docs: some updates to reflect changes

Author: Undiluted7027

.env.example

@@ -36,7 +36,7 @@ GEMINI_API_KEY="your_gemini_api_key_here"
 OLLAMA_HOST="your_ollama_server_url_here"
 
 # Vertex AI
-# GOGLE_API_KEY="your_google_vertex_api_key_here"
+# GOOGLE_API_KEY="your_google_vertex_api_key_here"
 # GOOGLE_CLOUD_PROJECT="your_google_cloud_project_id_here"
 # GOOGLE_CLOUD_LOCATION="your_google_cloud_location_here"
 # GOOGLE_GENAI_VERTEXAI="True"

.github/workflows/publish.yml

@@ -36,10 +36,7 @@ jobs:
         enable-cache: true
 
     - name: Install dependencies and project
-      run: uv sync --locked --all-extras --dev
-
-    - name: Install build dependencies
-      run: uv pip install build setuptools wheel
+      run: uv sync --locked --all-extras
 
     - name: Build package
       run: uv run python -m build

CHANGELOG.md

@@ -57,7 +57,6 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   - Gemini 3.x models (preview): `gemini-3-flash-preview`, `gemini-3-pro-preview`
   - Gemini 2.5.x models: `gemini-2.5-flash`, `gemini-2.5-pro`
   - Gemini 2.0.x models: `gemini-2.0-flash`
-  - Gemini 1.5.x models: `gemini-1.5-flash`, `gemini-1.5-pro`
   - Multiple authentication methods: API key, service accounts, ADC, impersonation
 - **Thought Signatures**: Automatic handling of Google Gemini 3.x thought signatures
   - New optional fields in `Message`, `ProviderResponse`, `ProviderChunk`

README.md

@@ -261,6 +261,7 @@ Allos supports a massive range of models through native integrations and a unive
 | **Cohere** | ✅ Ready | Command R+ |
 | **Ollama (compat)** | ✅ Ready | OpenAI-compatible endpoint mode |
 | **LocalAI / vLLM** | ✅ Ready | Custom endpoints via `chat_completions` |
+| **Google** | ✅ Ready | Gemini-3.0 Pro, Gemini-2.5 Flash |
 
 ## 🛠️ Built-in Tools
 
@@ -339,7 +340,7 @@ content_agent.run("Research AI trends and write a blog post")
 
