release: v0.3.2 — confirm auto-detect endpoints/model ids vs live OpenCode registry

- Verified against /config/providers: Zen (opencode/opencode-go) ->
  https://opencode.ai/zen/v1 + free deepseek-v4-flash-free; Google -> gemini-2.5-flash.
- Fixed stale OpenAI default gpt-4o-mini -> gpt-5.4-mini (model no longer exists);
  updated all docs/examples. 97 tests pass.

Author: doeixd

.opencode/ralph-provider.example.json

@@ -1,12 +1,12 @@
 {
   "supervisor": {
     "baseUrl": "https://api.openai.com/v1",
-    "modelID": "gpt-4o-mini",
+    "modelID": "gpt-5.4-mini",
     "maxToolRounds": 8
   },
   "worker": {
     "agent": "build",
-    "providerID": "anthropic",
-    "modelID": "claude-sonnet-4-5-20250929"
+    "providerID": "opencode",
+    "modelID": "deepseek-v4-flash-free"
   }
 }

CHANGELOG.md

@@ -4,6 +4,13 @@ All notable changes to this project are documented here.
 
 Format based on [Keep a Changelog](https://keepachangelog.com/).
 
+## [0.3.2] - 2026-06-24
+
+### Fixed
+
+- **Confirmed supervisor auto-detect endpoints + model ids** against the live OpenCode provider registry (`/config/providers`): OpenCode Zen (`opencode` / `opencode-go`) → `https://opencode.ai/zen/v1` with the free `deepseek-v4-flash-free` model; Google → `gemini-2.5-flash`. Prefer the free Zen model first.
+- Updated the stale OpenAI default model (`gpt-4o-mini` → `gpt-5.4-mini`) in the supervisor default and all docs/examples.
+
 ## [0.3.1] - 2026-06-24
 
 ### Added

README.md

@@ -95,7 +95,7 @@ Supervisor LLM credentials (provider process). **If you've already authenticated
 
 ```bash
 export RALPH_SUPERVISOR_API_KEY="..."
-export RALPH_SUPERVISOR_MODEL="gpt-4o-mini"   # + RALPH_SUPERVISOR_BASE_URL for non-OpenAI endpoints
+export RALPH_SUPERVISOR_MODEL="gpt-5.4-mini"   # + RALPH_SUPERVISOR_BASE_URL for non-OpenAI endpoints
 ```
 
 Or `.opencode/ralph-provider.json` — see [`.opencode/ralph-provider.example.json`](.opencode/ralph-provider.example.json). Check what resolved with `curl http://127.0.0.1:8787/api/health` (`supervisor.ready` / `model` / `source`).

docs/GETTINGSTARTEDGUIDE.md

@@ -249,7 +249,7 @@ The provider calls an LLM for supervisor chat turns (separate from your worker m
 
 ```bash
 export RALPH_SUPERVISOR_API_KEY="sk-..."
-export RALPH_SUPERVISOR_MODEL="gpt-4o-mini"
+export RALPH_SUPERVISOR_MODEL="gpt-5.4-mini"
 # export RALPH_SUPERVISOR_BASE_URL="https://api.openai.com/v1"   # for non-OpenAI endpoints
 ```
 

docs/INSTALLATION.md

@@ -165,5 +165,5 @@ Use this path when you need to review or vendor every file.
 
   ```bash
   export RALPH_SUPERVISOR_API_KEY="..."
-  export RALPH_SUPERVISOR_MODEL="gpt-4o-mini"
+  export RALPH_SUPERVISOR_MODEL="gpt-5.4-mini"
   ```

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@doeixd/opencode-ralph-rlm",
-  "version": "0.3.1",
+  "version": "0.3.2",
   "description": "Ralph RLM v0.2: OpenCode supervisor provider + loop engine + worker plugin",
   "type": "module",
   "workspaces": [

packages/engine/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@doeixd/opencode-ralph-rlm-engine",
   "private": true,
-  "version": "0.3.1",
+  "version": "0.3.2",
   "description": "Ralph RLM loop engine — verify, rollover, OpenCode SDK worker lifecycle",
   "license": "MIT",
   "repository": {

packages/engine/src/setup.ts

@@ -313,7 +313,7 @@ export async function setupProject(options: SetupOptions): Promise<SetupResult>
         jsonBlock({
           supervisor: {
             baseUrl: "https://api.openai.com/v1",
-            modelID: "gpt-4o-mini",
+            modelID: "gpt-5.4-mini",
             maxToolRounds: 8,
           },
           worker: {

packages/provider/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@doeixd/opencode-ralph-rlm-provider",
   "private": true,
-  "version": "0.3.1",
+  "version": "0.3.2",
   "description": "Ralph RLM OpenAI-compatible supervisor provider (Nitro)",
   "license": "MIT",
   "repository": {

packages/provider/server/lib/opencode-auth.ts

@@ -20,21 +20,28 @@ export type DetectedSupervisorCreds = {
   source: string;
 };
 
-/** Known OpenAI-compatible endpoints + a sensible default model per provider. */
+/**
+ * Known OpenAI-compatible endpoints + a sensible (tool-calling capable) default
+ * model per provider. Endpoints and model ids confirmed against the live
+ * OpenCode provider registry (`/config/providers`) and provider docs:
+ * - OpenCode Zen gateway uses OPENCODE_API_KEY; free models (`*-free`) are $0.
+ * - Google Gemini exposes an OpenAI-compatible endpoint; key = GEMINI_API_KEY.
+ */
 const PROVIDER_ENDPOINTS: Record<string, { baseUrl: string; model: string }> = {
-  // OpenCode Zen gateway (the user's keyed Zen access).
-  "opencode-go": { baseUrl: "https://opencode.ai/zen/v1", model: "deepseek-v4-flash" },
+  // OpenCode Zen — free model (same gateway/key as opencode-go).
+  opencode: { baseUrl: "https://opencode.ai/zen/v1", model: "deepseek-v4-flash-free" },
+  "opencode-go": { baseUrl: "https://opencode.ai/zen/v1", model: "deepseek-v4-flash-free" },
   // Google Gemini OpenAI-compatible endpoint.
   google: {
     baseUrl: "https://generativelanguage.googleapis.com/v1beta/openai",
     model: "gemini-2.5-flash",
   },
   // OpenAI (only used if stored as a static api key, not OAuth).
-  openai: { baseUrl: "https://api.openai.com/v1", model: "gpt-4o-mini" },
+  openai: { baseUrl: "https://api.openai.com/v1", model: "gpt-5.4-mini" },
 };
 
-/** Order to prefer when several keyed providers are available. */
-const PROVIDER_PRIORITY = ["opencode-go", "google", "openai"];
+/** Order to prefer when several keyed providers are available (free Zen first). */
+const PROVIDER_PRIORITY = ["opencode", "opencode-go", "google", "openai"];
 
 function authFilePath(): string {
   const override = process.env.RALPH_OPENCODE_AUTH_PATH?.trim();