feat: native Google provider (Gemini API and Vertex AI)

CHANGELOG.md

@@ -8,6 +8,12 @@ uses [Semantic Versioning](https://semver.org/).
 
 ### Added
 
+- **Google provider.** `google` (Gemini API key) and `google-vertex` (Vertex AI
+  with Google credentials: ADC or Workload Identity Federation, plus
+  `--google-project` and `--google-location`) run on Gemini without a base
+  URL. Agent loops use pydantic-ai's native `GoogleModel`, so Gemini 3 thought
+  signatures survive tool-call turns, and the cost summary prices Gemini runs
+  from genai-prices ([#20](https://github.com/FSoft-AI4Code/CodeWiki/issues/20)).
 - **Rust analyzer.** Structs, unions, enums, traits, free functions, and
   impl/trait methods become components, with trait-impl, field-type,
   struct-literal, and call edges. `#[cfg(test)]` modules are skipped, and

README.md

@@ -73,8 +73,8 @@ codewiki config set --provider claude-code \
   --main-model claude-sonnet-4-6 --cluster-model claude-sonnet-4-6
 ```
 
-Anthropic, Azure OpenAI, AWS Bedrock, Atlas Cloud, and Codex are also
-supported. See [Providers and models](./guides/providers.md).
+Anthropic, Azure OpenAI, AWS Bedrock, Google (Gemini API and Vertex AI),
+Atlas Cloud, and Codex are also supported. See [Providers and models](./guides/providers.md).
 
 **3. Generate**
 

codewiki/cli/adapters/doc_generator.py

@@ -151,6 +151,8 @@ def generate(self) -> DocumentationJob:
                 fallback_model=self.config.get("fallback_model"),
                 provider=self.config.get("provider", "openai-compatible"),
                 aws_region=self.config.get("aws_region", "us-east-1"),
+                google_project=self.config.get("google_project", ""),
+                google_location=self.config.get("google_location", ""),
                 max_tokens=self.config.get("max_tokens", 32768),
                 max_token_per_module=self.config.get("max_token_per_module", 36369),
                 max_token_per_leaf_module=self.config.get("max_token_per_leaf_module", 16000),

codewiki/cli/commands/config.py

@@ -69,12 +69,16 @@ def config_group():
             "azure-openai",
             "claude-code",
             "codex",
+            "google",
+            "google-vertex",
         ],
         case_sensitive=False,
     ),
     help=(
         "LLM provider type (default: openai-compatible). "
         "Use 'atlas-cloud' for Atlas Cloud (base URL auto-set; reads ATLASCLOUD_API_KEY). "
+        "Use 'google' for the Gemini API (reads GEMINI_API_KEY / GOOGLE_API_KEY) or "
+        "'google-vertex' for Vertex AI with Google credentials (no API key). "
         "Use 'claude-code' or 'codex' to run on a CLI subscription instead of an API key."
     ),
 )
@@ -83,6 +87,17 @@ def config_group():
     "--api-version", type=str, help="Azure OpenAI API version (default: 2024-12-01-preview)"
 )
 @click.option("--azure-deployment", type=str, help="Azure OpenAI deployment name")
+@click.option(
+    "--google-project",
+    type=str,
+    help="Google Cloud project for google-vertex (default: GOOGLE_CLOUD_PROJECT)",
+)
+@click.option(
+    "--google-location",
+    type=str,
+    help="Vertex AI location for google-vertex, e.g. global, eu, us "
+    "(default: GOOGLE_CLOUD_LOCATION, else global)",
+)
 @click.option(
     "--use-gitignore/--no-gitignore",
     default=None,
@@ -110,6 +125,8 @@ def config_set(
     aws_region: Optional[str] = None,
     api_version: Optional[str] = None,
     azure_deployment: Optional[str] = None,
+    google_project: Optional[str] = None,
+    google_location: Optional[str] = None,
     use_gitignore: Optional[bool] = None,
     prompt_caching: Optional[bool] = None,
 ):
@@ -142,6 +159,16 @@ def config_set(
     # Subscription mode (Codex)
     $ codewiki config set --provider codex --main-model gpt-5.2-codex
 
+    \b
+    # Vertex AI (Google credentials, no API key or base URL)
+    $ codewiki config set --provider google-vertex --google-project my-proj \\
+        --google-location eu --main-model gemini-3.8-flash --cluster-model gemini-3.8-flash
+
+    \b
+    # Gemini API (key read from GEMINI_API_KEY if --api-key is omitted)
+    $ codewiki config set --provider google --main-model gemini-3.8-flash \\
+        --cluster-model gemini-3.8-flash
+
     \b
     # Update only API key
     $ codewiki config set --api-key sk-new-key
@@ -181,6 +208,8 @@ def config_set(
                 aws_region,
                 api_version,
                 azure_deployment,
+                google_project,
+                google_location,
                 use_gitignore is not None,
                 prompt_caching is not None,
             ]
@@ -199,6 +228,20 @@ def config_set(
             if not api_key:
                 api_key = os.getenv("ATLASCLOUD_API_KEY")
 
+        manager = ConfigManager()
+        manager.load()  # Load existing config if present
+
+        if provider and provider.lower() == "google" and not api_key:
+            api_key = os.getenv("GEMINI_API_KEY") or os.getenv("GOOGLE_API_KEY")
