Added OpenAI-compatible proxy for Claude Code

CHANGELOG.md

@@ -1,3 +1,5 @@
+- Added OpenAI-compatible proxy: profiles can now route Claude Code through any OpenAI-compatible API provider (OpenAI, Azure, Groq, DeepSeek, Ollama, local models, etc.)
+
 # 1.1.7 (2026-05-22)
 
 - Fixed profile combo not refreshing when profiles are added or removed after session start

CLAUDE.md

@@ -176,7 +176,8 @@ The plugin follows an MVC structure for session management:
 | `ui/common/` | Shared input components: `AtCompletionPopup` — @-triggered path popup; `AtPathHighlighter` — blue token highlight; `FileDropHandler` — DnD + Ctrl+V → @path insertion; `ShortcutMatcher` — key→shortcut match with KEY_TYPED suppression; `TextComponentDecorator` — wires all the above; `DecoratedTextArea/TextField`, `TextContextMenu`, `RangeHighlightable` |
 | `settings/` | `ClaudeCodePreferences` — NbPreferences wrapper (keys: `contextMenuSessionMode` default CONTINUE_LAST, `sessionListLimit` default 30); `ClaudeCodeOptionsPanelController` / `ClaudeCodeOptionsPanel` — Tools→Options (General + Profiles tabs); `ClaudeProfile`, `ClaudeProfileStore` — named profiles with isolated `CLAUDE_CONFIG_DIR`, auth, proxy, extra env vars; `ClaudeProjectProperties` — per-project profile assignment |
 | `actions/` | `ClaudeCodeAction` — toolbar button; `OpenWithClaudeAction` — project node context menu; `PromptFavoriteAction` — dynamically registered action per favorite for Keymap API |
-| `io.github.nbplugins.claudecodegui.mcp` | `MCPSseServer` — Jetty `/sse` `/messages` `/hook`; `NetBeansMCPHandler` — MCP dispatcher + PreToolUse hook; `tools/` — `OpenDiff`, `PermissionPromptTool`, `DiffTabTracker`, `GetDiagnostics`, `GetOpenEditors`, `GetCurrentSelection`, `OpenFile` |
+| `io.github.nbplugins.claudecodegui.mcp` | `MCPSseServer` — Jetty `/sse` `/messages` `/hook` `/openai-proxy/*`; `NetBeansMCPHandler` — MCP dispatcher + PreToolUse hook; `tools/` — `OpenDiff`, `PermissionPromptTool`, `DiffTabTracker`, `GetDiagnostics`, `GetOpenEditors`, `GetCurrentSelection`, `OpenFile` |
+| `io.github.nbplugins.claudecodegui.openaiproxy` | `OpenAIProxyConfig` — per-session config snapshot; `OpenAIProxyServlet` — translates Anthropic → OpenAI requests and responses; `AnthropicToOpenAITranslator` — stateless translation logic including streaming SSE state machine |
 | `org.openbeans.claude.netbeans` | **Legacy package — do not add new code here and avoid modifying existing code.** Contains: `ClaudeCodeStatusService`, `ClaudeCodeStatusLineElement`, `EditorUtils`, `NbUtils`; `tools/` — `AsyncHandler`, `AsyncResponse`, `GetWorkspaceFolders`, `CheckDocumentDirty`, `SaveDocument`, `CloseTab`, `CloseAllDiffTabs` |
 
 ### Session lifecycle
@@ -255,6 +256,8 @@ After any action the originating `ClaudeSessionTab` is re-activated automaticall
 | "Start new session when opening with Claude" checkbox | `contextMenuSessionMode` | `CONTINUE_LAST` | Checked → `SessionMode.NEW`; unchecked → `SessionMode.CONTINUE_LAST`. Controls the mode used by the toolbar button and project context menu. |
 | "Session list limit" spinner (1–500) | `sessionListLimit` | 30 | Max number of past sessions shown in the resume list. |
 
+**Note:** Changes to profile connection parameters (API key, base URL, proxy settings) take effect only after restarting the session. This applies to all connection types, including OpenAI Compatible Proxy.
+
 ### Registration
 `layer.xml` registers the Options category (position 1500) and toolbar action. Icon PNGs are generated at build time from `cc-gui-icon.svg` via the Groovy script in `pom.xml` using Apache Batik.
 

