@ssweens/pi-vertex 1.0.1 → 1.1.3

package/CHANGELOG.md

@@ -0,0 +1,17 @@
+# Changelog
+
+All notable changes to this project will be documented in this file.
+
+## [1.1.3] - 2026-03-26
+### Fixed
+- Hardened Claude-on-Vertex replay for mid-session model switching (tool ID normalization, tool result adjacency, thinking signature validation).
+- Prevented Anthropic tool replay errors by inserting synthetic tool results when missing.
+
+### Updated
+- Claude 4.6 models use native Anthropic Vertex SDK streaming.
+- Claude 4.6 context window updated to 1M.
+- Model list order in the selector is now alphabetized by ID.
+
+## [1.1.2] - 2026-03-24
+### Changed
+- Initial Claude 4.x support on Vertex.

package/README.md

@@ -22,10 +22,11 @@ Set your GCP project and credentials. Vertex AI models (Gemini, Claude, Llama, D
   - **Other MaaS** (20): AI21 Jamba, Mistral, DeepSeek, Qwen, OpenAI GPT-OSS, Kimi, MiniMax, GLM
 
 - **Unified streaming**: Single provider, multiple model families
-- **Full tool calling support**: All models marked with tools support
+- **Full tool calling support**: All models with multi-turn tool use and proper tool result handling
+- **Thinking/reasoning**: Gemini 3 thinking levels, Gemini 2.5 thinking budgets, thought signature preservation
 - **Automatic auth**: Uses Google Application Default Credentials
 - **Region awareness**: Global endpoints where supported, regional where required
-- **Pricing tracking**: Built-in cost per token for all models
+- **Pricing tracking**: Built-in cost per token for all models (including thinking tokens)
 
 ## Installation
 
@@ -115,31 +116,29 @@ alias pil="GOOGLE_CLOUD_PROJECT=your-project pi --provider vertex --model llama-
 
 | Model | Context | Max Tokens | Input | Reasoning | Price (in/out) |
 |-------|---------|------------|-------|-----------|----------------|
-| gemini-3.1-pro | 1M | 64,000 | text, image | ✅ | $2.00/$12.00 |
-| gemini-3-pro | 2M | 8,192 | text, image | ✅ | $1.25/$10.00 |
-| gemini-3-flash | 1M | 8,192 | text, image | ✅ | $0.15/$0.60 |
-| gemini-2.5-pro | 1M | 64,000 | text, image | ✅ | $1.25/$10.00 |
-| gemini-2.5-flash | 1M | 64,000 | text, image | ✅ | $0.30/$2.50 |
-| gemini-2.5-flash-lite | 1M | 64,000 | text, image | ✅ | $0.10/$0.40 |
+| gemini-3.1-pro | 1M | 65,536 | text, image | ✅ | $2.00/$12.00 |
+| gemini-3.1-flash-lite | 1M | 65,535 | text, image | ✅ | $0.25/$1.50 |
+| gemini-3-pro | 1M | 65,536 | text, image | ✅ | $2.00/$12.00 |
+| gemini-3-flash | 1M | 65,536 | text, image | ✅ | $0.50/$3.00 |
+| gemini-2.5-pro | 1M | 65,536 | text, image | ✅ | $1.25/$10.00 |
+| gemini-2.5-flash | 1M | 65,536 | text, image | ✅ | $0.30/$2.50 |
+| gemini-2.5-flash-lite | 1M | 65,536 | text, image | ✅ | $0.10/$0.40 |
 | gemini-2.0-flash | 1M | 8,192 | text, image | ❌ | $0.15/$0.60 |
-| gemini-2.0-flash-lite | 1M | 8,192 | text | ❌ | $0.075/$0.30 |
+| gemini-2.0-flash-lite | 1M | 8,192 | text, image | ❌ | $0.075/$0.30 |
 
