@ssweens/pi-vertex 1.0.0

package/models/maas.ts

@@ -0,0 +1,462 @@
+/**
+ * MaaS (Model-as-a-Service) open model definitions for Vertex AI
+ * Pricing: https://cloud.google.com/vertex-ai/generative-ai/pricing#open-models
+ * All prices per 1M tokens (as of Feb 2025)
+ */
+
+import type { VertexModelConfig } from "../types.js";
+
+export const MAAS_MODELS: VertexModelConfig[] = [
+  // Llama models (Meta)
+  {
+    id: "llama-4-maverick",
+    name: "Llama 4 Maverick",
+    apiId: "meta/llama-4-maverick-17b-128e-instruct-maas",
+    publisher: "meta",
+    endpointType: "maas",
+    contextWindow: 524288,
+    maxTokens: 32000,
+    input: ["text"],
+    reasoning: true,
+    tools: true,
+    cost: {
+      input: 0.35,
+      output: 1.15,
+      cacheRead: 0,
+      cacheWrite: 0,
+    },
+    region: "global",
+  },
+  {
+    id: "llama-4-scout",
+    name: "Llama 4 Scout",
+    apiId: "meta/llama-4-scout-17b-16e-instruct-maas",
+    publisher: "meta",
+    endpointType: "maas",
+    contextWindow: 1310720,
+    maxTokens: 32000,
+    input: ["text"],
+    reasoning: true,
+    tools: true,
+    cost: {
+      input: 0.25,
+      output: 0.70,
+      cacheRead: 0,
+      cacheWrite: 0,
+    },
+    region: "global",
+  },
+  {
+    id: "llama-3.3-70b",
+    name: "Llama 3.3 70B",
+    apiId: "meta/llama-3.3-70b-instruct-maas",
+    publisher: "meta",
+    endpointType: "maas",
+    contextWindow: 128000,
+    maxTokens: 8192,
+    input: ["text"],
+    reasoning: false,
+    tools: true,
+    cost: {
+      input: 0.72,
+      output: 0.72,
+      cacheRead: 0,
+      cacheWrite: 0,
+    },
+    region: "global",
+  },
+
+  // Mistral models
+  {
+    id: "mistral-medium-3",
+    name: "Mistral Medium 3",
+    apiId: "mistralai/mistral-medium-3",
+    publisher: "mistralai",
+    endpointType: "maas",
+    contextWindow: 128000,
+    maxTokens: 32000,
+    input: ["text"],
+    reasoning: false,
+    tools: true,
+    cost: {
+      input: 0.40,
+      output: 2.00,
+      cacheRead: 0,
+      cacheWrite: 0,
+    },
+    region: "global",
+  },
+  {
+    id: "mistral-small-3.1",
+    name: "Mistral Small 3.1",
+    apiId: "mistralai/mistral-small-2503",
+    publisher: "mistralai",
+    endpointType: "maas",
+    contextWindow: 128000,
+    maxTokens: 32000,
+    input: ["text"],
+    reasoning: false,
+    tools: true,
+    cost: {
+      input: 0.10,
+      output: 0.30,
+      cacheRead: 0,
+      cacheWrite: 0,
+    },
+    region: "global",
+  },
+  {
+    id: "mistral-ocr",
+    name: "Mistral OCR",
+    apiId: "mistralai/mistral-ocr-2505",
+    publisher: "mistralai",
+    endpointType: "maas",
+    contextWindow: 128000,
+    maxTokens: 32000,
+    input: ["text", "image"],
+    reasoning: false,
+    tools: false,
+    cost: {
+      input: 0.50,  // Per page: $0.0005/page, shown as approx per 1K pages
+      output: 0.50,  // Per page pricing
+      cacheRead: 0,
+      cacheWrite: 0,
+    },
+    region: "global",
+  },
+  {
+    id: "codestral-2",
+    name: "Codestral 2",
+    apiId: "mistralai/codestral-2",
+    publisher: "mistralai",
+    endpointType: "maas",
+    contextWindow: 256000,
+    maxTokens: 32000,
+    input: ["text"],
+    reasoning: false,
+    tools: true,
+    cost: {
+      input: 0.30,
+      output: 0.90,
+      cacheRead: 0,
+      cacheWrite: 0,
+    },
+    region: "global",
+  },
+
+  // DeepSeek models
+  {
+    id: "deepseek-v3.2",
+    name: "DeepSeek V3.2",
+    apiId: "deepseek-ai/deepseek-v3.2-maas",
+    publisher: "deepseek-ai",
