diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
new file mode 100644
index 0000000..ca5ac99
--- /dev/null
+++ b/.github/workflows/ci.yml
@@ -0,0 +1,27 @@
+name: CI
+
+on:
+  push:
+    branches: ['**']
+  pull_request:
+
+jobs:
+  gemini-compat-tests:
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: server
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Setup Node.js
+        uses: actions/setup-node@v4
+        with:
+          node-version: '20'
+
+      - name: Install dependencies
+        run: npm install
+
+      - name: Run Gemini compatibility tests
+        run: npm run test:gemini-compat
diff --git a/README.md b/README.md
index b92a74b..b07557c 100644
--- a/README.md
+++ b/README.md
@@ -49,3 +49,10 @@ yarn gen:types:watch
 ## License
 
 [MIT](LICENSE)
+
+## Gemini CLI -> ChatGPT-style API compatibility
+
+A helper was added at `server/src/modules/common/geminiChatCompat.ts` to bridge Gemini-style request/response shapes to OpenAI Chat Completions-compatible payloads.
+
+- `toGeminiContents(messages)` converts ChatGPT-style messages (`system`/`user`/`assistant`) into Gemini `contents`.
+- `toOpenAIChatCompletion(geminiResponse, model, id?)` converts Gemini response candidates and usage metadata into a Chat Completions response object.
diff --git a/server/package.json b/server/package.json
index 3071370..ebc101b 100644
--- a/server/package.json
+++ b/server/package.json
@@ -6,7 +6,8 @@
   "scripts": {
     "start": "ts-node src/index.ts",
     "watch": "nodemon -e ts -w ./src -x npm run watch:serve",
-    "watch:serve": "ts-node src/index.ts"
+    "watch:serve": "ts-node src/index.ts",
+    "test:gemini-compat": "ts-node src/modules/common/geminiChatCompat.test.ts"
   },
   "devDependencies": {
     "@babel/preset-env": "^7.4.5",
diff --git a/server/src/modules/common/geminiChatCompat.test.ts b/server/src/modules/common/geminiChatCompat.test.ts
new file mode 100644
index 0000000..97fc54c
--- /dev/null
+++ b/server/src/modules/common/geminiChatCompat.test.ts
@@ -0,0 +1,60 @@
+import { strict as assert } from 'assert'
+import {
+  toGeminiContents,
+  toOpenAIChatCompletion,
+  OpenAIChatMessage,
+  GeminiGenerateContentResponse,
+} from './geminiChatCompat'
+
+const messages: OpenAIChatMessage[] = [
+  { role: 'system', content: 'You are concise.' },
+  { role: 'user', content: 'Say hello' },
+  { role: 'assistant', content: 'Hello' },
+]
+
+const geminiContents = toGeminiContents(messages)
+assert.equal(geminiContents.length, 3)
+assert.equal(geminiContents[0].role, 'user')
+assert.ok(
+  geminiContents[0].parts[0].text &&
+    geminiContents[0].parts[0].text.indexOf('System instructions:') === 0
+)
+assert.equal(geminiContents[1].role, 'user')
+assert.equal(geminiContents[1].parts[0].text, 'Say hello')
+assert.equal(geminiContents[2].role, 'model')
+assert.equal(geminiContents[2].parts[0].text, 'Hello')
+
+const geminiResponse: GeminiGenerateContentResponse = {
+  candidates: [
+    {
+      finishReason: 'MAX_TOKENS',
+      content: {
+        role: 'model',
+        parts: [{ text: 'Hi ' }, { text: 'there' }],
+      },
+    },
+  ],
+  usageMetadata: {
+    promptTokenCount: 3,
+    candidatesTokenCount: 2,
+    totalTokenCount: 5,
+  },
+}
+
+const chatCompletion = toOpenAIChatCompletion(geminiResponse, 'gpt-compat')
+assert.equal(chatCompletion.object, 'chat.completion')
+assert.equal(chatCompletion.model, 'gpt-compat')
+assert.equal(chatCompletion.choices[0].message.role, 'assistant')
+assert.equal(chatCompletion.choices[0].message.content, 'Hi there')
+assert.equal(chatCompletion.choices[0].finish_reason, 'length')
+assert.equal(chatCompletion.usage.prompt_tokens, 3)