README.md

@@ -13,6 +13,7 @@ NetBeans Claude Code GUI is a NetBeans IDE plugin that embeds the Claude Code CL
 - **Prompt history and favorites** — recall past prompts with Ctrl+Up/Down; save reusable prompts as favorites with optional keyboard shortcuts
 - **File attachments** — attach files via `@path` tokens with auto-completion popup, drag-and-drop, and clipboard paste support
 - **Multiple profiles** — run Claude Code under separate accounts or API keys for different projects, each with an isolated config directory, authentication, proxy, and model settings
+- **OpenAI-compatible proxy** — route Claude Code through any OpenAI-compatible provider (OpenAI, Azure OpenAI, Groq, DeepSeek, Ollama, LM Studio, and others) via a built-in translation proxy; no changes to Claude Code CLI required
 - **Session management** — start new sessions, continue the last session, or resume a specific past session; sessions persist across IDE restarts
 - **Markdown Preview** — live-rendered markdown tab for plan files and MCP-initiated previews; includes a find bar (Ctrl+F) and font zoom (Alt+Scroll)
 - **Auto Plan Preview** — when Claude writes a plan file, a live preview tab opens automatically as soon as you accept the diff

docs/testing-openai-proxy-opencode-zen.md

@@ -0,0 +1,60 @@
+# Testing OpenAI Compatible Proxy with OpenCode Zen
+
+OpenCode Zen provides free OpenAI-compatible models that can be used to test the proxy feature without spending money.
+
+## 1. Registration and API Key
+
+1. Go to https://opencode.ai/auth
+2. Sign in with Google or GitHub
+3. Add billing details (a card is required even for free models — no charges for free-tier usage)
+4. Copy your API key from the account dashboard
+
+## 2. Plugin Profile Setup
+
+Open **Tools → Options → Claude Code → Profiles**, create a new profile:
+
+| Field | Value |
+|---|---|
+| Connection type | OpenAI Compatible Proxy |
+| Base URL | `https://opencode.ai/zen` |
+| API Key | your key from step 1 |
+
+> **Note:** do not add `/v1` to the Base URL — the plugin appends it automatically when routing requests (`/v1/chat/completions`) and fetching models (`/v1/models`).
+
+## 3. Free Models
+
+Click **Model Aliases** in the profile editor. Use **Fetch** to load the current model list, or add manually:
+
+| Model name | Model ID |
+|---|---|
+| Big Pickle | `opencode/big-pickle` |
+| DeepSeek V4 Flash Free | `opencode/deepseek-v4-flash-free` |
+| Nemotron 3 Super Free | `opencode/nemotron-3-super-free` |
+
+All three models are $0 for input and output, context window 200K+. The list is time-limited while model teams collect feedback — check https://opencode.ai/docs/zen/ for the current list.
+
+Map Claude model tiers to OpenCode models, for example:
+
+| Alias (Claude tier) | Target model |
+|---|---|
+| `claude-sonnet-4-5` | `opencode/big-pickle` |
+| `claude-haiku-4-5` | `opencode/deepseek-v4-flash-free` |
+| `claude-opus-4-7` | `opencode/nemotron-3-super-free` |
+
+Claude Code sends requests using names like `claude-sonnet-4-5` — Model Aliases redirects them to the actual provider model. Without an alias the request will fail because OpenCode does not recognize Claude model names.
+
+## 4. Start a Session
+
+Start a Claude Code session with this profile. The plugin will:
+- Set `ANTHROPIC_BASE_URL` to the internal proxy (`http://127.0.0.1:<mcp-port>/openai-proxy/<uuid>`)
+- Translate Anthropic Messages API requests → OpenAI Chat Completions
+- Translate responses back, including streaming SSE and tool use
+
+## 5. What to Verify
+
+- **Basic conversation** — send a simple question, verify a response arrives
+- **Streaming** — response should appear incrementally, not all at once
+- **Tool use** — ask Claude to read or edit a file, verify the tool_use → tool_result round-trip works
+- **Debug logging** — enable debug mode in preferences, then check `~/.netbeans/28/var/log/messages.log` for lines showing request/response metadata (headers, model, token counts) with no message content
+- **Two simultaneous sessions** — open two sessions with the same profile, verify both work independently
+- **Profile edit during session** — change the API key while a session is running, verify the running session is unaffected; restart to apply the new key

