feat(llm): add Atlas Cloud provider

.env.example

@@ -20,7 +20,7 @@ db_mode=json
 ########################################
 # LLM Core Settings
 ########################################
-LLM_PROVIDER=gemini            # gemini | openai | claude | grok | ollama | openrouter | requesty | cheaperinference | minimax | apiroute
+LLM_PROVIDER=gemini            # gemini | openai | claude | grok | ollama | openrouter | requesty | cheaperinference | atlascloud | minimax | apiroute
 EMB_PROVIDER=openai             # openai | gemini | ollama etc.
 LLM_TEMP=1
 LLM_MAXTOK=16384
@@ -74,6 +74,16 @@ CHEAPER_INFERENCE_API_KEY=
 CHEAPER_INFERENCE_MODEL=gpt-5.4-mini
 CHEAPER_INFERENCE_BASE_URL="https://api.cheaperinference.com/v1"
 
+########################################
+# Atlas Cloud (https://atlascloud.ai)
+########################################
+# Get a key at https://atlascloud.ai -- the model catalog at
+# https://api.atlascloud.ai/v1/models is public and needs no key.
+# Embeddings use the OpenAI settings above
+ATLASCLOUD_API_KEY=
+ATLASCLOUD_MODEL=deepseek-ai/deepseek-v4-flash
+ATLASCLOUD_BASE_URL="https://api.atlascloud.ai/v1"
+
 ########################################
 # Grok (xAI)
 ########################################

README.md

@@ -79,7 +79,7 @@ The platform provides a modern interface for students, educators, and researcher
 
 ### Supported AI Models
 
-- Google Gemini • OpenAI GPT • Anthropic Claude • xAI Grok • [MiniMax](https://www.minimax.io/) • [API Route](https://www.api-route.com) • Ollama (local) • OpenRouter • [Requesty](https://requesty.ai) • [Cheaper Inference](https://cheaperinference.com)
+- Google Gemini • OpenAI GPT • Anthropic Claude • xAI Grok • [MiniMax](https://www.minimax.io/) • [API Route](https://www.api-route.com) • Ollama (local) • OpenRouter • [Requesty](https://requesty.ai) • [Cheaper Inference](https://cheaperinference.com) • [Atlas Cloud](https://atlascloud.ai)
 
 ### Embedding Providers
 

backend/src/config/env.ts

@@ -17,6 +17,9 @@ export const config = {
   cheaperinference: process.env.CHEAPER_INFERENCE_API_KEY || '',
   cheaperinference_model: process.env.CHEAPER_INFERENCE_MODEL || 'gpt-5.4-mini',
   cheaperinference_base: process.env.CHEAPER_INFERENCE_BASE_URL || 'https://api.cheaperinference.com/v1',
+  atlascloud: process.env.ATLASCLOUD_API_KEY || '',
+  atlascloud_model: process.env.ATLASCLOUD_MODEL || 'deepseek-ai/deepseek-v4-flash',
+  atlascloud_base: process.env.ATLASCLOUD_BASE_URL || 'https://api.atlascloud.ai/v1',
   gemini: process.env.gemini || process.env.GOOGLE_API_KEY || '',
   gemini_model: process.env.gemini_model || 'gemini-1.5-pro',
   gemini_embed_model: process.env.gemini_embed_model || 'text-embedding-004',

