opencode-jev-model-router 0.1.0 → 0.1.2

package/README.md

@@ -2,17 +2,26 @@
 
 Automatic per-turn model picking for [OpenCode](https://opencode.ai), powered by [Jev](https://typesafe.ai). Built for **[OpenCode Go](https://opencode.ai/docs/go/)**: easy turns burn generous models; hard turns get Luna.
 
+[npm](https://www.npmjs.com/package/opencode-jev-model-router) · [GitHub](https://github.com/aaronshaf/opencode-jev-model-router)
+
 ## Quick start
 
 **1. Install the plugin**
 
 ```bash
-git clone https://github.com/aaronshaf/opencode-jev-model-router.git
-cd opencode-jev-model-router
-npm install && npm run build
-opencode plugin "$(pwd)" -g
+opencode plugin opencode-jev-model-router -g
+```
+
+Or add it to `~/.config/opencode/opencode.json`:
+
+```json
+{
+  "plugin": ["opencode-jev-model-router"]
+}
 ```
 
+Pin a version if you prefer: `"opencode-jev-model-router@0.1.2"`.
+
 **2. Add your Jev key** (OpenCode usually does not see shell `export`s)
 
 ```bash
@@ -20,10 +29,16 @@ printf '%s\n' "$JEV_KEY" > ~/.config/opencode/opencode-jev-router.key
 chmod 600 ~/.config/opencode/opencode-jev-router.key
 ```
 
-**3. Optional:** copy defaults you can edit later
+**3. Optional config** — built-in defaults are **OpenCode Go only** (Muse / DeepSeek Flash / Luna). Copy an example only if you want to change that:
 
 ```bash
-cp opencode-jev-router.example.json ~/.config/opencode/opencode-jev-router.json
+# Go-only (same as built-in defaults)
+curl -fsSL https://raw.githubusercontent.com/aaronshaf/opencode-jev-model-router/main/opencode-jev-router.example.json \
+  -o ~/.config/opencode/opencode-jev-router.json
+
+# Go + Claude + Codex (see below)
+curl -fsSL https://raw.githubusercontent.com/aaronshaf/opencode-jev-model-router/main/opencode-jev-router.go-claude-codex.example.json \
+  -o ~/.config/opencode/opencode-jev-router.json
 ```
 
 **4. Restart OpenCode**, then check:
@@ -34,13 +49,19 @@ cp opencode-jev-router.example.json ~/.config/opencode/opencode-jev-router.json
 
 You want: `Jev key present; routing ON; …`
 
+## How routing picks models
+
+Jev chooses a **tier** (`fast` / `balanced` / `strong` / `long`). The plugin then picks the first eligible model from that tier’s `model` + `fallbacks` in the plugin config (or the built-in Go defaults if you have no config file).
+
+If your current model is **not** in that list, the turn is **pinned** — no routing. So Claude/Codex only participate after you add them to a tier (or use the example above).
+
 ## Day to day
 
-1. Select a Go model the router manages (DeepSeek Flash, Muse, Luna, …). Avoid one-off pins like `hy3` if you want routing.
+1. Start on a **managed** model (one listed in the defaults or your config). Unlisted models pin.
 2. Chat as usual.
 3. Watch for a toast such as `Routed to opencode-go/muse-spark-1.3-contributor · jev fast 98% 280ms`.
 
-| Kind of ask | Typical tier | Default model |
+| Kind of ask | Typical tier | Default model (Go) |
 |---|---|---|
 | Typo, rename, “say hi” | `fast` | Muse Spark |
 | Normal feature / fix | `balanced` | DeepSeek V4.1 Flash |
@@ -49,6 +70,12 @@ You want: `Jev key present; routing ON; …`
 
 If Jev is unreachable, your current model stays put.
 
+### Also using Claude or Codex?
+
+Connect them in OpenCode (`/connect`), then use a config that lists those models under a tier — e.g. [opencode-jev-router.go-claude-codex.example.json](./opencode-jev-router.go-claude-codex.example.json): Go stays primary for fast/balanced; `strong` falls back to Sonnet then Codex if Luna is unavailable.
+
+Adjust the `anthropic/…` and `openai/…` IDs to whatever `/models` shows for your account.
+
 ### Force a tier in the prompt
 
 ```text
@@ -62,7 +89,7 @@ use balanced for this endpoint
 
 - `/jev-off` — stay on whatever model you picked  
 - `/jev-on` — turn automatic routing back on  
-- Or pick an unmanaged model (e.g. `opencode-go/hy3`) to **pin**
+- Or pick an unmanaged model (not in your tier lists) to **pin**
 
 ### When a model hits its Go allowance
 

package/opencode-jev-router.go-claude-codex.example.json

@@ -0,0 +1,60 @@
+{
+  "$schema": "https://github.com/aaronshaf/opencode-jev-model-router/schemas/config.schema.json",
+  "enabled": true,
+  "allowProjectModels": false,
+  "tiers": {
+    "fast": {
+      "model": "opencode-go/muse-spark-1.3-contributor",
+      "aliases": ["fast", "flash", "muse"],
+      "fallbacks": [
+        "opencode-go/mimo-v2.5",
+        "opencode-go/deepseek-v4.1-flash",
+        "opencode-go/glm-5.3-flash"
+      ]
+    },
+    "balanced": {
+      "model": "opencode-go/deepseek-v4.1-flash",
+      "aliases": ["balanced"],
+      "fallbacks": [
+        "opencode-go/muse-spark-1.3-contributor",
+        "opencode-go/mimo-v2.5",
+        "opencode-go/qwen3.7-plus"
+      ]
+    },
+    "strong": {
+      "model": "opencode-go/gpt-5.6-luna",
+      "aliases": ["strong", "luna", "sonnet", "opus", "codex"],
+      "fallbacks": [
+        "anthropic/claude-sonnet-4-20250514",
+        "openai/gpt-5.1-codex",
+        "opencode-go/kimi-k2.7-code",
+        "opencode-go/qwen3.7-plus"
+      ]
+    },
+    "long": {
+      "model": "opencode-go/kimi-k3",
+      "enabled": false,
+      "aliases": ["long", "kimi", "k3"],
+      "fallbacks": ["opencode-go/kimi-k2.7-code"]
+    }
+  },
+  "routing": {
+    "failOpen": true,
+    "maxPromptBytes": 16384,
+    "timeoutMs": 1500,
+    "deadlineMs": 3000,
+    "maxRetries": 1,
+    "minimumConfidence": 0.3,
+    "uncertainCeiling": "balanced",
+    "downgradeMaxContextTokens": 20000
+  },
+  "quota": {
+    "cooldownHours": 5
+  },
+  "history": {
+    "enabled": true,
+    "retainPrompt": false,
+    "maxEntries": 20
+  },
+  "echoRouting": true
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "opencode-jev-model-router",
-  "version": "0.1.0",
+  "version": "0.1.2",
   "description": "Jev-based automatic per-turn model routing for OpenCode",
   "type": "module",
   "license": "MIT",
@@ -22,6 +22,7 @@
     "dist",
     "schemas",
     "opencode-jev-router.example.json",
+    "opencode-jev-router.go-claude-codex.example.json",
     "README.md",
     "LICENSE"
   ],