Skip to main content

Custom Endpoints don’t work, and haven’t for months

BoltAI Mobile

this is not a new issue, i can reach the endpoint with other apps.

Bolt.ai fails to even attempt to reach the host because im watching the hosts logs and nothing is hitting the endpoint.

It fails using the https domain name url AND the IP address.

It is able to probe the models successfully but errors out with “can’t reach host” error when i attempt to chat. I am confident that this is a bolt.ai problem

Extremely disappointing, I was hoping itd work by now

Posted by \} ·
Status: Completed6 comments

Log in to comment and vote

Comments6

  • Daniel Nguyen changed status to Completed
    Team•

    Aug 27, 2025

    Pinned

    I’ve improved support for local & custom servers in build 20. Update when you have a chance.

    • kernkraft

      •

      Aug 28, 2025

      Thank you so much, it works now. Currently bolt.ai is the best general purpose chat app that I have found for iOS. The next best option I’ve found is TypingMind, which is pretty mature, but it’s a PWA which Apple seems to regret supporting in the first place. It has more features than Bolt.ai but the restrictions placed on PWAs render them irrelevant due to added friction and lack of sane backgrounding. Bolt.ai in it’s current state offers enough benefit for me to switch.

      PS: Sorry for my tone in the initial report, I have been very frustrated with my software experience on Apple devices lately, almost exclusively due to Apple themselves, and sometimes my frustration leaks out on smaller independent developers.

  • Daniel Nguyen

    Team•

    Aug 25, 2025

    Thanks for the feedback. The mobile app is still in beta and it’s hard for me to cover all use cases. It’s free! Please be patient :)

    Currently the app uses AI SDK by Vercel (https://ai-sdk.dev) and it requires a back-end server to handle AI service routing / streaming / processing. This is a known limitation at the moment.

    In your case, since you use tailscale, it’s likely that you can access the endpoint from your computer but BoltAI server wasn’t able to.

    Please share with me your setup so I can debug further.

    • kernkraft

      •

      Aug 28, 2025

      What was odd was that it was able to retrieve the models so /v1/models clearly worked. It didn’t occur to me to try tailscale funnel which would have worked due to the tscl endpoint becoming publically reachable, but I would then have had to create another proxy to add api key auth

      • Daniel Nguyen

        Team•

        Aug 29, 2025

        Yes, the model fetching call is direct, not via the proxy (not needed). Anyway in the latest build, this is irrelevant as all connections are direct now. I’ve found a way to remove the proxy server.

  • \}

    •

    Aug 24, 2025

    my endpoint is a tailscale endpoint.

    https://next.ghost-morpho.ts.net/lms/v1

    i tried it without using the lms subpath as well, still fails.

    i don’t need an API key, but the app forces me to have one, however there are other apps that do this as well and lm studios endpoint will gladly take the request with an unneeded and made up api key