 ### 🏗️ Phase 2: Upcoming
 - [x] **Native Ollama**: Advanced local model support with tool calling.
-- [ ] Google Gemini support
+- [x] Google Gemini support
 - [ ] Web search and fetch tools
 - [ ] Advanced context management
 - [ ] Plugin system
@@ -404,6 +405,12 @@ git clone https://github.com/Undiluted7027/allos-agent-sdk.git
 cd allos-agent-sdk
 ```
 
+For better DX, if you are on Linux/MacOS/WSL2, consider:
+```bash
+chmod +x scripts/setup_dev.sh
+./scripts/setup_dev
+```
+
 #### Python Environment
 
 ##### With `pip`
@@ -416,7 +423,7 @@ source venv/bin/activate
 # On Windows: venv\Scripts\activate
 
 # Install in development mode
-pip install -e ".[dev]"
+pip install -e ".[all, dev]"
 
 # Make the test script executable
 chmod +x scripts/run_tests.sh

docs/contributing/adding-providers.md

@@ -109,42 +109,10 @@ from . import nexusai # Add this line
 -   Add comprehensive unit tests for your provider in `tests/unit/test_nexusai_provider.py`. Use mocking to avoid making real API calls.
 -   Add integration tests in `tests/integration/test_nexusai_provider.py`. Mark them with the `run_integration_tests` decorator.
 
-Congratulations! You've successfully extended the Allos SDK with a new provider.
-
 ## Advanced: Supporting Thought Signatures
 
-Some LLM providers (like Google Gemini 3.x) use "thought signatures" - encrypted reasoning
-state that must be preserved across conversation turns.
-
-If you're implementing a provider that requires this:
-
-1. **Extract thought signatures from the API response**:
-   ```python
-   def _parse_response(self, response):
-       thought_signatures = {}
-       for part in response.parts:
-           if part.thought_signature:
-               thought_signatures[part.id] = part.thought_signature
-       return ProviderResponse(
-           content=...,
-           tool_calls=...,
-           thought_signatures=thought_signatures if thought_signatures else None
-       )
-   ```
-
-2. **Preserve thought signatures in message conversion**:
-   ```python
-   def _convert_messages(self, messages: List[Message]):
-       for msg in messages:
-           if msg.thought_signatures and msg.tool_calls:
-               # Include thought signatures when reconstructing tool call parts
-               for tc in msg.tool_calls:
-                   sig = msg.thought_signatures.get(tc.id)
-                   # Add signature to your API's request format
-   ```
-
-3. **Handle streaming**:
-   - Accumulate thought signatures across chunks
-   - Include in final metadata or in specific chunks
+Some LLM providers (like Google Gemini 3.x) use "thought signatures" - encrypted reasoning state that must be preserved across conversation turns.
 
 See `allos/providers/google.py` for a complete reference implementation.
+
+Congratulations! You've successfully extended the Allos SDK with a new provider.

docs/contributing/testing.md

@@ -6,6 +6,9 @@ Our test suite is built with `pytest` and is divided into three main categories.
 -   **End-to-End (E2E) Tests (`tests/e2e/`):** These tests validate the full application flow, primarily through the CLI. They use a **mocked LLM provider** but interact with the **real filesystem and tools**. They **do not** require real API keys to run.
 -   **Integration Tests (`tests/integration/`):** These tests verify the interaction between different parts of our SDK by making **real API calls** to external services (like OpenAI and Anthropic). They are slower, may incur costs, and require API keys and a special flag to run.
 
+> [!IMPORTANT]
+> **Python 3.10 or higher is required** for the Google (Gemini/Vertex AI) provider related tests. All other provider tests work with Python 3.9+. Use `-m "not_requires_python310"` to skip `GoogleProvider` related tests.
+
 ## Running Tests
 
 We provide a helper script for the most common testing scenario and command-line flags for more specific needs.
@@ -53,6 +56,9 @@ If you are missing a required API key, pytest will skip the relevant tests and p
 **Markers:**
 *   `@pytest.mark.requires_openai`: Used for tests hitting OpenAI directly OR using the `ChatCompletionsProvider` (which uses the `openai` library).
 *   `@pytest.mark.requires_anthropic`: Used for tests hitting Anthropic.
+* `@pytest.mark.requires_ollama`: Used for tests hitting Ollama.
+* `@pytest.mark.requires_gemini`: Used for tests hitting Gemini API.
+* `@pytest.mark.requires_vertexai`: Used for tests hitting Google Vertex AI API.
 
 ## Writing Tests
 

docs/guides/quickstart.md

@@ -10,7 +10,7 @@ Create and run your first "Hello, World!" application using the Allos agent in u
     ```
 2.  **Choose your provider**:
 
-    **Option A: Cloud Provider (OpenAI, Anthropic, Groq)**
+    **Option A: Cloud Provider (OpenAI, Anthropic, Groq, Google)**
 
     Create a `.env` file in your project directory with your API key:
     ```env

docs/installation.md

@@ -2,9 +2,15 @@
 
 This guide will walk you through installing the Allos Agent SDK. We recommend using `uv` for the fastest and most reliable experience.
 
+## Python Version Requirements
+
+Most providers work with Python 3.10+. However, **Google Gemini/Vertex AI provider requires Python 3.10+** due to dependencies on the Google GenAI SDK and [google-auth library](https://github.com/googleapis/google-auth-library-python).
+
+If you need to use Python 3.9, you can still use all other providers (OpenAI, Anthropic, Ollama, etc.).
+
 ## Prerequisites
 
-- Python 3.10+
+- Python 3.9+
 - We recommend installing `uv`, a next-generation Python package manager.
 
 ### Installing `uv`
@@ -36,12 +42,6 @@ This installs the core SDK without any specific provider libraries.
 uv pip install allos-agent-sdk
 ```
 