+    endpointType: "maas",
+    contextWindow: 163840,
+    maxTokens: 32000,
+    input: ["text"],
+    reasoning: true,
+    tools: true,
+    cost: {
+      input: 0.56,
+      output: 1.68,
+      cacheRead: 0.056,
+      cacheWrite: 0,
+    },
+    region: "global",
+  },
+  {
+    id: "deepseek-v3.1",
+    name: "DeepSeek V3.1",
+    apiId: "deepseek-ai/deepseek-v3.1-maas",
+    publisher: "deepseek-ai",
+    endpointType: "maas",
+    contextWindow: 163840,
+    maxTokens: 32000,
+    input: ["text"],
+    reasoning: true,
+    tools: true,
+    cost: {
+      input: 0.60,
+      output: 1.70,
+      cacheRead: 0.06,
+      cacheWrite: 0,
+    },
+    region: "global",
+  },
+  {
+    id: "deepseek-r1",
+    name: "DeepSeek R1",
+    apiId: "deepseek-ai/deepseek-r1-0528-maas",
+    publisher: "deepseek-ai",
+    endpointType: "maas",
+    contextWindow: 163840,
+    maxTokens: 32000,
+    input: ["text"],
+    reasoning: true,
+    tools: true,
+    cost: {
+      input: 1.35,
+      output: 5.40,
+      cacheRead: 0,
+      cacheWrite: 0,
+    },
+    region: "global",
+  },
+
+  // AI21 Labs models
+  {
+    id: "jamba-1.5-large",
+    name: "Jamba 1.5 Large",
+    apiId: "ai21/jamba-1.5-large",
+    publisher: "ai21",
+    endpointType: "maas",
+    contextWindow: 256000,
+    maxTokens: 256000,
+    input: ["text"],
+    reasoning: false,
+    tools: true,
+    cost: {
+      input: 2.00,
+      output: 8.00,
+      cacheRead: 0,
+      cacheWrite: 0,
+    },
+    region: "global",
+  },
+  {
+    id: "jamba-1.5-mini",
+    name: "Jamba 1.5 Mini",
+    apiId: "ai21/jamba-1.5-mini",
+    publisher: "ai21",
+    endpointType: "maas",
+    contextWindow: 256000,
+    maxTokens: 256000,
+    input: ["text"],
+    reasoning: false,
+    tools: true,
+    cost: {
+      input: 0.20,
+      output: 0.40,
+      cacheRead: 0,
+      cacheWrite: 0,
+    },
+    region: "global",
+  },
+
+  // OpenAI models (gpt-oss)
+  {
+    id: "gpt-oss-120b",
+    name: "GPT-OSS 120B",
+    apiId: "openai/gpt-oss-120b-maas",
+    publisher: "openai",
+    endpointType: "maas",
+    contextWindow: 131072,
+    maxTokens: 32000,
+    input: ["text"],
+    reasoning: true,
+    tools: true,
+    cost: {
+      input: 0.09,
+      output: 0.36,
+      cacheRead: 0,
+      cacheWrite: 0,
+    },
+    region: "global",
+  },
+  {
+    id: "gpt-oss-20b",
+    name: "GPT-OSS 20B",
+    apiId: "openai/gpt-oss-20b-maas",
+    publisher: "openai",
+    endpointType: "maas",
+    contextWindow: 131072,
+    maxTokens: 32000,
+    input: ["text"],
+    reasoning: false,
+    tools: true,
+    cost: {
+      input: 0.07,
+      output: 0.25,
+      cacheRead: 0.007,
+      cacheWrite: 0,
+    },
+    region: "global",
+  },
+
+  // DeepSeek OCR
+  {
+    id: "deepseek-ocr",
+    name: "DeepSeek OCR",
+    apiId: "deepseek-ai/deepseek-ocr-maas",
+    publisher: "deepseek-ai",
+    endpointType: "maas",
+    contextWindow: 163840,
+    maxTokens: 32000,
+    input: ["text", "image"],
+    reasoning: false,
+    tools: false,
+    cost: {
+      input: 0.30,  // Per page: $0.0003/page
+      output: 1.20,  // Per page pricing
+      cacheRead: 0,
+      cacheWrite: 0,
+    },
+    region: "global",
+  },
+
+  // Qwen models
+  {
+    id: "qwen3-235b",
+    name: "Qwen 3 235B",
+    apiId: "qwen/qwen3-235b-a22b-instruct-2507-maas",
+    publisher: "qwen",
+    endpointType: "maas",
+    contextWindow: 262144,
+    maxTokens: 32000,
+    input: ["text"],
+    reasoning: true,