 ### Claude Models
 
 | Model | Context | Max Tokens | Input | Reasoning | Price (in/out) | Region |
 |-------|---------|------------|-------|-----------|----------------|--------|
-| claude-opus-4-6 | 200K | 32,000 | text, image | ✅ | $5.00/$25.00 | global |
-| claude-sonnet-4-6 | 200K | 64,000 | text, image | ✅ | $3.00/$15.00 | global |
+| claude-opus-4-6 | 1M | 32,000 | text, image | ✅ | $5.00/$25.00 | global |
+| claude-sonnet-4-6 | 1M | 64,000 | text, image | ✅ | $3.00/$15.00 | global |
 | claude-opus-4-5 | 200K | 32,000 | text, image | ✅ | $5.00/$25.00 | global |
 | claude-sonnet-4-5 | 200K | 64,000 | text, image | ✅ | $3.00/$15.00 | global |
 | claude-haiku-4-5 | 200K | 64,000 | text, image | ✅ | $1.00/$5.00 | global |
-| claude-opus-4-1 | 200K | 32,000 | text, image | ✅ | $15.00/$75.00 | us-east5 |
-| claude-opus-4 | 200K | 32,000 | text, image | ✅ | $15.00/$75.00 | us-east5 |
-| claude-sonnet-4 | 200K | 64,000 | text, image | ✅ | $3.00/$15.00 | us-east5 |
-| claude-3-7-sonnet | 200K | 64,000 | text, image | ✅ | $3.00/$15.00 | us-east5 |
-| claude-3-5-sonnet-v2 | 200K | 8,192 | text, image | ❌ | $3.00/$15.00 | us-east5 |
-| claude-3-5-sonnet | 200K | 8,192 | text, image | ❌ | $3.00/$15.00 | us-east5 |
-| claude-3-haiku | 200K | 4,096 | text | ❌ | $0.25/$1.25 | us-east5 |
+| claude-opus-4-1 | 200K | 32,000 | text, image | ✅ | $15.00/$75.00 | global |
+| claude-opus-4 | 200K | 32,000 | text, image | ✅ | $15.00/$75.00 | global |
+| claude-sonnet-4 | 200K | 64,000 | text, image | ✅ | $3.00/$15.00 | global |
+| claude-3-5-sonnet-v2 | 200K | 8,192 | text, image | ❌ | $3.00/$15.00 | global |
 
 ### Llama Models
 
@@ -153,8 +152,6 @@ alias pil="GOOGLE_CLOUD_PROJECT=your-project pi --provider vertex --model llama-
 
 | Model | Context | Publisher | Price (in/out) | Region |
 |-------|---------|-----------|----------------|--------|
-| jamba-1.5-large | 256K | ai21 | $2.00/$8.00 | global |
-| jamba-1.5-mini | 256K | ai21 | $0.20/$0.40 | global |
 | mistral-medium-3 | 128K | mistralai | $0.40/$2.00 | global |
 | mistral-small-3.1 | 128K | mistralai | $0.10/$0.30 | global |
 | mistral-ocr | 30 pages | mistralai | $0.0005/page | global |
@@ -183,8 +180,7 @@ Models use different endpoints based on availability:
 
 Default regions by model:
 - Gemini: `global`
-- Claude 4.6/4.5: `global`
-- Claude 4/4.1/3.7/3.5/3: `us-east5`
+- Claude (all): `global`
 - MaaS: `global`
 
 Override with:
@@ -217,6 +213,7 @@ export GOOGLE_CLOUD_LOCATION=us-central1
 ## Dependencies
 
 - `@google/genai`: Google GenAI SDK for Gemini models
+- `@anthropic-ai/vertex-sdk`: Official Anthropic-on-Vertex SDK for Claude models (native streaming)
 - `google-auth-library`: ADC authentication for all models
 - `@mariozechner/pi-ai`: Peer dependency
 - `@mariozechner/pi-coding-agent`: Peer dependency

package/TEST_COVERAGE.md

@@ -0,0 +1,13 @@
+# Test Coverage
+
+## Current Status
+- Automated tests: not yet implemented in this package.
+- Lint/type checks: `npm run check` (currently a no-op placeholder).
+
+## Manual Verification
+- Claude 4.6 streaming verified via Anthropic Vertex SDK.
+- Mid-session model switching (tool call replay) verified interactively in pi.
+
+## Gaps / Next Steps
+- Add automated integration tests for Anthropic Vertex streaming and tool replay.
+- Add unit tests for message normalization and replay sequencing.