docs/user-manual.md

@@ -78,7 +78,7 @@ Contains the input area, **▶ Send** / **✖ Cancel** buttons, and **☰ Histor
 
 Press **Shift+Tab** anywhere in the prompt panel to cycle through modes.
 
-**Model selector** — lists the models available in the current session. Standard entries (`sonnet`, `opus`, `haiku`) are discovered by opening the `/model` menu. For **Other API** profiles, any models assigned the `custom` alias in **Model Aliases** also appear here, shown by their full provider ID. Selecting a model switches to it immediately.
+**Model selector** — lists the models available in the current session. Standard entries (`sonnet`, `opus`, `haiku`) are discovered by opening the `/model` menu. For **Claude-compatible API** profiles, any models assigned the `custom` alias in **Model Aliases** also appear here, shown by their full provider ID. Selecting a model switches to it immediately.
 
 On the right side of the status bar:
 
@@ -487,9 +487,12 @@ Select the authentication type using the radio buttons:
 | **Claude Managed** | None | Claude manages authentication itself. Run `claude` in a terminal once to set it up, or Claude will prompt you on first launch. Use `/login` in the session terminal to re-authenticate. |
 | **Subscription** | OAuth token | Obtain the token at [claude.ai](https://claude.ai) (requires Pro, Max, Team, or Enterprise subscription). **Note:** OAuth tokens expire periodically and must be regenerated and re-entered manually. For a smoother experience, prefer **Claude Managed** instead. |
 | **Claude API** | API key | Obtain the key at [console.anthropic.com](https://console.anthropic.com) (requires an Anthropic Console account with API access). |
-| **Other API** | API key + Base URL | Use the API key and base URL provided by your third-party provider. |
+| **Claude-compatible API** | API key + Base URL | Use the API key and base URL provided by your third-party provider (Anthropic-compatible endpoint). |
+| **OpenAI-compatible API** | API key + Base URL | Route Claude Code through any OpenAI-compatible provider. The plugin starts an internal proxy that translates Anthropic API calls to OpenAI Chat Completions format. See [OpenAI-compatible API](#openai-compatible-api) below. |
 
-### Model aliases (Other API only)
+> **Note:** Changes to connection parameters (API key, Base URL) take effect only after restarting the session.
+
+### Model aliases (Claude-compatible API only)
 
 If your provider names models differently from Anthropic's standard names, the plugin cannot match them to the `sonnet`, `opus`, and `haiku` aliases used by Claude Code. In that case, set the alias to the actual model ID used by your provider (for example, with an `anthropic/` prefix).
 
@@ -546,6 +549,49 @@ The dialog shows a table with three columns:
 
 **Note:** Claude Code communicates with the plugin via `localhost`. External proxies are not aware of this. If you fill in the **NO_PROXY** field, make sure it includes `localhost` — otherwise the plugin integration will stop working. If you leave the field empty, the plugin adds `localhost` automatically.
 