-## Python Version Requirements
-
-Most providers work with Python 3.10+. However, **Google Gemini/Vertex AI provider requires Python 3.10+** due to dependencies on the Google GenAI SDK and [google-auth library](https://github.com/googleapis/google-auth-library-python).
-
-If you need to use Python 3.9, you can still use all other providers (OpenAI, Anthropic, Ollama, etc.).
-
 ### Installation with Providers
 
 To use a specific provider, you can install it as an "extra".

docs/providers/google.md

@@ -239,12 +239,11 @@ provider = ProviderRegistry.get_provider(
 
 | Model | Context Window | Best For |
 |-------|---------------|----------|
-| `gemini-2.5-flash` | 1,048,576 tokens | Fastest responses, cost-effective |
-| `gemini-2.5-pro` | 1,048,576 tokens | Balanced performance |
-| `gemini-2.0-flash` | 1,048,576 tokens | Latest flash model |
-| `gemini-1.5-flash` | 1,048,576 tokens | Standard fast model |
-| `gemini-1.5-pro` | 2,097,152 tokens | Highest quality, largest context |
-| `gemini-1.0-pro` | 32,768 tokens | Legacy model |
+| `gemini-3-flash-preview` | 1,048,576 tokens | Speed, scale |
+| `gemini-3-pro-preview` | 1,048,576 tokens | Multimodality, agentic |
+| `gemini-2.5-flash` | 1,048,576 tokens | Low-latency, cost-effective |
+| `gemini-2.5-pro` | 1,048,576 tokens | Complex problems |
+| `gemini-2.5-flash-lite` | 1,048,576 tokens | Fastest model |
 
 For the latest model availability, refer to:
 - [Gemini API Models](https://ai.google.dev/gemini-api/docs/models)
@@ -350,7 +349,7 @@ Function call is missing a thought_signature in functionCall parts
 
 ---
 
-## Authentication Priority
+## Authentication Precedence
 
 The Google provider checks for authentication in this order:
 
@@ -475,44 +474,6 @@ allos --active-providers
 
 ---
 
-## Migration from Other Providers
-
-### From OpenAI
-
-```python
-# OpenAI
-from allos import Agent, AgentConfig
-
-config = AgentConfig(
-    provider_name="openai",
-    model="gpt-4"
-)
-
-# Google (simple replacement)
-config = AgentConfig(
-    provider_name="google",
-    model="gemini-2.0-flash"
-)
-```
-
-### From Anthropic
-
-```python
-# Anthropic
-config = AgentConfig(
-    provider_name="anthropic",
-    model="claude-3-5-sonnet-20240620"
-)
-
-# Google
-config = AgentConfig(
-    provider_name="google",
-    model="gemini-1.5-pro"  # Similar quality tier
-)
-```
-
----
-
 ## Best Practices
 
 1. **Use Gemini API for Development**: Simpler setup, faster iteration

docs/reference/provider-api.md

@@ -67,8 +67,8 @@ Different providers have different ways of identifying tool calls. The Allos SDK
 >
 >   -   **For most providers (like Anthropic):** The `id` is a straightforward unique identifier for the tool use block.
 >    -   **For OpenAI's Responses API:** The API uses two types of IDs:
-        -   An **item ID** (e.g., `"fc_..."`) that uniquely identifies the `function_call` object in the response list.
-        -   A **correlation ID** or `call_id` (e.g., `"call_..."`) that is used to link the call to its eventual result.
+>         -   An **item ID** (e.g., `"fc_..."`) that uniquely identifies the `function_call` object in the response list.
+>        -   A **correlation ID** or `call_id` (e.g., `"call_..."`) that is used to link the call to its eventual result.
 >
 >    **The `ToolCall.id` in the Allos SDK always stores the correlation ID (`call_id`).** Our `OpenAIProvider` handles the translation and management of the item IDs (`fc_...`) internally, so you only need to work with the correlation ID when building custom tools or inspecting the agent's context.