@openenthrium/oe-runtime-sdk 1.7.5

package/src/providers/llm/index.js

@@ -0,0 +1,136 @@
+const { PrismaClient } = require("@prisma/client");
+const db = new PrismaClient();
+
+async function getSetting(key) {
+  const s = await db.setting.findUnique({ where: { key } });
+  return s?.value || null;
+}
+
+// Providers that use the OpenAI SDK with a custom baseURL
+const OPENAI_COMPATIBLE = {
+  groq:        { baseURL: "https://api.groq.com/openai/v1" },
+  togetherai:  { baseURL: "https://api.together.xyz/v1" },
+  mistral:     { baseURL: "https://api.mistral.ai/v1" },
+  openrouter:  { baseURL: "https://openrouter.ai/api/v1" },
+  perplexity:  { baseURL: "https://api.perplexity.ai" },
+  deepseek:    { baseURL: "https://api.deepseek.com/v1" },
+  xai:         { baseURL: "https://api.x.ai/v1" },
+  fireworks:   { baseURL: "https://api.fireworks.ai/inference/v1" },
+  // Google Gemini exposes an OpenAI-compatible endpoint
+  gemini:      { baseURL: "https://generativelanguage.googleapis.com/v1beta/openai/" },
+  nvidiaNim:   { baseURL: "https://integrate.api.nvidia.com/v1" },
+  sambanova:   { baseURL: "https://api.sambanova.ai/v1" },
+  litellm:     { baseURL: null, apiKey: "no-key", localBaseUrlKey: "llm_base_url", localDefault: "http://localhost:4000" },
+  // Local providers
+  ollama:      { baseURL: null, apiKey: "ollama", localBaseUrlKey: "llm_base_url", localDefault: "http://localhost:11434/v1" },
+  lmstudio:    { baseURL: null, apiKey: "lmstudio", localBaseUrlKey: "llm_base_url", localDefault: "http://localhost:1234/v1" },
+  "generic-openai": { baseURL: null, localBaseUrlKey: "llm_base_url", localDefault: "" },
+};
+
+async function getLLMClient() {
+  const { default: OpenAI } = await import("openai");
+  const provider = (await getSetting("llm_provider")) || process.env.LLM_PROVIDER || "openai";
+  const apiKey = (await getSetting("llm_api_key")) || null;
+
+  // Azure (special case — needs endpoint + deployment in URL)
+  if (provider === "azure") {
+    const endpoint = (await getSetting("llm_azure_endpoint")) || process.env.AZURE_OPENAI_ENDPOINT;
+    const deployment = (await getSetting("llm_azure_deployment")) || process.env.AZURE_OPENAI_DEPLOYMENT;
+    const azureKey = (await getSetting("llm_api_key")) || process.env.AZURE_OPENAI_API_KEY;
+    return {
+      provider: "azure",
+      client: new OpenAI({
+        apiKey: azureKey,
+        baseURL: `${endpoint}/openai/deployments/${deployment}`,
+        defaultQuery: { "api-version": "2024-08-01-preview" },
+        defaultHeaders: { "api-key": azureKey }
+      })
+    };
+  }
+
+  // Anthropic (separate SDK)
+  if (provider === "anthropic") {
+    const Anthropic = require("@anthropic-ai/sdk");
+    const anthropicKey = apiKey || process.env.ANTHROPIC_API_KEY;
+    return { provider: "anthropic", client: new Anthropic({ apiKey: anthropicKey }) };
+  }
+
+  // Standard OpenAI
+  if (provider === "openai") {
+    return { provider: "openai", client: new OpenAI({ apiKey: apiKey || process.env.OPENAI_API_KEY }) };
+  }
+
+  // All OpenAI-compatible providers
+  const compat = OPENAI_COMPATIBLE[provider];
+  if (compat) {
+    let baseURL = compat.baseURL;
+    if (!baseURL) {
+      baseURL = (await getSetting(compat.localBaseUrlKey)) || compat.localDefault;
+    }
+    const key = apiKey || compat.apiKey || "no-key";
+    return { provider, client: new OpenAI({ apiKey: key, baseURL }) };
+  }
+
+  throw new Error(`Unknown LLM provider: ${provider}`);
+}
+
+async function chat(messages, { stream = false, systemPrompt = null, temperature = 0.7 } = {}) {
+  const { provider, client } = await getLLMClient();
+  const model = (await getSetting("llm_model"))
+    || process.env.OPENAI_MODEL
+    || process.env.OLLAMA_MODEL
+    || "gpt-4o";
+
+  const allMessages = systemPrompt
+    ? [{ role: "system", content: systemPrompt }, ...messages]
+    : messages;
+
+  if (provider === "anthropic") {
+    const system = allMessages.find(m => m.role === "system")?.content;
+    const userMsgs = allMessages.filter(m => m.role !== "system");
+    const response = await client.messages.create({
+      model: model || "claude-3-5-sonnet-20241022",
+      max_tokens: 4096,
+      system: system || undefined,
+      messages: userMsgs,
+      temperature,
+      stream
+    });
+    if (stream) return response;
+    return response.content[0].text;
+  }
+
+  const response = await client.chat.completions.create({ model, messages: allMessages, temperature, stream });
+  if (stream) return response;
+  return response.choices[0].message.content;
+}
+
+/**
+ * Read LLM settings from DB and return a plain config object
+ * that can be passed directly to engine.run() or createLLMClient().
+ */
+async function getLLMConfig() {
+  const provider = (await getSetting("llm_provider")) || process.env.LLM_PROVIDER || "openai";
+  const apiKey   = (await getSetting("llm_api_key"))  || null;
+  const model    = (await getSetting("llm_model"))    || process.env.OPENAI_MODEL || process.env.OLLAMA_MODEL || "gpt-4o";
+
+  if (provider === "azure") {
+    return {
+      provider,
+      apiKey:          apiKey || process.env.AZURE_OPENAI_API_KEY,
+      model,
+      azureEndpoint:   (await getSetting("llm_azure_endpoint"))   || process.env.AZURE_OPENAI_ENDPOINT,
+      azureDeployment: (await getSetting("llm_azure_deployment"))  || process.env.AZURE_OPENAI_DEPLOYMENT,
+    };
+  }
+
+  const compat = OPENAI_COMPATIBLE[provider];
+  if (compat?.localBaseUrlKey) {
+    return { provider, apiKey: apiKey || compat.apiKey || "no-key", model,
+             baseURL: (await getSetting(compat.localBaseUrlKey)) || compat.localDefault };
+  }
+
+  return { provider, apiKey, model };
+}
+
+module.exports = { chat, getLLMClient, getLLMConfig, getSetting };