Update Vertex docs for provider-based Anthropic Vertex support

docs/models/index.md

@@ -21,7 +21,7 @@ Model specifications in fast-agent follow this precedence order (highest to lowe
 
 Model strings follow this format: `provider.model_name[?reasoning=value][&query=value...]`
 
-- **provider**: The LLM provider (e.g., `anthropic`, `openai`, `azure`, `deepseek`, `generic`,`openrouter`, `tensorzero`)
+- **provider**: The LLM provider (for example `anthropic`, `anthropic-vertex`, `openai`, `azure`, `deepseek`, `generic`, `openrouter`, `tensorzero`)
 - **model_name**: The specific model to use in API calls (for Azure, this is your deployment name)
 - **query parameters** (optional): provider/model-specific runtime overrides such as
   `reasoning`, `structured`, `context`, `transport`, `temperature` (`temp` alias),
@@ -32,6 +32,7 @@ Model strings follow this format: `provider.model_name[?reasoning=value][&query=
 Examples:
 
 - `anthropic.claude-4-5-sonnet-latest`
+- `anthropic-vertex.claude-sonnet-4-6`
 - `openai.gpt-5.2`
 - `openai.o3-mini?reasoning=high`
 - `sonnet?reasoning=4096`

docs/models/llm_providers.md

@@ -54,6 +54,81 @@ anthropic:
 - `ANTHROPIC_API_KEY`: Your Anthropic API key
 - `ANTHROPIC_BASE_URL`: Override the API endpoint
 
+### Anthropic on Vertex AI
+
+Anthropic-on-Vertex is now exposed as a dedicated provider: `anthropic-vertex`.
+
+Use provider-prefixed model strings such as:
+
+- `anthropic-vertex.claude-sonnet-4-6`
+- `anthropic-vertex.claude-haiku-4-5`
+- `anthropic-vertex.claude-opus-4-6`
+
+Do not use the old query-string routing forms such as `?via=vertex` or `?source=vertex`.
+
+When you select `anthropic-vertex`:
+
+- fast-agent uses the dedicated Anthropic Vertex client path
+- **Google Application Default Credentials (ADC)** are required
+- an Anthropic direct API key is **not** required
+- the provider appears separately in model selection and diagnostics as **Anthropic (Vertex)**
+
+**YAML Configuration:**
+
+```yaml
+anthropic:
+  vertex_ai:
+    enabled: true
+    project_id: "my-gcp-project" # Optional if resolved from env/ADC
+    location: "global" # Optional, defaults to global
+    # base_url: "https://example-vertex-endpoint"
+```
+
+The `anthropic-vertex` provider reads its Vertex connection settings from
+`anthropic.vertex_ai`.
+
+**Environment Variables:**
+
+Project ID can be provided by any of:
+
+- `ANTHROPIC_VERTEX_PROJECT_ID`
+- `GOOGLE_CLOUD_PROJECT`
+- `GOOGLE_PROJECT_ID`
+- `GCLOUD_PROJECT`
+- `GCP_PROJECT`
+
+Location can be provided by any of:
+
+- `ANTHROPIC_VERTEX_LOCATION`
+- `GOOGLE_CLOUD_LOCATION`
+- `GOOGLE_CLOUD_REGION`
+- `CLOUD_ML_REGION`
+- `VERTEX_REGION`
+
+If location is not set, fast-agent defaults to `global`.
+
+**Authentication:**
+
+Anthropic-on-Vertex requires Google ADC, for example:
+
+```bash
+gcloud auth application-default login
+```
+
+or a service account exposed through the usual Google credential environment.
+
+**Example:**
+
+```yaml
+default_model: "anthropic-vertex.claude-sonnet-4-6"
+
+anthropic:
+  vertex_ai:
+    enabled: true
+    project_id: "my-gcp-project"
+    location: "global"
+```
+
 **Caching Options:**
 
 The `cache_mode` setting controls how prompt caching is applied:
@@ -528,6 +603,30 @@ google:
 