+    tools: true,
+    cost: {
+      input: 0.22,
+      output: 0.88,
+      cacheRead: 0,
+      cacheWrite: 0,
+    },
+    region: "global",
+  },
+  {
+    id: "qwen3-next-instruct",
+    name: "Qwen 3 Next Instruct",
+    apiId: "qwen/qwen3-next-instruct-80b-maas",
+    publisher: "qwen",
+    endpointType: "maas",
+    contextWindow: 262144,
+    maxTokens: 32000,
+    input: ["text"],
+    reasoning: true,
+    tools: true,
+    cost: {
+      input: 0.15,
+      output: 1.20,
+      cacheRead: 0,
+      cacheWrite: 0,
+    },
+    region: "global",
+  },
+  {
+    id: "qwen3-next-thinking",
+    name: "Qwen 3 Next Thinking",
+    apiId: "qwen/qwen3-next-thinking-80b-maas",
+    publisher: "qwen",
+    endpointType: "maas",
+    contextWindow: 262144,
+    maxTokens: 32000,
+    input: ["text"],
+    reasoning: true,
+    tools: true,
+    cost: {
+      input: 0.15,
+      output: 1.20,
+      cacheRead: 0,
+      cacheWrite: 0,
+    },
+    region: "global",
+  },
+  {
+    id: "qwen3-coder",
+    name: "Qwen 3 Coder",
+    apiId: "qwen/qwen3-coder-480b-a35b-instruct-maas",
+    publisher: "qwen",
+    endpointType: "maas",
+    contextWindow: 262144,
+    maxTokens: 32000,
+    input: ["text"],
+    reasoning: true,
+    tools: true,
+    cost: {
+      input: 0.22,
+      output: 1.80,
+      cacheRead: 0.022,
+      cacheWrite: 0,
+    },
+    region: "global",
+  },
+
+  // Other models
+  {
+    id: "kimi-k2-thinking",
+    name: "Kimi K2 Thinking",
+    apiId: "moonshotai/kimi-k2-thinking-maas",
+    publisher: "moonshotai",
+    endpointType: "maas",
+    contextWindow: 262144,
+    maxTokens: 32000,
+    input: ["text"],
+    reasoning: true,
+    tools: true,
+    cost: {
+      input: 0.60,
+      output: 2.50,
+      cacheRead: 0.06,
+      cacheWrite: 0,
+    },
+    region: "global",
+  },
+  {
+    id: "minimax-m2",
+    name: "MiniMax M2",
+    apiId: "minimaxai/minimax-m2-maas",
+    publisher: "minimaxai",
+    endpointType: "maas",
+    contextWindow: 196608,
+    maxTokens: 32000,
+    input: ["text"],
+    reasoning: true,
+    tools: true,
+    cost: {
+      input: 0.30,
+      output: 1.20,
+      cacheRead: 0.03,
+      cacheWrite: 0,
+    },
+    region: "global",
+  },
+  {
+    id: "glm-5",
+    name: "GLM 5",
+    apiId: "zai-org/glm-5-maas",
+    publisher: "zai-org",
+    endpointType: "maas",
+    contextWindow: 200000,
+    maxTokens: 32000,
+    input: ["text"],
+    reasoning: true,
+    tools: true,
+    cost: {
+      input: 1.00,
+      output: 3.20,
+      cacheRead: 0.10,
+      cacheWrite: 0,
+    },
+    region: "global",
+  },
+  {
+    id: "glm-4.7",
+    name: "GLM 4.7",
+    apiId: "zai-org/glm-4.7-maas",
+    publisher: "zai-org",
+    endpointType: "maas",
+    contextWindow: 200000,
+    maxTokens: 32000,
+    input: ["text"],
+    reasoning: true,
+    tools: true,
+    cost: {
+      input: 0.60,
+      output: 2.20,
+      cacheRead: 0,
+      cacheWrite: 0,
+    },
+    region: "global",
+  },
+];

package/package.json

@@ -0,0 +1,47 @@
+{
+  "name": "@ssweens/pi-vertex",
+  "version": "1.0.0",
+  "description": "Google Vertex AI provider for Pi coding agent - supports Gemini, Claude, and all MaaS models",
+  "type": "module",
+  "main": "index.ts",
+  "files": [
+    "index.ts",
+    "auth.ts",
+    "config.ts",
+    "types.ts",
+    "utils.ts",
+    "models/",
+    "streaming/",
+    "README.md",
+    "LICENSE",
+    "screenshot.png"
+  ],
+  "scripts": {
+    "clean": "echo 'nothing to clean'",