+### OpenAI-compatible API
+
+Claude Code sends requests in Anthropic Messages API format. Many AI providers — including OpenAI, Azure OpenAI, Groq, DeepSeek, and local models via Ollama or LM Studio — use OpenAI Chat Completions format instead. The **OpenAI-compatible API** connection type bridges this gap by running an internal HTTP proxy inside the plugin.
+
+**How it works:**
+
+1. When a session starts, the plugin registers an internal proxy path within the already-running MCP server.
+2. Claude Code is directed to send requests to that path (`ANTHROPIC_BASE_URL=http://127.0.0.1:<mcp-port>/openai-proxy/<session-id>`).
+3. The proxy translates each Anthropic Messages API request to OpenAI Chat Completions format, forwards it to your provider, and translates the response back — including streaming and tool calls.
+4. When the session stops, the proxy path is deregistered. No separate port or process is used.
+
+**Setup:**
+
+1. Create a new profile and select **OpenAI-compatible API** as the connection type.
+2. Fill in **Base URL** — the base URL of your provider's API (e.g. `https://api.openai.com/v1`, `http://localhost:11434/v1` for Ollama).
+3. Fill in **API Key** — the API key for your provider (enter any non-blank value if the provider does not require a key).
+4. Open **Model Aliases…** and map Claude model tiers to your provider's model names:
+
+| Alias | Provider model example |
+|-------|----------------------|
+| `sonnet` | `gpt-4o` |
+| `opus`   | `gpt-4o` |
+| `haiku`  | `gpt-4o-mini` |
+
+Without model aliases, Claude Code will send Anthropic model names (e.g. `claude-sonnet-4-5`) to your provider. Most OpenAI-compatible providers will reject these — configure aliases to avoid errors.
+
+**Provider examples:**
+
+| Provider | Base URL | Notes |
+|----------|----------|-------|
+| OpenAI | `https://api.openai.com/v1` | Standard OpenAI key |
+| Azure OpenAI | `https://<resource>.openai.azure.com/openai/deployments/<deployment>` | Azure API key |
+| Groq | `https://api.groq.com/openai/v1` | Groq API key |
+| DeepSeek | `https://api.deepseek.com/v1` | DeepSeek API key |
+| Ollama (local) | `http://localhost:11434/v1` | Any value for API key |
+| LM Studio (local) | `http://localhost:1234/v1` | Any value for API key |
+
+**Limitations:**
+
+- Not all OpenAI-compatible models support all Claude Code features (tool use, large context windows). Using an incompatible model may cause errors.
+- Images in conversations (base64 content) are forwarded in OpenAI vision format; provider must support it.
+- Changes to the profile's Base URL or API Key take effect only after restarting the session.
+
 ### Extra environment variables
 
 Arbitrary `KEY=VALUE` pairs injected into the Claude process environment. Applied last — they override all other profile variables. Use this for provider-specific configuration.

pom.xml

@@ -6,7 +6,7 @@
 
     <groupId>io.github.nbplugins</groupId>
     <artifactId>netbeans-plugin-claude-code-gui</artifactId>
-    <version>1.1.4-SNAPSHOT</version>
+    <version>1.2.7-SNAPSHOT</version>
     <packaging>nbm</packaging>
     <name>Netbeans Plugin Claude Code GUI</name>
     <description>NetBeans plugin that provides a GUI for Claude Code CLI — run claude code sessions directly inside the IDE.</description>

src/main/java/io/github/nbplugins/claudecodegui/ClaudeCodeInstaller.java

@@ -247,4 +247,24 @@ public int getServerPort() {
     public boolean isLockFileValid() {
         return false; // lock file no longer used
     }
+
+    @Override
+    public void registerOpenAIProxy(String uuid, String baseUrl, String apiKey,
+            io.github.nbplugins.claudecodegui.settings.ProxyConfiguration proxy) {
+        if (mcpServer != null) {
+            mcpServer.registerOpenAIProxy(uuid, baseUrl, apiKey, proxy);
+        }
+    }
+
+    @Override
+    public void deregisterOpenAIProxy(String uuid) {
+        if (mcpServer != null) {
+            mcpServer.deregisterOpenAIProxy(uuid);
+        }
+    }
+
+    @Override
+    public io.github.nbplugins.claudecodegui.openaiproxy.OpenAIProxyConfig getOpenAIProxyConfig(String uuid) {
+        return mcpServer != null ? mcpServer.getOpenAIProxyConfig(uuid) : null;
+    }
 }