backend/src/utils/llm/models/__tests__/atlascloud.test.ts

@@ -0,0 +1,71 @@
+import { describe, it, expect, vi, beforeEach } from 'vitest'
+
+vi.mock('@langchain/openai', () => {
+  const MockChatOpenAI = vi.fn(function (this: any, opts: any) {
+    this.invoke = vi.fn().mockResolvedValue({ content: 'mock response' })
+    this._opts = opts
+  })
+  const MockOpenAIEmbeddings = vi.fn(function (this: any, opts: any) {
+    this.embedDocuments = vi.fn().mockResolvedValue([[0.1, 0.2]])
+    this.embedQuery = vi.fn().mockResolvedValue([0.1, 0.2])
+    this._opts = opts
+  })
+  return { ChatOpenAI: MockChatOpenAI, OpenAIEmbeddings: MockOpenAIEmbeddings }
+})
+
+import { ChatOpenAI, OpenAIEmbeddings } from '@langchain/openai'
+import { makeLLM, makeEmbeddings } from '../atlascloud'
+
+describe('Atlas Cloud LLM provider', () => {
+  beforeEach(() => {
+    vi.clearAllMocks()
+  })
+
+  it('should create a ChatOpenAI instance with Atlas Cloud defaults', () => {
+    const llm = makeLLM({ atlascloud: 'test-key' })
+
+    expect(ChatOpenAI).toHaveBeenCalledWith(
+      expect.objectContaining({
+        model: 'deepseek-ai/deepseek-v4-flash',
+        apiKey: 'test-key',
+        configuration: { baseURL: 'https://api.atlascloud.ai/v1' },
+      }),
+    )
+    expect(typeof llm.invoke).toBe('function')
+    expect(typeof llm.call).toBe('function')
+  })
+
+  it('should use configured model and base URL', () => {
+    makeLLM({
+      atlascloud: 'key',
+      atlascloud_model: 'zai-org/glm-5.3-flash',
+      atlascloud_base: 'https://gateway.example.test/v1',
+    })
+
+    expect(ChatOpenAI).toHaveBeenCalledWith(
+      expect.objectContaining({
+        model: 'zai-org/glm-5.3-flash',
+        configuration: { baseURL: 'https://gateway.example.test/v1' },
+      }),
+    )
+  })
+
+  it('should pass temperature and max tokens through', () => {
+    makeLLM({ atlascloud: 'key', temp: 0.2, max_tokens: 1024 })
+
+    expect(ChatOpenAI).toHaveBeenCalledWith(
+      expect.objectContaining({ temperature: 0.2, maxTokens: 1024 }),
+    )
+  })
+
+  it('should create embeddings with the OpenAI settings', () => {
+    makeEmbeddings({ openai: 'openai-key' })
+
+    expect(OpenAIEmbeddings).toHaveBeenCalledWith(
+      expect.objectContaining({
+        model: 'text-embedding-3-large',
+        apiKey: 'openai-key',
+      }),
+    )
+  })
+})

backend/src/utils/llm/models/atlascloud.ts

@@ -0,0 +1,24 @@
+import { ChatOpenAI, OpenAIEmbeddings } from '@langchain/openai'
+import { wrapChat } from './util'
+import type { MkLLM, MkEmb, EmbeddingsLike } from './types'
+
+export const makeLLM: MkLLM = (cfg: any) => {
+  const m = new ChatOpenAI({
+    model: cfg.atlascloud_model || 'deepseek-ai/deepseek-v4-flash',
+    apiKey: cfg.atlascloud || '',
+    configuration: { baseURL: cfg.atlascloud_base || 'https://api.atlascloud.ai/v1' },
+    temperature: cfg.temp ?? 0.7,
+    maxTokens: cfg.max_tokens,
+  })
+  return wrapChat(m)
+}
+
+// Atlas Cloud serves chat completions only -- it has no /v1/embeddings
+// endpoint -- so embeddings fall back to the OpenAI settings, the same way
+// the other chat-only gateways here do.
+export const makeEmbeddings: MkEmb = (cfg: any): EmbeddingsLike => {
+  return new OpenAIEmbeddings({
+    model: cfg.openai_embed_model || 'text-embedding-3-large',
+    apiKey: cfg.openai || process.env.OPENAI_API_KEY,
+  })
+}

backend/src/utils/llm/models/index.ts

@@ -6,6 +6,7 @@ import * as claude from './claude'
 import * as openrouter from './openrouter'
 import * as requesty from './requesty'
 import * as cheaperinference from './cheaperinference'
+import * as atlascloud from './atlascloud'
 import * as minimax from './minimax'
 import * as apiroute from './apiroute'
 import { config } from '../../../config/env'
@@ -23,6 +24,7 @@ function pick(p: string) {
     case 'openrouter': return openrouter
     case 'requesty': return requesty
     case 'cheaperinference': return cheaperinference
+    case 'atlascloud': return atlascloud
     case 'minimax': return minimax
     case 'apiroute':
     case 'api_route':