+            previous = manager.get_config()
+            if not api_key and not (previous and previous.provider == "google"):
+                # The keyring has one key slot: reusing it would send another
+                # provider's key to Google.
+                raise ConfigurationError(
+                    "--provider google needs a Gemini API key: pass --api-key or set "
+                    "GEMINI_API_KEY (or GOOGLE_API_KEY)."
+                )
+
         # Validate inputs before saving
         validated_data = {}
 
@@ -265,9 +308,11 @@ def config_set(
         if prompt_caching is not None:
             validated_data["prompt_caching"] = prompt_caching
 
-        # Create config manager and save
-        manager = ConfigManager()
-        manager.load()  # Load existing config if present
+        if google_project is not None:
+            validated_data["google_project"] = google_project
+
+        if google_location is not None:
+            validated_data["google_location"] = google_location
 
         manager.save(
             api_key=validated_data.get("api_key"),
@@ -285,6 +330,8 @@ def config_set(
             aws_region=validated_data.get("aws_region"),
             api_version=validated_data.get("api_version"),
             azure_deployment=validated_data.get("azure_deployment"),
+            google_project=validated_data.get("google_project"),
+            google_location=validated_data.get("google_location"),
             use_gitignore=validated_data.get("use_gitignore"),
             prompt_caching=validated_data.get("prompt_caching"),
         )
@@ -352,6 +399,12 @@ def config_set(
         if azure_deployment:
             click.secho(f"✓ Azure Deployment: {azure_deployment}", fg="green")
 
+        if google_project:
+            click.secho(f"✓ Google project: {google_project}", fg="green")
+
+        if google_location:
+            click.secho(f"✓ Google location: {google_location}", fg="green")
+
         if use_gitignore is not None:
             click.secho(f"✓ Use gitignore: {use_gitignore}", fg="green")
 
@@ -405,6 +458,9 @@ def config_show(output_json: bool):
                 "api_key": mask_api_key(api_key) if api_key else "Not set",
                 "api_key_storage": "keychain" if manager.keyring_available else "encrypted_file",
                 "base_url": config.base_url if config else "",
+                "provider": config.provider if config else "",
+                "google_project": config.google_project if config else "",
+                "google_location": config.google_location if config else "",
                 "main_model": config.main_model if config else "",
                 "cluster_model": config.cluster_model if config else "",
                 "fallback_model": config.fallback_model if config else "glm-4p5",
@@ -430,12 +486,14 @@ def config_show(output_json: bool):
             click.echo("━" * 40)
             click.echo()
 
-            from codewiki.src.be.backend import is_caw_provider
+            from codewiki.src.be.backend import is_caw_provider, needs_base_url
 
             caw_mode = bool(config) and is_caw_provider(config.provider)
 
             click.secho("Credentials", fg="cyan", bold=True)
-            if caw_mode:
+            if config and config.provider == "google-vertex":
+                click.secho("  Google ADC / Workload Identity (no API key needed)", fg="cyan")
+            elif caw_mode:
                 cli_name = "claude" if config.provider == "claude-code" else "codex"
                 click.secho(
                     f"  Subscription mode: authenticate via '{cli_name} login' (no API key needed)",
@@ -453,10 +511,18 @@ def config_show(output_json: bool):
                 click.echo(f"  Provider:         {config.provider}")
                 click.echo(f"  Main Model:       {config.main_model or 'Not set'}")
                 if not caw_mode:
-                    click.echo(f"  Base URL:         {config.base_url or 'Not set'}")
+                    if needs_base_url(config.provider):
+                        click.echo(f"  Base URL:         {config.base_url or 'Not set'}")
                     click.echo(f"  Cluster Model:    {config.cluster_model or 'Not set'}")
                     click.echo(f"  Fallback Model:   {config.fallback_model or 'Not set'}")
-                    if config.provider == "bedrock":
+                    if config.provider == "google-vertex":
+                        project = config.google_project or "from GOOGLE_CLOUD_PROJECT"
+                        location = config.google_location or (
+                            "from GOOGLE_CLOUD_LOCATION, else global"
+                        )
+                        click.echo(f"  Google Project:   {project}")
+                        click.echo(f"  Google Location:  {location}")
+                    elif config.provider == "bedrock":
                         click.echo(f"  AWS Region:       {config.aws_region}")
                     elif config.provider == "azure-openai":
