Open Router - Hugging Face - UI Improvement

.github/workflows/ci.yml

@@ -56,3 +56,45 @@ jobs:
           path: artifacts/*.vsix
           if-no-files-found: error
           retention-days: 14
+
+  jetbrains:
+    name: JetBrains build, test, verify
+    runs-on: ubuntu-latest
+
+    strategy:
+      fail-fast: false
+      matrix:
+        project: [jetbrains, jetbrains-apiclient]
+
+    steps:
+      - name: Check out repository
+        uses: actions/checkout@v4
+
+      - name: Set up JDK 21
+        uses: actions/setup-java@v4
+        with:
+          distribution: temurin
+          java-version: 21
+
+      - name: Cache Gradle
+        uses: actions/cache@v4
+        with:
+          path: |
+            ~/.gradle/caches
+            ~/.gradle/wrapper
+          key: gradle-${{ runner.os }}-${{ matrix.project }}-${{ hashFiles(format('apps/{0}/**/*.gradle.kts', matrix.project), format('apps/{0}/gradle.properties', matrix.project), format('apps/{0}/gradle/wrapper/gradle-wrapper.properties', matrix.project)) }}
+          restore-keys: |
+            gradle-${{ runner.os }}-${{ matrix.project }}-
+            gradle-${{ runner.os }}-
+
+      - name: Compile, test, build, and verify plugin
+        working-directory: apps/${{ matrix.project }}
+        run: ./gradlew compileKotlin compileTestKotlin test buildPlugin verifyPlugin --no-daemon
+
+      - name: Upload plugin zip (smoke build)
+        uses: actions/upload-artifact@v4
+        with:
+          name: codesetu-${{ matrix.project }}-zip
+          path: apps/${{ matrix.project }}/build/distributions/*.zip
+          if-no-files-found: error
+          retention-days: 14

.github/workflows/release-jetbrains.yml

@@ -48,9 +48,9 @@ jobs:
           echo "tag=jetbrains-v${VERSION}" >> "$GITHUB_OUTPUT"
           echo "Computed version: ${VERSION}"
 
-      - name: Build plugin
+      - name: Build and verify plugin
         working-directory: apps/jetbrains
-        run: ./gradlew -PpluginVersion=${{ steps.version.outputs.version }} buildPlugin --no-daemon
+        run: ./gradlew -PpluginVersion=${{ steps.version.outputs.version }} buildPlugin verifyPlugin --no-daemon
 
       - name: Verify plugin zip exists
         run: |

.github/workflows/release-vscode.yml

@@ -15,6 +15,10 @@ jobs:
   release:
     name: Build and publish
     runs-on: ubuntu-latest
+    # Job-level env so the Open VSX step's `if` can read it (step-level env is
+    # not visible in that same step's `if` condition).
+    env:
+      OVSX_PAT: ${{ secrets.OVSX_PAT }}
 
     steps:
       - name: Check out repository
@@ -72,8 +76,6 @@ jobs:
 
       - name: Publish to Open VSX
         if: ${{ !inputs.dry-run && env.OVSX_PAT != '' }}
-        env:
-          OVSX_PAT: ${{ secrets.OVSX_PAT }}
         run: |
           VSIX=$(ls artifacts/codesetu-*.vsix | head -n1)
           pnpm dlx ovsx publish "$VSIX" --pat "$OVSX_PAT"

README.md

@@ -7,7 +7,7 @@
 [![Discord](https://img.shields.io/badge/Discord-Join-5865F2?logo=discord&logoColor=white)](https://discord.gg/sjVKU8cpC6)
 [![PRs Welcome](https://img.shields.io/badge/PRs-welcome-brightgreen.svg)](CONTRIBUTING.md)
 
-A **Copilot / Cursor alternative** designed for Indian developers, enterprises, public sector teams, and air-gapped deployments. Bring your own model — CodeSetu works with **Sarvam, OpenAI-compatible APIs (Ollama, vLLM, OpenRouter, SGLang)**, and local self-hosted deployments. AI chat, repo-aware context, selected-code actions, inline FIM completions, tool-calling, and an extensible plugin and skill system across VSCode and JetBrains.
+A **Copilot / Cursor alternative** designed for Indian developers, enterprises, public sector teams, and air-gapped deployments. Bring your own model — CodeSetu works with **Sarvam, Hugging Face (any served chat model, dedicated endpoints, or self-hosted TGI), OpenAI-compatible APIs (Ollama, vLLM, OpenRouter, SGLang)**, and local self-hosted deployments. AI chat, repo-aware context, selected-code actions, inline FIM completions, tool-calling, and an extensible plugin and skill system across VSCode and JetBrains.
 