 - `GOOGLE_API_KEY`: Your Google API key
 
+### Vertex AI mode
+
+For Vertex AI, configure `google.vertex_ai` and authenticate with Google ADC/IAM instead of a
+Gemini API key:
+
+```yaml
+google:
+  vertex_ai:
+    enabled: true
+    project_id: "your-gcp-project"
+    location: "global" # or a supported regional location
+```
+
+In this mode, `fast-agent` uses the Google native provider without requiring `GOOGLE_API_KEY`.
+
+Vertex partner model ids such as Anthropic Claude are passed through unchanged, so with Vertex
+enabled you can use either:
+
+- `google.claude-sonnet-4-6`
+- `google.publishers/anthropic/models/claude-sonnet-4-6`
+
+If you want the dedicated Anthropic-flavoured route instead, use the separate
+`anthropic-vertex` provider described above.
+
 **Model Name Aliases:**
 
 --8<-- "_generated/model_aliases_google.md"

docs/models/models_reference.md

@@ -12,4 +12,11 @@ provider prefix when required). The provider column reflects the default
 provider used by fast-agent (models may become available with multiple
 providers in the future).
 
+Anthropic-on-Vertex is exposed as a separate provider using model strings such
+as `anthropic-vertex.claude-sonnet-4-6`. With `google.vertex_ai.enabled`, the
+native Google provider can also pass Vertex partner model ids such as
+`google.claude-sonnet-4-6` through unchanged. The table below still shows the
+default provider for each base model id; Vertex-specific routes are documented
+on [LLM Providers](llm_providers.md#anthropic-on-vertex-ai).
+
 --8<-- "_generated/models_reference.md"

docs/ref/config_file.md

@@ -145,6 +145,11 @@ anthropic:
   base_url: "https://api.anthropic.com/v1"  # Optional, only include to override
   reasoning: auto  # Adaptive models: auto/low/medium/high/max. Budget models: integer tokens or off.
   structured_output_mode: auto  # auto (default), json, or tool_use
+  vertex_ai:
+    enabled: false  # Optional; enables Anthropic-on-Vertex configuration intent
+    project_id: "my-gcp-project"  # Optional if resolved from env/ADC
+    location: "global"  # Optional; defaults to global
+    # base_url: "https://example-vertex-endpoint"
   web_search:
     enabled: false
     max_uses: 3  # Optional, must be > 0
@@ -193,6 +198,38 @@ Anthropic built-in web tools can also be toggled per run in the model string:
 
 Allowed values: `on`/`off` (also accepts `true`/`false`, `1`/`0`).
 
+Anthropic-on-Vertex now uses a dedicated provider namespace instead of model query routing.
+
+Use:
+
+- `anthropic-vertex.claude-sonnet-4-6`
+- `anthropic-vertex.claude-haiku-4-5`
+- `anthropic-vertex.claude-opus-4-6`
+
+Do not use the legacy `?via=vertex` / `?source=vertex` query parameters.
+
+The `anthropic-vertex` provider uses Google Application Default Credentials
+(ADC) and reads its connection settings from `anthropic.vertex_ai`. It does not
+require a direct Anthropic API key.
+
+Project ID can be sourced from:
+
+- `ANTHROPIC_VERTEX_PROJECT_ID`
+- `GOOGLE_CLOUD_PROJECT`
+- `GOOGLE_PROJECT_ID`
+- `GCLOUD_PROJECT`
+- `GCP_PROJECT`
+
+Location can be sourced from:
+
+- `ANTHROPIC_VERTEX_LOCATION`
+- `GOOGLE_CLOUD_LOCATION`
+- `GOOGLE_CLOUD_REGION`
+- `CLOUD_ML_REGION`
+- `VERTEX_REGION`
+
+If location is omitted, fast-agent defaults to `global`.
+
 ### OpenAI
 
 ```yaml
@@ -313,8 +350,19 @@ deepseek:
 google:
   api_key: "your_google_key"  # Can also use GOOGLE_API_KEY env var
   base_url: "https://generativelanguage.googleapis.com/v1beta/openai"  # Optional
+  vertex_ai:
+    enabled: true
+    project_id: "your-gcp-project"
+    location: "global"
 ```
 
+When `google.vertex_ai.enabled` is true, fast-agent uses ADC/IAM auth for the native Google
+provider and does not require `GOOGLE_API_KEY`.
+
+With Vertex enabled, the Google native provider also accepts partner model ids such as
+`google.claude-sonnet-4-6`. If you want the dedicated Anthropic-on-Vertex route instead, use the
+separate `anthropic-vertex` provider.
+
 ### xAI (Grok)
 
 ```yaml