Read custom providers from `opencode.json` from central `~/.config/opencode/` config instead of root of each project

[GitMurf]

@thdxr I followed the instructions from the readme for custom options for a openai compatible provider (as you pointed to in your readme) and that works when I add it to a opencode.json file in the root of a project. But I want to be able to use it for all projects but cannot seem to get it to read it from ~/.config/opencode/ or from ~ (home) or from ~/.opencode/ etc... I tried all the potential opencode config locations with no luck.

Am I doing something wrong or do I need to copy the custom opencode.json to each of my project locations in every place I want to use OpenCode with my custom provider options?

Thanks!!

Originally posted by @GitMurf in #140

Figured easier for you all to track @thdxr if I open a new issue for this. Thanks!

p.s. I have dev repo pulled locally and seeing your minute by minute pushes ;) so whenever you are ready for me to test and I fetch and pull and confirm if working!

[GitMurf]

Similar issue opened here #153 (closed to consolidate here)

[thdxr]

yep im going to implement this - my idea was letting you mirror the same structure as models.dev inside ~/.config/opencode/providers

does that work for you?

https://github.com/sst/models.dev/tree/dev/providers

[GitMurf]

does that work for you?

@thdxr I was actually thinking of requesting exactly that but thought it may sound too complicated 🤣 Perfect!

One additional note, I have my OpenAI api key set as env variable for other tools, but I don't want it to show up in the model chooser for opencode. In previous "go version" of opencode we could explicitly disable providers. Is there a way you could implement that? I only want specific providers to show up when using /model but if OpenAI api key env variable is set, opencode seems to auto "load it". Not a huge deal, but would be nice if this was possible!

[thdxr]

good idea can you make a seperate issue for that

[GitMurf]

good idea can you make a seperate issue for that

@thdxr see #157

[thdxr]

implemented in 0.1.72 and updated README

[GitMurf]

@thdxr the new providers config setup inside of ~/.config/opencode/providers/... works well except for one part. It does not have a way to pass in options like the baseURL for proxying / openai compatible custom providers.

I don't know if there is a better way, but here is the patch that got it working for me if I have a provider.toml like this:

name = "GitHub Copilot"
npm = "@ai-sdk/openai-compatible"

[options]
baseURL = "http://localhost:3000"

Here is the patch that fixes it:

diff --git a/packages/opencode/src/provider/provider.ts b/packages/opencode/src/provider/provider.ts
index c155891..1121a0b 100644
--- a/packages/opencode/src/provider/provider.ts
+++ b/packages/opencode/src/provider/provider.ts
@@ -150,6 +150,7 @@ export namespace Provider {
         name: provider.name ?? existing?.name ?? providerID,
         env: provider.env ?? existing?.env ?? [],
         models: existing?.models ?? {},
+        ...(provider.options ? { options: provider.options } : {}),
       }
 
       for (const [modelID, model] of Object.entries(provider.models ?? {})) {
@@ -178,6 +179,12 @@ export namespace Provider {
       database[providerID] = parsed
     }
 
+    for (const [providerID, provider] of Object.entries(database)) {
+      if ("options" in provider && provider.options) {
+        mergeProvider(providerID, provider.options, "config")
+      }
+    }
+
     const disabled = await GlobalConfig.get().then(
       (cfg) => new Set(cfg.disabled_providers ?? []),
     )

Let me know if you have any questions or thoughts?

[thdxr]

yeah this makes sense i'll fix

[thdxr]

fixed in 0.1.81 i believe