 **Highlights**: AI chat in IDE · Repo-aware context · Selected-code actions · Inline (FIM) code completions · Provider setup and diagnostics · Workspace skills/checks · Air-gapped friendly · Hindi / Indic-aware · Plugin + skill SDK · 100% open-source (Apache 2.0)
 
@@ -27,7 +27,7 @@ This repository is a pnpm + Gradle monorepo organized as:
 
 - Node.js 18+
 - pnpm 9+
-- A provider — Sarvam API key, OpenRouter key, or a local OpenAI-compatible endpoint (Ollama, vLLM, SGLang)
+- A provider — Sarvam API key, a Hugging Face token (`hf_…`), OpenRouter key, or a local OpenAI-compatible endpoint (Ollama, vLLM, SGLang)
 
 ## Setup
 
@@ -50,10 +50,10 @@ Marketplace, Open VSX, and private VSIX hosting, see
 
 The VSCode extension reads these settings:
 
-- `codesetu.provider` - `sarvam` or `openai-compatible`
-- `codesetu.apiKey` - optional provider API key
-- `codesetu.baseUrl` - optional OpenAI-compatible base URL
-- `codesetu.model` - optional model name
+- `codesetu.provider` - `sarvam`, `openai-compatible`, or `huggingface`
+- `codesetu.baseUrl` - optional base URL (e.g. `https://router.huggingface.co/v1`, a dedicated HF endpoint, or a local server)
+- `codesetu.model` - optional model name (for Hugging Face, the model repo id, e.g. `meta-llama/Llama-3.3-70B-Instruct`)
+- API key — set via the **CodeSetu: Setup Provider** command (stored in the OS secret store), or the `SARVAM_API_KEY` / `HF_TOKEN` / `CODESETU_API_KEY` environment variables
 - `codesetu.inlineCompletions.enabled` - enable FIM inline completions
 - `codesetu.chat.maxTokens` / `codesetu.chat.temperature`
 

apps/jetbrains-apiclient/.intellijPlatform/self-update.lock

@@ -1 +1 @@
-2026-05-27
+2026-05-28

apps/jetbrains-apiclient/build.gradle.kts

@@ -1,3 +1,5 @@
+import org.jetbrains.intellij.platform.gradle.IntelliJPlatformType
+
 plugins {
   // Kotlin version must be >= the version bundled in the target IntelliJ Platform.
   // IDEA 2025.2 (IC-252.x) ships with Kotlin 2.2.x.
@@ -36,6 +38,8 @@ dependencies {
       intellijIdeaCommunity("2025.2.5")
     }
     instrumentationTools()
+    // CLI used by the `verifyPlugin` task (JetBrains Plugin Verifier).
+    pluginVerifier()
   }
 }
 
@@ -61,6 +65,14 @@ intellijPlatform {
     }
   }
   buildSearchableOptions = false
+  // `verifyPlugin` runs the JetBrains Plugin Verifier. Pin it to the build
+  // target so it's deterministic and always resolvable (recommended() can pick
+  // an unreleased IDE that fails to download).
+  pluginVerification {
+    ides {
+      ide(IntelliJPlatformType.IntellijIdeaCommunity, "2025.2.5")
+    }
+  }
   publishing {
     token = providers.environmentVariable("JETBRAINS_MARKETPLACE_TOKEN")
   }

apps/jetbrains/build.gradle.kts

@@ -1,3 +1,5 @@
+import org.jetbrains.intellij.platform.gradle.IntelliJPlatformType
+
 plugins {
   // Kotlin version must be >= the version bundled in the target IntelliJ Platform.
   // IDEA 2025.2 (IC-252.x) ships with Kotlin 2.2.x.
@@ -35,6 +37,8 @@ dependencies {
       intellijIdeaCommunity("2025.2.5")
     }
     instrumentationTools()
+    // CLI used by the `verifyPlugin` task (JetBrains Plugin Verifier).
+    pluginVerifier()
   }
 }
 
@@ -65,6 +69,14 @@ intellijPlatform {
   // compute a settings-search index; our scaffold has no settings panel, so
   // it's pure overhead. Re-enable once we add settings UI.
   buildSearchableOptions = false
+  // `verifyPlugin` runs the JetBrains Plugin Verifier. Pin it to the build
+  // target so it's deterministic and always resolvable (recommended() can pick
+  // an unreleased IDE that fails to download).
+  pluginVerification {
+    ides {
+      ide(IntelliJPlatformType.IntellijIdeaCommunity, "2025.2.5")
+    }
+  }
   publishing {
     token = providers.environmentVariable("JETBRAINS_MARKETPLACE_TOKEN")
   }

apps/jetbrains/gradle.properties

@@ -6,7 +6,7 @@ org.gradle.parallel=true
 # Plugin version baseline. CI overrides with -PpluginVersion=<major>.<minor>.<run_number>.
 # Bump the major.minor here when you want a semantic version bump; CI keeps the
 # patch incrementing automatically.
-pluginVersion=0.1.0
+pluginVersion=0.2.0
 