+    "build": "echo 'nothing to build'",
+    "check": "echo 'nothing to check'"
+  },
+  "dependencies": {
+    "@google/genai": "^1.42.0",
+    "google-auth-library": "^9.0.0"
+  },
+  "peerDependencies": {
+    "@mariozechner/pi-ai": "*",
+    "@mariozechner/pi-coding-agent": "*"
+  },
+  "pi": {
+    "extensions": [
+      "./index.ts"
+    ]
+  },
+  "keywords": [
+    "pi-package",
+    "pi-extension",
+    "vertex-ai",
+    "gemini",
+    "claude",
+    "maas"
+  ],
+  "author": "ssweens",
+  "license": "MIT"
+}

package/streaming/gemini.ts

@@ -0,0 +1,164 @@
+/**
+ * Gemini streaming handler using @google/genai SDK
+ */
+
+import { GoogleGenAI } from "@google/genai";
+import type { VertexModelConfig, Context, StreamOptions } from "../types.js";
+import { getAuthConfig, resolveLocation } from "../auth.js";
+import { sanitizeText, convertToGeminiMessages, calculateCost } from "../utils.js";
+import { createAssistantMessageEventStream, type AssistantMessageEventStream, type AssistantMessage } from "@mariozechner/pi-ai";
+
+export function streamGemini(
+  model: VertexModelConfig,
+  context: Context,
+  options?: StreamOptions
+): AssistantMessageEventStream {
+  const stream = createAssistantMessageEventStream();
+  
+  (async () => {
+    const output: AssistantMessage = {
+      role: "assistant",
+      content: [],
+      api: "google-generative-ai",
+      provider: "vertex",
+      model: model.id,
+      usage: {
+        input: 0,
+        output: 0,
+        cacheRead: 0,
+        cacheWrite: 0,
+        totalTokens: 0,
+        cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
+      },
+      stopReason: "stop",
+      timestamp: Date.now(),
+    };
+    
+    try {
+      // Priority: config file > env var > model region > default
+      const location = resolveLocation(model.region);
+      const auth = getAuthConfig(location);
+
+      // Create client
+      const client = new GoogleGenAI({
+        vertexai: true,
+        project: auth.projectId,
+        location: auth.location,
+      });
+      
+      // Convert messages
+      const contents = convertToGeminiMessages(context.messages);
+      
+      // Build config
+      const config: any = {
+        maxOutputTokens: options?.maxTokens || Math.floor(model.maxTokens / 2),
+        temperature: options?.temperature ?? 0.7,
+      };
+      
+      // Add system prompt if present
+      if (context.systemPrompt) {
+        config.systemInstruction = sanitizeText(context.systemPrompt);
+      }
+      
+      // Add tools if present
+      if (context.tools && context.tools.length > 0) {
+        config.tools = [
+          {
+            functionDeclarations: context.tools.map((tool) => ({
+              name: tool.name,
+              description: tool.description,
+              parameters: tool.parameters,
+            })),
+          },
+        ];
+      }
+      
+      stream.push({ type: "start", partial: output });
+      
+      // Start streaming
+      const response = await client.models.generateContentStream({
+        model: model.apiId,
+        contents,
+        config,
+      });
+      
+      let textContent = "";
+      let textIndex = 0;
+      
+      for await (const chunk of response) {
+        if (options?.signal?.aborted) {
+          throw new Error("Request was aborted");
+        }
+        
+        // Update usage
+        if (chunk.usageMetadata) {
+          output.usage.input = chunk.usageMetadata.promptTokenCount || output.usage.input;