+assert.equal(chatCompletion.usage.completion_tokens, 2)
+assert.equal(chatCompletion.usage.total_tokens, 5)
+
+const emptyResponse = toOpenAIChatCompletion({}, 'gpt-compat', 'chatcmpl-fixed')
+assert.equal(emptyResponse.id, 'chatcmpl-fixed')
+assert.equal(emptyResponse.choices[0].message.content, '')
+assert.equal(emptyResponse.choices[0].finish_reason, 'stop')
+assert.equal(emptyResponse.usage.total_tokens, 0)
+
+console.log('geminiChatCompat tests passed')
diff --git a/server/src/modules/common/geminiChatCompat.ts b/server/src/modules/common/geminiChatCompat.ts
new file mode 100644
index 0000000..1757dda
--- /dev/null
+++ b/server/src/modules/common/geminiChatCompat.ts
@@ -0,0 +1,115 @@
+export type OpenAIChatRole = 'system' | 'user' | 'assistant' | 'tool'
+
+export interface OpenAIChatMessage {
+  role: OpenAIChatRole
+  content: string
+}
+
+export interface GeminiContentPart {
+  text?: string
+}
+
+export interface GeminiContent {
+  role?: 'user' | 'model'
+  parts: GeminiContentPart[]
+}
+
+export interface GeminiGenerateContentResponse {
+  candidates?: Array<{
+    content?: GeminiContent
+    finishReason?: string
+  }>
+  usageMetadata?: {
+    promptTokenCount?: number
+    candidatesTokenCount?: number
+    totalTokenCount?: number
+  }
+  modelVersion?: string
+}
+
+export const toGeminiContents = (messages: OpenAIChatMessage[]): GeminiContent[] => {
+  const systemMessages = messages.filter(message => message.role === 'system')
+  const nonSystem = messages.filter(message => message.role !== 'system')
+
+  const normalized: GeminiContent[] = nonSystem.map(message => ({
+    role: (message.role === 'assistant' ? 'model' : 'user') as 'user' | 'model',
+    parts: [{ text: message.content }],
+  }))
+
+  if (systemMessages.length === 0) {
+    return normalized
+  }
+
+  return [
+    {
+      role: 'user',
+      parts: [
+        {
+          text: `System instructions:\n${systemMessages
+            .map(message => message.content)
+            .join('\n\n')}`,
+        },
+      ],
+    },
+    ...normalized,
+  ]
+}
+
+export const toOpenAIChatCompletion = (
+  geminiResponse: GeminiGenerateContentResponse,
+  model: string,
+  id: string = `chatcmpl-${Date.now()}`
+) => {
+  const firstCandidate = geminiResponse.candidates && geminiResponse.candidates[0]
+  const parts =
+    firstCandidate && firstCandidate.content && firstCandidate.content.parts
+      ? firstCandidate.content.parts
+      : []
+
+  const textContent = parts
+    .map(part => part.text || '')
+    .join('')
+    .trim()
+
+  return {
+    id,
+    object: 'chat.completion',
+    created: Math.floor(Date.now() / 1000),
+    model,
+    choices: [
+      {
+        index: 0,
+        message: {
+          role: 'assistant',
+          content: textContent,
+        },
+        finish_reason: normalizeFinishReason(firstCandidate && firstCandidate.finishReason),
+      },
+    ],
+    usage: {
+      prompt_tokens:
+        (geminiResponse.usageMetadata && geminiResponse.usageMetadata.promptTokenCount) || 0,
+      completion_tokens:
+        (geminiResponse.usageMetadata && geminiResponse.usageMetadata.candidatesTokenCount) || 0,
+      total_tokens:
+        (geminiResponse.usageMetadata && geminiResponse.usageMetadata.totalTokenCount) || 0,
+    },
+  }
+}
+
+const normalizeFinishReason = (finishReason?: string): string => {
+  if (!finishReason) {
+    return 'stop'
+  }
+
+  switch (finishReason) {
+    case 'STOP':
+      return 'stop'
+    case 'MAX_TOKENS':
+      return 'length'
+    case 'SAFETY':
+      return 'content_filter'
+    default:
+      return 'stop'
+  }
+}