 # Tell Gradle's toolchain detection where to find JDK 21 installed via Homebrew.
 # Falls back to other auto-detected JDKs if this path doesn't exist on a contributor's machine.

apps/jetbrains/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@codesetu/jetbrains",
-  "version": "0.1.0",
+  "version": "0.2.0",
   "private": true,
   "description": "JetBrains plugin for CodeSetu. This Kotlin/Gradle project is integrated into the pnpm workspace via thin script delegations — actual build/test runs through Gradle.",
   "license": "Apache-2.0",

apps/jetbrains/src/main/kotlin/ai/codesetu/model/CodeSetuModels.kt

@@ -5,7 +5,8 @@ import kotlinx.serialization.Serializable
 
 enum class ProviderKind(val id: String) {
   SARVAM("sarvam"),
-  OPENAI_COMPATIBLE("openai-compatible");
+  OPENAI_COMPATIBLE("openai-compatible"),
+  HUGGING_FACE("huggingface");
 
   companion object {
     fun fromId(id: String): ProviderKind =

apps/jetbrains/src/main/kotlin/ai/codesetu/prompts/PromptBuilder.kt

@@ -4,13 +4,27 @@ import ai.codesetu.model.IdeActionId
 import ai.codesetu.model.IdeContextPayload
 import ai.codesetu.model.WorkspaceInstruction
 
-fun buildSystemMessage(instructions: List<WorkspaceInstruction>): String =
-  if (instructions.isEmpty()) {
-    "You are CodeSetu, an AI coding assistant for Indian developers. Be concise, correct, practical, and privacy-aware."
-  } else {
-    "You are CodeSetu, an AI coding assistant for Indian developers. Be concise, correct, practical, and privacy-aware.\n\nFollow applicable workspace guidance when it helps the user's request."
+fun buildSystemMessage(instructions: List<WorkspaceInstruction>): String {
+  val parts = mutableListOf(
+    "You are CodeSetu, an AI coding assistant for Indian developers. Be concise, correct, practical, and privacy-aware.",
+    "Use the supplied IDE context as the source of truth. Ask for missing context when needed.",
+  )
+
+  if (instructions.isNotEmpty()) {
+    parts.add(formatWorkspaceInstructions(instructions))
   }
 
+  return parts.joinToString("\n\n")
+}
+
+private fun formatWorkspaceInstructions(instructions: List<WorkspaceInstruction>): String {
+  val rendered = instructions.joinToString("\n\n") { instruction ->
+    listOf("${instruction.name} (${instruction.id})", instruction.description, instruction.body)
+      .joinToString("\n")
+  }
+  return "Workspace instructions\n\n$rendered"
+}
+
 fun buildActionInstruction(actionId: IdeActionId): String =
   when (actionId) {
     IdeActionId.EXPLAIN -> "Explain the selected code clearly and concisely. Include key control flow, inputs, outputs, and risks."

apps/jetbrains/src/main/kotlin/ai/codesetu/provider/CodeSetuProviderClient.kt

@@ -4,6 +4,7 @@ import ai.codesetu.model.ChatCompletionRequest
 import ai.codesetu.model.ChatCompletionResponse
 import ai.codesetu.model.ChatCompletionChunk
 import ai.codesetu.model.ChatMessage
+import ai.codesetu.model.ProviderKind
 import ai.codesetu.settings.CodeSetuSettingsState
 import ai.codesetu.settings.resolveCodeSetuModel
 import java.net.URI
@@ -24,12 +25,12 @@ class CodeSetuProviderClient(
       messages = messages,
       maxTokens = maxTokens,
       temperature = temperature,
-      reasoningEffort = "low",
+      reasoningEffort = reasoningEffortFor(state.provider),
       json = json,
     )
     val request = HttpRequest.newBuilder()
       .uri(URI.create(state.baseUrl.trimEnd('/') + "/chat/completions"))
-      .header("Authorization", "Bearer ${state.apiKey}")
+      .header("Authorization", "Bearer ${CodeSetuSettingsState.getInstance().getApiKey()}")
       .header("Content-Type", "application/json")
       .POST(HttpRequest.BodyPublishers.ofString(body))
       .build()
@@ -54,13 +55,13 @@ class CodeSetuProviderClient(
       messages = messages,
       maxTokens = maxTokens,
       temperature = temperature,
-      reasoningEffort = "low",
+      reasoningEffort = reasoningEffortFor(state.provider),
       stream = true,
       json = json,
     )
     val request = HttpRequest.newBuilder()
       .uri(URI.create(state.baseUrl.trimEnd('/') + "/chat/completions"))
-      .header("Authorization", "Bearer ${state.apiKey}")
+      .header("Authorization", "Bearer ${CodeSetuSettingsState.getInstance().getApiKey()}")
       .header("Content-Type", "application/json")
       .POST(HttpRequest.BodyPublishers.ofString(body))
       .build()
@@ -87,7 +88,12 @@ class CodeSetuProviderClient(
           break
         }
 