package/index.ts

@@ -112,8 +112,8 @@ export default function (pi: ExtensionAPI) {
 
   // Show startup info as a widget that clears on first user input
   const vertexStartupLines = [
-    `[pi-vertex] Initializing with project: ${projectId}`,
-    `[pi-vertex] Registered ${ALL_MODELS.length} models`,
+    `   [pi-vertex] Initializing with project: ${projectId}`,
+    `   [pi-vertex] Registered ${ALL_MODELS.length} models`,
   ];
   pi.on("session_start", async (_event, ctx) => {
     ctx.ui.setWidget("pi-vertex-startup", (_tui, theme) => ({

package/models/claude.ts

@@ -1,21 +1,22 @@
 /**
  * Claude model definitions for Vertex AI
+ * Source: https://cloud.google.com/vertex-ai/generative-ai/docs/partner-models/use-partner-models
  * Pricing: https://cloud.google.com/vertex-ai/generative-ai/pricing#partner-models
- * All prices per 1M tokens (as of Feb 2025)
- * Cache write prices shown are for 5-minute TTL
+ * All prices per 1M tokens (global endpoint, <= 200K input tokens)
+ * Cache write prices are for 5-minute TTL
  */
 
 import type { VertexModelConfig } from "../types.js";
 
 export const CLAUDE_MODELS: VertexModelConfig[] = [
-  // Claude 4.6 series - latest, supports global endpoint
+  // Claude 4.6 series
   {
     id: "claude-opus-4-6",
     name: "Claude Opus 4.6",
     apiId: "claude-opus-4-6",
     publisher: "anthropic",
     endpointType: "maas",
-    contextWindow: 200000,
+    contextWindow: 1000000,
     maxTokens: 32000,
     input: ["text", "image"],
     reasoning: true,
@@ -34,7 +35,7 @@ export const CLAUDE_MODELS: VertexModelConfig[] = [
     apiId: "claude-sonnet-4-6",
     publisher: "anthropic",
     endpointType: "maas",
-    contextWindow: 200000,
+    contextWindow: 1000000,
     maxTokens: 64000,
     input: ["text", "image"],
     reasoning: true,
@@ -47,7 +48,8 @@ export const CLAUDE_MODELS: VertexModelConfig[] = [
     },
     region: "global",
   },
-  // Claude 4.5 series - supports global endpoint
+
+  // Claude 4.5 series
   {
     id: "claude-opus-4-5",
     name: "Claude Opus 4.5",
@@ -105,7 +107,8 @@ export const CLAUDE_MODELS: VertexModelConfig[] = [
     },
     region: "global",
   },
-  // Claude 4.1 series - regional pricing
+
+  // Claude 4.1 series
   {
     id: "claude-opus-4-1",
     name: "Claude Opus 4.1",
@@ -123,9 +126,10 @@ export const CLAUDE_MODELS: VertexModelConfig[] = [
       cacheRead: 1.50,
       cacheWrite: 18.75,
     },
-    region: "us-east5",
+    region: "global",
   },
-  // Claude 4.0 series - regional pricing
+
+  // Claude 4.0 series
   {
     id: "claude-opus-4",
     name: "Claude Opus 4",
@@ -143,7 +147,7 @@ export const CLAUDE_MODELS: VertexModelConfig[] = [
       cacheRead: 1.50,
       cacheWrite: 18.75,
     },
-    region: "us-east5",
+    region: "global",
   },
   {
     id: "claude-sonnet-4",
@@ -162,29 +166,10 @@ export const CLAUDE_MODELS: VertexModelConfig[] = [
       cacheRead: 0.30,
       cacheWrite: 3.75,
     },
-    region: "us-east5",
-  },
-  // Claude 3.7 series - regional pricing
-  {
-    id: "claude-3-7-sonnet",
-    name: "Claude 3.7 Sonnet",
-    apiId: "claude-3-7-sonnet@20250219",
-    publisher: "anthropic",
-    endpointType: "maas",
-    contextWindow: 200000,
-    maxTokens: 64000,
-    input: ["text", "image"],
-    reasoning: true,
-    tools: true,
-    cost: {
-      input: 3.0,
-      output: 15.0,
-      cacheRead: 0.3,
-      cacheWrite: 3.75,
-    },
-    region: "us-east5",
+    region: "global",
   },
-  // Claude 3.5 series - regional pricing
+
+  // Claude 3.5 series
   {
     id: "claude-3-5-sonnet-v2",
     name: "Claude 3.5 Sonnet v2",
@@ -197,50 +182,11 @@ export const CLAUDE_MODELS: VertexModelConfig[] = [
     reasoning: false,
     tools: true,
     cost: {
-      input: 3.0,
-      output: 15.0,
-      cacheRead: 0.3,
-      cacheWrite: 3.75,
-    },
-    region: "us-east5",
-  },
-  {
-    id: "claude-3-5-sonnet",
-    name: "Claude 3.5 Sonnet",
-    apiId: "claude-3-5-sonnet@20240620",
-    publisher: "anthropic",
-    endpointType: "maas",
-    contextWindow: 200000,
-    maxTokens: 8192,
-    input: ["text", "image"],
-    reasoning: false,
-    tools: true,
-    cost: {
-      input: 3.0,
-      output: 15.0,
-      cacheRead: 0.3,
+      input: 3.00,
+      output: 15.00,
+      cacheRead: 0.30,
       cacheWrite: 3.75,
     },
-    region: "us-east5",
-  },
-  // Claude 3 Haiku - regional pricing
-  {
-    id: "claude-3-haiku",
-    name: "Claude 3 Haiku",
-    apiId: "claude-3-haiku@20240307",
-    publisher: "anthropic",
-    endpointType: "maas",
-    contextWindow: 200000,
-    maxTokens: 4096,
-    input: ["text"],
-    reasoning: false,
-    tools: true,
-    cost: {
-      input: 0.25,
-      output: 1.25,
-      cacheRead: 0.03,
-      cacheWrite: 0.3,
-    },
-    region: "us-east5",
+    region: "global",
   },
 ];