+          output.usage.output = chunk.usageMetadata.candidatesTokenCount || output.usage.output;
+          output.usage.totalTokens = chunk.usageMetadata.totalTokenCount || 
+            (output.usage.input + output.usage.output);
+          calculateCost(model.cost.input, model.cost.output, model.cost.cacheRead, model.cost.cacheWrite, output.usage);
+        }
+        
+        // Handle text
+        const text = chunk.text;
+        if (text) {
+          if (!textContent) {
+            // First text chunk
+            output.content.push({ type: "text", text: "" });
+            textIndex = output.content.length - 1;
+            stream.push({ type: "text_start", contentIndex: textIndex, partial: output });
+          }
+          textContent += text;
+          (output.content[textIndex] as any).text = textContent;
+          stream.push({ type: "text_delta", contentIndex: textIndex, delta: text, partial: output });
+        }
+        
+        // Handle function calls (tools)
+        if (chunk.functionCalls && chunk.functionCalls.length > 0) {
+          for (const call of chunk.functionCalls) {
+            output.content.push({
+              type: "toolCall",
+              id: call.id || `call_${Date.now()}`,
+              name: call.name,
+              arguments: call.args || {},
+            });
+            stream.push({
+              type: "toolcall_end",
+              contentIndex: output.content.length - 1,
+              toolCall: output.content[output.content.length - 1] as any,
+              partial: output,
+            });
+          }
+        }
+        
+        // Handle finish reason
+        if (chunk.candidates && chunk.candidates[0]?.finishReason) {
+          const reason = chunk.candidates[0].finishReason;
+          if (reason === "STOP") {
+            output.stopReason = "stop";
+          } else if (reason === "MAX_TOKENS") {
+            output.stopReason = "length";
+          } else if (reason === "SAFETY") {
+            output.stopReason = "error";
+            output.errorMessage = "Content blocked by safety filters";
+          }
+        }
+      }
+      
+      // End text if we had any
+      if (textContent) {
+        stream.push({ type: "text_end", contentIndex: textIndex, content: textContent, partial: output });
+      }
+      
+      stream.push({ type: "done", reason: output.stopReason as any, message: output });
+      stream.end();
+    } catch (error) {
+      output.stopReason = options?.signal?.aborted ? "aborted" : "error";
+      output.errorMessage = error instanceof Error ? error.message : String(error);
+      stream.push({ type: "error", reason: output.stopReason, error: output });
+      stream.end();
+    }
+  })();
+  
+  return stream;
+}

package/streaming/index.ts

@@ -0,0 +1,25 @@
+/**
+ * Streaming handler dispatcher
+ */
+
+import type { VertexModelConfig, Context, StreamOptions } from "../types.js";
+import type { AssistantMessageEventStream } from "@mariozechner/pi-ai";
+import { streamGemini } from "./gemini.js";
+import { streamMaaS } from "./maas.js";
+
+export function streamVertex(
+  model: VertexModelConfig,
+  context: Context,
+  options?: StreamOptions
+): AssistantMessageEventStream {
+  switch (model.endpointType) {
+    case "gemini":
+      return streamGemini(model, context, options);
+    case "maas":
+      return streamMaaS(model, context, options);
+    default:
+      throw new Error(`Unknown endpoint type: ${(model as any).endpointType}`);
+  }
+}
+
+export { streamGemini, streamMaaS };