-        val text = getAssistantChunkText(json.decodeFromString<ChatCompletionChunk>(data))
+        // Skip malformed/partial SSE payloads instead of aborting the whole stream.
+        val text = try {
+          getAssistantChunkText(json.decodeFromString<ChatCompletionChunk>(data))
+        } catch (error: Exception) {
+          ""
+        }
 
         if (text.isNotEmpty()) {
           assistantText.append(text)
@@ -100,6 +106,11 @@ class CodeSetuProviderClient(
   }
 }
 
+// Sarvam needs a low reasoning effort to avoid exhausting its token budget;
+// other providers (OpenAI-compatible, Hugging Face) may reject an unknown field.
+private fun reasoningEffortFor(providerId: String): String? =
+  if (ProviderKind.fromId(providerId) == ProviderKind.SARVAM) "low" else null
+
 fun getAssistantText(response: ChatCompletionResponse): String {
   val message = response.choices.firstOrNull()?.message
   return message?.content ?: message?.refusal.orEmpty()

apps/jetbrains/src/main/kotlin/ai/codesetu/settings/CodeSetuApiKeyStore.kt

@@ -0,0 +1,26 @@
+package ai.codesetu.settings
+
+import com.intellij.credentialStore.CredentialAttributes
+import com.intellij.credentialStore.Credentials
+import com.intellij.credentialStore.generateServiceName
+import com.intellij.ide.passwordSafe.PasswordSafe
+
+/**
+ * Stores the CodeSetu provider API key in the IDE's [PasswordSafe] (OS keychain
+ * / encrypted credential store) instead of the plaintext settings XML.
+ */
+object CodeSetuApiKeyStore {
+  private val attributes: CredentialAttributes =
+    CredentialAttributes(generateServiceName("CodeSetu", "apiKey"))
+
+  fun get(): String = PasswordSafe.instance.getPassword(attributes).orEmpty()
+
+  fun set(value: String) {
+    val trimmed = value.trim()
+    if (trimmed.isEmpty()) {
+      PasswordSafe.instance.set(attributes, null)
+    } else {
+      PasswordSafe.instance.set(attributes, Credentials(null, trimmed))
+    }
+  }
+}

apps/jetbrains/src/main/kotlin/ai/codesetu/settings/CodeSetuModelCatalog.kt

@@ -0,0 +1,22 @@
+package ai.codesetu.settings
+
+import ai.codesetu.model.ProviderKind
+
+/**
+ * Curated, reliably-served chat models offered in the model picker. Users can
+ * still type any other model id (Hub repo id, dedicated endpoint model, etc.).
+ */
+object CodeSetuModelCatalog {
+  val HUGGINGFACE_MODELS: List<String> = listOf(
+    "meta-llama/Llama-3.3-70B-Instruct",
+    "Qwen/Qwen2.5-72B-Instruct",
+    "Qwen/Qwen2.5-Coder-32B-Instruct",
+    "deepseek-ai/DeepSeek-V3-0324",
+    "meta-llama/Llama-3.1-8B-Instruct",
+    "google/gemma-2-27b-it",
+    "mistralai/Mistral-Small-24B-Instruct-2501",
+  )
+
+  fun suggestionsFor(providerId: String): List<String> =
+    if (ProviderKind.fromId(providerId) == ProviderKind.HUGGING_FACE) HUGGINGFACE_MODELS else emptyList()
+}

apps/jetbrains/src/main/kotlin/ai/codesetu/settings/CodeSetuProviderDefaults.kt

@@ -1,7 +1,20 @@
 package ai.codesetu.settings
 
+import ai.codesetu.model.ProviderKind
+
 const val DEFAULT_CODESETU_BASE_URL = "https://api.sarvam.ai/v1"
 const val DEFAULT_CODESETU_MODEL = "sarvam-30b"
 
 fun resolveCodeSetuModel(model: String): String =
   model.ifBlank { DEFAULT_CODESETU_MODEL }
+
+data class ProviderDefaults(val baseUrl: String, val model: String)
+
+fun providerDefaults(providerId: String): ProviderDefaults =
+  when (ProviderKind.fromId(providerId)) {
+    ProviderKind.SARVAM -> ProviderDefaults("https://api.sarvam.ai/v1", "sarvam-30b")
+    ProviderKind.OPENAI_COMPATIBLE ->
+      ProviderDefaults("http://localhost:11434/v1", "qwen2.5-coder:7b")
+    ProviderKind.HUGGING_FACE ->
+      ProviderDefaults("https://router.huggingface.co/v1", "meta-llama/Llama-3.3-70B-Instruct")
+  }