package/models/gemini.ts

@@ -1,77 +1,83 @@
 /**
  * Gemini model definitions for Vertex AI
- * Pricing: https://cloud.google.com/vertex-ai/generative-ai/pricing#gemini-models
- * All prices per 1M tokens (Standard tier pricing, as of Feb 2026)
+ * Source: https://docs.cloud.google.com/vertex-ai/generative-ai/docs/models
+ * Pricing: https://cloud.google.com/vertex-ai/generative-ai/pricing
+ * All prices per 1M tokens (standard tier, <= 200K input tokens)
  */
 
 import type { VertexModelConfig } from "../types.js";
 
 export const GEMINI_MODELS: VertexModelConfig[] = [
+  // --- Gemini 3.1 (Preview) ---
   {
     id: "gemini-3.1-pro",
     name: "Gemini 3.1 Pro",
     apiId: "gemini-3.1-pro-preview",
     publisher: "google",
     endpointType: "gemini",
-    contextWindow: 1000000,
-    maxTokens: 64000,
+    contextWindow: 1048576,
+    maxTokens: 65536,
     input: ["text", "image"],
     reasoning: true,
     tools: true,
     cost: {
       input: 2.00,
       output: 12.00,
-      cacheRead: 0,
+      cacheRead: 0.20,
       cacheWrite: 0,
     },
     region: "global",
   },
   {
-    id: "gemini-3-pro",
-    name: "Gemini 3 Pro",
-    apiId: "gemini-3-pro-preview",
+    id: "gemini-3.1-flash-lite",
+    name: "Gemini 3.1 Flash Lite",
+    apiId: "gemini-3.1-flash-lite-preview",
     publisher: "google",
     endpointType: "gemini",
-    contextWindow: 2000000,
-    maxTokens: 8192,
+    contextWindow: 1048576,
+    maxTokens: 65535,
     input: ["text", "image"],
     reasoning: true,
     tools: true,
     cost: {
-      input: 1.25,
-      output: 10.00,
-      cacheRead: 0.125,
+      input: 0.25,
+      output: 1.50,
+      cacheRead: 0.025,
       cacheWrite: 0,
     },
     region: "global",
   },
+
+  // --- Gemini 3 (Preview) ---
   {
     id: "gemini-3-flash",
     name: "Gemini 3 Flash",
     apiId: "gemini-3-flash-preview",
     publisher: "google",
     endpointType: "gemini",
-    contextWindow: 1000000,
-    maxTokens: 8192,
+    contextWindow: 1048576,
+    maxTokens: 65536,
     input: ["text", "image"],
     reasoning: true,
     tools: true,
     cost: {
-      input: 0.15,
-      output: 0.60,
-      cacheRead: 0.0375,
+      input: 0.50,
+      output: 3.00,
+      cacheRead: 0.05,
       cacheWrite: 0,
     },
     region: "global",
   },
+
+  // --- Gemini 2.5 (GA) ---
   {
     id: "gemini-2.5-pro",
     name: "Gemini 2.5 Pro",
     apiId: "gemini-2.5-pro",
     publisher: "google",
     endpointType: "gemini",
-    contextWindow: 1000000,