                         click.echo(f"  API Version:      {config.api_version}")
@@ -565,20 +631,29 @@ def config_validate(quick: bool, verbose: bool):
 
         # Load config early so we know the provider for the rest of the checks.
         config = manager.get_config()
-        from codewiki.src.be.backend import is_caw_provider
+        from codewiki.src.be.backend import (
+            is_caw_provider,
+            is_google_provider,
+            needs_api_key,
+            needs_base_url,
+        )
 
         caw_mode = bool(config) and is_caw_provider(config.provider)
+        key_needed = bool(config) and needs_api_key(config.provider)
+        url_needed = bool(config) and needs_base_url(config.provider)
+        api_key = None
 
-        # Step 2: Check API key (skipped for subscription providers)
+        # Step 2: Check API key (skipped for subscription providers and Vertex AI)
         if verbose:
             click.echo()
             click.echo("[2/5] Checking API key...")
 
-        if caw_mode:
+        if not key_needed:
+            reason = "subscription mode" if caw_mode else "Google credentials"
             if verbose:
-                click.secho("      ✓ API key not required (subscription mode)", fg="green")
+                click.secho(f"      ✓ API key not required ({reason})", fg="green")
             else:
-                click.secho("✓ API key not required (subscription mode)", fg="green")
+                click.secho(f"✓ API key not required ({reason})", fg="green")
         else:
             if verbose:
                 storage = "system keychain" if manager.keyring_available else "encrypted file"
@@ -597,16 +672,17 @@ def config_validate(quick: bool, verbose: bool):
             else:
                 click.secho("✓ API key present (stored in keychain)", fg="green")
 
-        # Step 3: Check base URL (skipped for subscription providers)
+        # Step 3: Check base URL (skipped for subscription and Google providers)
         if verbose:
             click.echo()
             click.echo("[3/5] Checking base URL...")
 
-        if caw_mode:
+        if not url_needed:
+            reason = "subscription mode" if caw_mode else "Google endpoint"
             if verbose:
-                click.secho("      ✓ Base URL not required (subscription mode)", fg="green")
+                click.secho(f"      ✓ Base URL not required ({reason})", fg="green")
             else:
-                click.secho("✓ Base URL not required (subscription mode)", fg="green")
+                click.secho(f"✓ Base URL not required ({reason})", fg="green")
         else:
             if verbose:
                 click.echo(f"      URL: {config.base_url}")
@@ -657,7 +733,7 @@ def config_validate(quick: bool, verbose: bool):
             # Warn about non-top-tier cluster model
             if not is_top_tier_model(config.cluster_model):
                 click.secho(
-                    "⚠️  Cluster model is not top-tier. Consider using claude-sonnet-4 or gpt-4.",
+                    "⚠️  Cluster model is not top-tier. Consider a Claude Sonnet/Opus, GPT-5 or Gemini 3 model.",
                     fg="yellow",
                 )
 
@@ -690,6 +766,26 @@ def config_validate(quick: bool, verbose: bool):
                     f"✓ {cli_name} CLI available (run '{cli_name} login' if not yet authenticated)",
                     fg="green",
                 )
+        elif is_google_provider(config.provider) and not quick:
+            if verbose:
+                click.echo()
+                click.echo("[5/5] Testing Google API access...")
+            try:
+                from codewiki.src.be import google_provider
+
+                google_provider.check_access(
+                    config.provider,
+                    api_key if key_needed else None,
+                    config.main_model,
+                    config.google_project,
+                    config.google_location,
+                )
+                click.secho("✓ API connectivity test successful", fg="green")
+            except Exception as e:
+                click.secho("✗ API connectivity test failed", fg="red")
+                if verbose:
+                    click.echo(f"      Error: {e}")
+                sys.exit(EXIT_CONFIG_ERROR)
         elif not quick:
             if verbose:
                 click.echo()

codewiki/cli/commands/generate.py

@@ -793,6 +793,8 @@ def generate_command(
                 "api_key": api_key,
                 "provider": getattr(config, "provider", "openai-compatible"),
                 "aws_region": getattr(config, "aws_region", "us-east-1"),
+                "google_project": getattr(config, "google_project", ""),
+                "google_location": getattr(config, "google_location", ""),
                 "agent_instructions": agent_instructions_dict,
                 # Max token settings (runtime overrides take precedence)
                 "max_tokens": max_tokens if max_tokens is not None else config.max_tokens,