fetch_models_for_opencode

fetch_models_for_opencode

A small Go tool that fetches the model list from an OpenAI-compatible endpoint (GET <baseurl>/models) and generates an opencode.jsonc config file for OpenCode.

OpenCode only allows adding models manually via opencode.jsonc – there is no way to import them automatically. This is tedious when your provider offers many models.

Example: I run this tool against OmniRoute, a local AI gateway that exposes hundreds of models and combo models through a single OpenAI-compatible endpoint – exactly the models I want to use in OpenCode. Instead of writing every entry by hand, this tool generates the complete provider config in one go.

How it works

• Fetches /models from the given base URL
• Creates a provider entry (@ai-sdk/openai-compatible) with all models
• Context/output limits are read from model metadata (fallbacks: 131072 / 8192)

Usage

go run . --provider OmniRoute --baseurl http://localhost:20128/v1 --token sk-1234

Output: an opencode.jsonc file ready to use — either V2-native (default) or V1-compatible.

Formats

Flag	Top-level key	Provider key	Config key
--format v2 (default)	"provider"	"npm"	"options"
--format v1	"providers"	"package"	"settings"

Both shapes are accepted by OpenCode V2. Use v2 for new configs, v1 when you need compatibility with existing tooling or the V1-compatible global config.

# V2-native (default)
go run . --provider OmniRoute --baseurl http://localhost:20128/v1

# V1-compatible
go run . --provider OmniRoute --baseurl http://localhost:20128/v1 --format v1

Flags

Flag	Default	Description
-provider	Myprovider	Provider name
-baseurl	– (required)	Base URL, e.g. http://localhost:8080/v1
-output	opencode.jsonc	Output file
-token	–	API token sent as Authorization: Bearer and x-api-key
-apiurl	<baseurl>/models	Alternative URL for fetching the model list
-format	v2	Output format: v2 (default) or v1

Build

go build -o fetch_models_for_opencode .