-    maxTokens: 64000,
+    contextWindow: 1048576,
+    maxTokens: 65536,
     input: ["text", "image"],
     reasoning: true,
     tools: true,
@@ -89,15 +95,15 @@ export const GEMINI_MODELS: VertexModelConfig[] = [
     apiId: "gemini-2.5-flash",
     publisher: "google",
     endpointType: "gemini",
-    contextWindow: 1000000,
-    maxTokens: 64000,
+    contextWindow: 1048576,
+    maxTokens: 65536,
     input: ["text", "image"],
     reasoning: true,
     tools: true,
     cost: {
       input: 0.30,
       output: 2.50,
-      cacheRead: 0.030,
+      cacheRead: 0.03,
       cacheWrite: 0,
     },
     region: "global",
@@ -108,26 +114,28 @@ export const GEMINI_MODELS: VertexModelConfig[] = [
     apiId: "gemini-2.5-flash-lite",
     publisher: "google",
     endpointType: "gemini",
-    contextWindow: 1000000,
-    maxTokens: 64000,
+    contextWindow: 1048576,
+    maxTokens: 65536,
     input: ["text", "image"],
     reasoning: true,
     tools: true,
     cost: {
       input: 0.10,
       output: 0.40,
-      cacheRead: 0.010,
+      cacheRead: 0.01,
       cacheWrite: 0,
     },
     region: "global",
   },
+
+  // --- Gemini 2.0 (GA) ---
   {
     id: "gemini-2.0-flash",
     name: "Gemini 2.0 Flash",
     apiId: "gemini-2.0-flash",
     publisher: "google",
     endpointType: "gemini",
-    contextWindow: 1000000,
+    contextWindow: 1048576,
     maxTokens: 8192,
     input: ["text", "image"],
     reasoning: false,
@@ -135,7 +143,7 @@ export const GEMINI_MODELS: VertexModelConfig[] = [
     cost: {
       input: 0.15,
       output: 0.60,
-      cacheRead: 0.025,
+      cacheRead: 0,
       cacheWrite: 0,
     },
     region: "global",
@@ -146,15 +154,15 @@ export const GEMINI_MODELS: VertexModelConfig[] = [
     apiId: "gemini-2.0-flash-lite",
     publisher: "google",
     endpointType: "gemini",
-    contextWindow: 1000000,
+    contextWindow: 1048576,
     maxTokens: 8192,
-    input: ["text"],
+    input: ["text", "image"],
     reasoning: false,
     tools: true,
     cost: {
       input: 0.075,
       output: 0.30,
-      cacheRead: 0.01875,
+      cacheRead: 0,
       cacheWrite: 0,
     },
     region: "global",

package/models/index.ts

@@ -11,7 +11,7 @@ export const ALL_MODELS: VertexModelConfig[] = [
   ...GEMINI_MODELS,
   ...CLAUDE_MODELS,
   ...MAAS_MODELS,
-];
+].sort((a, b) => a.id.localeCompare(b.id));
 
 export function getModelById(id: string): VertexModelConfig | undefined {
   return ALL_MODELS.find((m) => m.id === id);