OpenCode plugin provider for 9Router — FREE AI Router & Token Saver. 40+ providers, 100+ models.

Registers 9Router as a custom provider in OpenCode with auto-discovery of models.

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["@vheins/opencode-9router@latest"]
}

1. Add the plugin to opencode.json
2. Restart OpenCode
3. /models → select a 9Router model

The plugin will auto-discover models from http://localhost:20128 (default).

• Auto-discover models — Models from 9Router are automatically detected on startup
• Smart caching — Discovery results cached for 3 hours (configurable) for instant subsequent loads
• Stale fallback — If the backend is unreachable, returns cached models instead of failing
• Configurable timeout — Adjustable discovery timeout (default 30s) for slow backends
• Dynamic model list — All models from 9Router are available, including custom combos
• Combo multimodal defaults — Combo models (owned_by: "combo") that expose no capability info are forced to accept text/image/audio input with tool calling, reasoning, 256K context, 128K output, and low/medium/high/xhigh/max/minimal/thinking variants
• OpenAI-compatible — Uses @ai-sdk/openai-compatible
• Type-safe — Uses the config hook for provider registration that conforms to the OpenCode config schema

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["@vheins/opencode-9router@latest"]
}

No need to define a provider manually — the plugin registers it automatically.

Add one or more 9Router-family providers with custom options:

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["@vheins/opencode-9router@latest"],
  "provider": {
    "9router": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Your Provider",
      "options": {
        "baseURL": "https://your-9router-endpoint.example.com/v1",
        "apiKey": "{env:ROUTER_API_KEY}",
        "cache": true,
        "cacheTTL": 10800000,
        "discoveryTimeout": 60000
      }
    },
    "9router-local": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Local Router",
      "options": {
        "baseURL": "http://127.0.0.1:20128/v1",
        "apiKey": "sk-...",
        "cache": true,
        "discoveryTimeout": 60000
      }
    }
  }
}

Each provider key must start with 9router. The npm must be @ai-sdk/openai-compatible. The name is a display label shown in OpenCode.

Never commit real API keys. Replace the baseURL/apiKey above with your own endpoint and a placeholder such as {env:ROUTER_API_KEY} (or a dummy value like sk-...). Real keys belong in environment variables, not in opencode.json or this README.

{
  "provider": {
    "9router": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Remote 9Router",
      "options": {
        "baseURL": "https://example.com/v1",
        "apiKey": "{env:ROUTER_API_KEY}",
        "cache": true,
        "discoveryTimeout": 30000
      }
    }
  }
}

export ROUTER_API_KEY=your-api-key-here
opencode

{
  "provider": {
    "9router": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Local Router",
      "options": {
        "baseURL": "http://localhost:20128",
        "cache": true,
        "discoveryTimeout": 60000
      }
    },
    "9router-remote": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Remote API",
      "options": {
        "baseURL": "https://your-9router-endpoint.example.com/v1",
        "apiKey": "{env:ROUTER_API_KEY}",
        "cacheTTL": 3600000
      }
    }
  }
}

{
  "provider": {
    "9router": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Local Router",
      "options": {
        "baseURL": "http://localhost:20128",
        "apiKey": "sk-...",
        "cache": false,
        "discoveryTimeout": 30000
      }
    }
  }
}

/models
  → Find the 9Router provider
  → Pick any model (e.g., kr/claude-sonnet-4.5, cc/claude-opus-4-7)

opencode.json "plugin": ["@vheins/opencode-9router@latest"]
  ↓
Plugin loads at startup detects 9Router-family providers in config
  ↓
For each provider:
  1. Check local cache (~/.cache/opencode-9router/discovery-*.json)
     └─ Fresh cache (≤3h) → return cached models instantly
  2. If cache miss/stale:
     └─ GET {baseURL}/v1/models with 30s timeout (configurable)
     └─ If OK → register live models + save to cache
     └─ If fail → fallback to stale cache (if exists), else skip
  ↓
config hook creates/updates provider with discovered models
  ↓
Provider appears in /models

Cache files are stored at ~/.cache/opencode-9router/discovery-{base64url}.json, one per unique baseURL. The cache directory also stores the models.dev capability catalog (models-dev.json).

• Combos count as enriched only on an exact models.dev segment match (no more loose substring hits like vision → gpt-4-turbo-vision)
• Ignore bare /models/info capability stubs ({tools: true}) for combos, so backends that return stubs still get forced multimodal defaults

• Detect combo models via owned_by: "combo" (fallback: no provider prefix)
• Force multimodal defaults on combos with no resolved capability info: text/image/audio input, tool calling, reasoning, 256K context, 128K output
• Add low/medium/high/xhigh/max/minimal/thinking variants, plus search: false, context_length, and max_completion_tokens
• Version discovery cache so older entries are refetched

• Add INFO-level logging for cache hit/miss, fetch status, timeout, and stale fallback

• Model discovery cache — 3-hour TTL with stale fallback when backend is unreachable
• Configurable timeout — 30s default, adjustable via discoveryTimeout option
• Provider options — cache, cacheTTL, discoveryTimeout
• New constant: DISCOVERY_CACHE_TTL, DISCOVERY_TIMEOUT

• Support multiple 9Router-family providers (9router, 9router-local, etc.)
• Fallback models catalog via models.dev API
• Per-model capability resolution (vision, tools, reasoning)

• Basic auto-discovery from localhost:20128
• Fallback models when API unreachable
• Single provider support

git clone https://github.com/vheins/opencode-9router
cd opencode-9router
npm install
npm run build

node test-minimal.mjs
opencode models 9router --print-logs

After publishing to npm, verify the plugin works end-to-end:

opencode models 9router-local --print-logs --log-level DEBUG

This runs model discovery against the published package with full debug logging to confirm caching, timeout, and capability resolution are working correctly.

npm login
npm version patch  # or minor, major
npm publish --access public
git push --follow-tags

opencode-9router/
  src/
    plugin.ts         # Plugin entry (provider registration + config hook)
    discovery.ts      # Model discovery pipeline
    capabilities.ts   # Capability resolution (catalog + per-model API)
    cache.ts          # Discovery + models.dev cache
    types.ts          # Shared types
    utils.ts          # Shared helpers
    constants.ts      # Models, defaults, prefixes
  dist/               # Compiled output (generated)
  package.json        # npm package config
  tsconfig.json       # TypeScript config
  README.md           # This file
  README.id.md        # Bahasa Indonesia version
  LICENSE             # MIT

• OpenCode Plugin Docs
• OpenCode Custom Provider
• 9Router GitHub
• 9Router Website