src/main/java/io/github/nbplugins/claudecodegui/mcp/MCPSseServer.java

@@ -4,6 +4,9 @@
 
 import com.fasterxml.jackson.databind.JsonNode;
 import com.fasterxml.jackson.databind.ObjectMapper;
+import io.github.nbplugins.claudecodegui.openaiproxy.OpenAIProxyConfig;
+import io.github.nbplugins.claudecodegui.openaiproxy.OpenAIProxyServlet;
+import io.github.nbplugins.claudecodegui.settings.ProxyConfiguration;
 import org.eclipse.jetty.server.Server;
 import org.eclipse.jetty.server.ServerConnector;
 import org.eclipse.jetty.servlet.ServletContextHandler;
@@ -62,6 +65,13 @@ public class MCPSseServer {
     private final ConcurrentHashMap<String, BlockingQueue<String>> sessionQueues =
             new ConcurrentHashMap<>();
 
+    /**
+     * Registry of active OpenAI-compatible proxy sessions, keyed by UUID.
+     * Each entry is a config snapshot taken at session start.
+     */
+    private final ConcurrentHashMap<String, OpenAIProxyConfig> openAIProxies =
+            new ConcurrentHashMap<>();
+
     /**
      * Creates a new MCPSseServer backed by the given MCP handler.
      *
@@ -103,6 +113,7 @@ public boolean start(int port) {
             ctx.addServlet(hookHolder, "/hook");
             ctx.addServlet(new ServletHolder(new StopServlet()), "/stop");
             ctx.addServlet(new ServletHolder(new PermissionRequestServlet()), "/permission-request");
+            ctx.addServlet(new ServletHolder(new OpenAIProxyServlet(this)), "/openai-proxy/*");
 
             server.setHandler(ctx);
             server.setRequestLog((request, response) ->
@@ -141,6 +152,43 @@ public void stop() {
      */
     public int getPort() { return port; }
 
+    // -------------------------------------------------------------------------
+    // OpenAI proxy registry
+    // -------------------------------------------------------------------------
+
+    /**
+     * Registers an OpenAI-compatible proxy session.
+     *
+     * @param uuid   unique session identifier
+     * @param baseUrl provider base URL
+     * @param apiKey  provider API key
+     * @param proxy   proxy settings from the profile (used by the servlet's HttpClient)
+     */
+    public void registerOpenAIProxy(String uuid, String baseUrl, String apiKey, ProxyConfiguration proxy) {
+        openAIProxies.put(uuid, new OpenAIProxyConfig(baseUrl, apiKey, proxy));
+        LOGGER.log(Level.FINE, "OpenAI proxy registered: uuid={0}", uuid);
+    }
+
+    /**
+     * Deregisters an OpenAI-compatible proxy session.
+     *
+     * @param uuid unique session identifier
+     */
+    public void deregisterOpenAIProxy(String uuid) {
+        openAIProxies.remove(uuid);
+        LOGGER.log(Level.FINE, "OpenAI proxy deregistered: uuid={0}", uuid);
+    }
+
+    /**
+     * Returns the proxy config for the given UUID, or {@code null} if not found.
+     *
+     * @param uuid unique session identifier
+     * @return config snapshot or {@code null}
+     */
+    public OpenAIProxyConfig getOpenAIProxyConfig(String uuid) {
+        return openAIProxies.get(uuid);
+    }
+
     /**
      * Returns {@code true} while the Jetty server is running.
      *