • Daniel Nguyen changed status to Completed
    Team•

    Jan 7

    Pinned

    I’ve fixed this in v2.5.0. Please confirm. Thank you.

    • Daniel Nguyen

      Team•

      Jan 7

      cc @Jourin Enwurk

      • Jourin Enwurk

        •

        Jan 7

        Can confirm it working, adding custom url to both openai and anthropic can get correct model list and chat without problems.

  • Daniel Nguyen changed status to In Progress
    Team•

    Jan 2

    Pinned

    I’m reopening this as @Jourin Enwurk is still running into the issue.

  • Daniel Nguyen changed status to Completed
    Team•

    Nov 8, 2025

    Pinned

    I haven’t heard back from you. LMK if it works for you. Feel free to reopen the issue if it doesn’t.

    • jd503

      •

      Nov 11, 2025

      Hi Daniel, I’ll check it out at work tomorrow. Glad to see it added either way! Between this and how great the iOS sync has been, I’ve renewed my license 🥳

    • Jourin Enwurk

      •

      Nov 29, 2025

      Hi, I think it is not working again in build 41, even if I set custom url for openai and anthropic, in my network request I see it is still requesting the official api, as a result I only get “Connection failed: Invalid API key - authentication failed” and cannot move forward. I am using a custom litellm endpoint provided from work, and as tested can work with claude code and other opensource client. Thanks for taking a look.

      • Daniel Nguyen

        Team•

        Nov 30, 2025

        Sorry. I’ll take a deeper look later today.

        • Jourin Enwurk

          •

          Jan 2

          I think this so far still hasn’t fixed (already on the Jan 1st version), should I make a new feedback to keep track? I also want to state why this is important for routing service(openrouter or in this case the litellm instance from my work) because newer models like claude-4.5 and gemini, you need to stream back all reasoning token to make tool calling function, if you go through the normal openai-compatible endpoint setup, you will get a reasoning missing error. Thanks and happy new year!

          • Daniel Nguyen

            Team•

            Jan 2

            Wait. I think I fixed the wrong bug. Do you use the OpenAI provider with custom base URL, or do you use Custom Provider (OpenAI-compatible endpoint)? They are different and are handled differently in BoltAI

            • Jourin Enwurk

              •

              Jan 2

              I am wanting to use openai and anthropic with a custom BaseURL, something like xxx.com/v1 that is fully openai and anthropic compatible. In other client and in v1 of bolt, I can just change api.openai.com/v1 to this xxx.com/v1 to use the proxying url(as for my usecase, a litellm instance provided from work), same for anthropic, and I can confirm this way of settings works on other clients or using openai/anthropic python SDK. However, when I try to set xxx.com/v1 as a BaseURL in bolt v2, it is not working and throw me a “Connection failed: Invalid API key - authentication failed” error, tracing the network log, I can see the app is not honoring the setting in BaseURL, but still using the default api.openai.com/v1 and the default anthropic url, as a result, I cannot go forward and save the provider.

              • Daniel Nguyen

                Team•

                Jan 2

                I see. Any reason why you don’t use the OpenAI-compatible server? (The Custom provider)?

                In BoltAI, the OpenAI provider is meant to be used with the official OpenAI API key etc as it has a strict schema check and it supports more endpoints than a custom server (speech etc etc). The app expects these endpoints to work and not all custom servers respect it.

                Ideally it should work regardless but in this case, I think the root cause is somewhere else.

                Can you share a screenshot of the error message etc? Sorry I find it easier to understand the issue this way. Is it on the chat interface, or during setup?

                • Jourin Enwurk

                  •

                  Jan 2

                  Hi, I understand this, so my main problem isn’t with openai, as most openai models through proxy works fine with openai compatible, what doesn’t work is anthropic, as the openai compatible isn’t streaming back the encrypted reasoning chain as the api expected and will fail on first message. That’s why for proxying service, as it is basically just passing through the api, it should be the same as the official api, I would expect to set it in base url. Here is a screenshot, I didn’t use the real url for obvious reason, but you can see the app is not requesting the url I set.

                  • Jourin Enwurk

                    •

                    Jan 2

                    And since it is not requesting the url I set, of course the api key is different making the auth fail, so I cannot click “Next“ and finish setting up the service. So I would say, it is less of a problem with custom api, but more of a bug that the app is not honoring Base URL settings during setup(not sure about during chat as I am cannot proceed to that step yet.)

                  • Daniel Nguyen

                    Team•

                    Jan 2

                    Thank you. I found the issue. You’re right that during setting up the service, BoltAI still use the hard-coded base URL for fetching model lists. It does honor the base URL when sending the chat message.

                    Working on the fix right now. I will release a new version soon.

                    • Jourin Enwurk

                      •

                      Jan 2

                      Thanks again for your hard work and the wonderful app, will report back upon update!

  • Daniel Nguyen

    Team•

    Oct 28, 2025

    Hey Jonathan. Sorry for the inconvenience. I’ve been porting features from v1 to v2 and it’s easy to forget details like this. BoltAI has too many features.

    I’ll prioritize this and release a new version soon.

    To be clear you can continue to use v1 without renewing. It’s a perpetual license with an update plan. You can renew later when the feature is released in v2.

    Stay tuned and thanks for your patience.

    • jd503

      •

      Nov 4, 2025

      I definitely should’ve kept v1, but I overwrote the app out of excitement and double owned myself on this one. I spend a lot of my day job losing track of esoteric AI implementation details and this is a bit of an edge case, so I get it.

      It’s not much, but I promise at least one (1) annual renewal for it if you add it 🙏

      • Daniel Nguyen

        Team•

        Nov 4, 2025

        I just released v2.0.15 to add support for it. Can you give it a try? The format is different than v1 though. You need to include the version path in the Base URL for example https://api.anthropic.com/v1

        Let me know if it works