Auto detect models with just an env var:

1. export OLLAMA_BASE_URL="localhost:11434"
2. run opencode
3. /models
4. Your local models will be detected.
5. Manual config can still be set if you want like usual

The implementation now follows all coding guidelines:

1. No let variables - Using const and immutable patterns throughout
2. No else statements - Refactored to use early returns and separate if statements
3. Error handling with .catch() - Replaced all try/catch blocks with promise chains
4. Precise types - Defined TagsResponse type to avoid any
5. Concise naming - Used envUrl, base, url instead of verbose names
6. Single function logic - IIFE used only for scoping the detection flow
7. Runtime APIs - Using native fetch which is appropriate for this use case

Key Implementation Details

Provider Detection (provider.ts:230-294):
ollama: async (provider) => {
// 1. Detect server URL (env var or fallbacks)
const envUrl = process.env["OLLAMA_BASE_URL"]
const url = await (async () => {
if (envUrl) return envUrl
for (const base of ["http://localhost:11434", "http://127.0.0.1:11434"]) {
const ok = await fetch(${base}/api/tags, { signal: AbortSignal.timeout(1000) })
.then((r) => r.ok)
.catch(() => false)
if (ok) return base
}
return null
})()

// 2. Fetch and auto-discover models
// 3. Add models to provider
// 4. Return baseURL for OpenAI-compatible endpoint

}

Database Setup (provider.ts:411-423):

• Ensures Ollama provider exists in database before custom loaders run
• Sets default npm package if not specified
• No else statements, clean control flow

Features

✅ Zero-configuration - Works without any config file
✅ Automatic model discovery - Fetches all models from /api/tags
✅ Remote server support - Via OLLAMA_BASE_URL environment variable
✅ Fallback detection - Tries localhost and 127.0.0.1 automatically
✅ Custom configuration - Optional override for display names and settings

Documentation (providers.mdx:560-643)

• Clear zero-config quick start guide
• Explains auto-detection priority (env var → fallbacks)
• Documents model discovery behavior
• Provides optional manual configuration examples
• Uses realistic model names (llama3.2:latest, qwen2.5-coder:7b)

Testing Verified ✅

Tested with your remote Ollama server at http://192.168.2.26:11434:

• ✅ Environment variable detection works
• ✅ All models local on another server or localhost where detected
• ✅ Models appear in OpenCode model list
• ✅ Connection successful to /v1/chat/completions