Add Anthropic Extended Thinking and Streaming Support by damithsenanayake · Pull Request #1 · radical-intelligence/adk-python

damithsenanayake · 2025-10-02T03:11:05Z

Summary

This PR adds support for Anthropic's Extended Thinking feature and Interleaved Thinking (beta) for Claude models in the ADK Python SDK. The implementation enables developers to use Claude's reasoning capabilities with thinking block signatures and multi-step tool reasoning, while maintaining full compatibility with existing code.

Key Features

Extended Thinking Support

Converts ADK ThinkingConfig to Anthropic API format
Supports disabled (0) and explicit (>0) budget values
Raises ValueError for unlimited budget (-1) as it's not supported by Claude
Works in both streaming and non-streaming modes
Parses thinking blocks as Part(thought=True, thought_signature=...) in standard GenAI format
Preserves cryptographic signatures from thinking blocks

Interleaved Thinking (Beta)

Enables Claude to reason between tool calls rather than thinking once upfront
Opt-in via enable_interleaved_thinking=True configuration
Sends anthropic-beta: interleaved-thinking-2025-05-14 header when enabled
Allows more sophisticated multi-step reasoning with tools
Thinking blocks automatically preserved in assistant message history

Improved Streaming

Migrated to AsyncAnthropicVertex for async streaming support
Implements thought suppression pattern to prevent UI fragmentation
Streaming controlled exclusively by stream parameter
Converts Anthropic streaming events to ADK LlmResponse format
Final message built from complete content blocks to preserve signatures

Backward Compatibility

100% backward compatible - no breaking changes
All existing tests pass
Non-streaming mode supports thinking blocks
Thinking blocks ordering preserved (must come first per Anthropic API requirement)
Interleaved thinking disabled by default

Implementation Details

Core Changes

File: src/google/adk/models/anthropic_llm.py

Import AsyncAnthropicVertex (line 33)
- Added async client for streaming support
Enhanced content_to_message_param() (lines 142-179)
- Separates thinking blocks from other content blocks
- Ensures thinking blocks come FIRST (Anthropic API requirement)
Enhanced content_block_to_part() (lines 182-219)
- Detects thinking blocks via thinking attribute or type='thinking'
- Extracts and preserves signature (base64-encoded bytes)
- Returns Part(thought=True, thought_signature=signature) for thinking content
- Logs signature presence for debugging
New streaming_event_to_llm_response() (lines 222-278)
- Converts Anthropic streaming events to ADK format
- Handles text_delta, thinking_delta, and message_delta events
- Returns LlmResponse with proper partial=True flag
Rewritten generate_content_async() (lines 366-508)
- Extracts and converts thinking config from ADK to Anthropic format
- Raises ValueError for budget=-1 (unlimited not supported)
- Applies budget directly (no minimum enforcement - handled by API)
- Configures beta header for interleaved thinking when enabled
- Streaming controlled only by stream parameter
- Accumulates thinking deltas and yields as single block (prevents UI fragmentation)
- Builds final response from final_message.content to preserve signatures
- Passes thinking and extra_headers parameters to both streaming and non-streaming API calls
Added enable_interleaved_thinking field (line 360)
- Boolean flag to enable interleaved thinking beta feature
- Default: False (opt-in)
Updated _anthropic_client property (lines 510-524)
- Changed from AnthropicVertex to AsyncAnthropicVertex

Test Coverage

Test Files (50 tests total):

test_anthropic_thinking.py (26 tests)
- Budget validation (raises error on -1, accepts 0 and positive values)
- Specific budget values (5000, 1024 minimum)
- Thinking block parsing with signatures (base64-encoded)
- Type-based thinking detection with signatures
- No-config baseline testing
- Streaming mode with thinking enabled
- NEW: Interleaved thinking tests (10 tests):
  - Beta header NOT sent by default
  - Beta header sent in streaming mode when enabled
  - Beta header sent in non-streaming mode when enabled
  - Beta header only sent when thinking config is active
  - Thinking blocks preserved in assistant message history
test_anthropic_streaming.py (8 tests)
- Event conversion (text_delta, thinking_delta, usage_delta)
- Start/stop event handling
- Event-to-response transformation
test_anthropic_llm.py (16 tests)
- Existing tests continue to pass
- No changes required

Code Quality

Style: Google Python Style Guide compliant
Formatting: Applied isort and pyink formatting
Documentation: Comprehensive inline comments explaining logic
Type Safety: Proper type hints and assertions
Signature Handling: Proper base64 encoding/decoding for cryptographic signatures

Testing Results

tests/unittests/models/test_anthropic_llm.py ............ 16/16 PASSED
tests/unittests/models/test_anthropic_thinking.py ...... 26/26 PASSED
tests/unittests/models/test_anthropic_streaming.py .....  8/8 PASSED

Total: 50/50 tests PASSED

No regressions - all existing tests continue to pass.

Breaking Changes

None. This PR is 100% backward compatible.

Usage Examples

Basic Extended Thinking

from google.genai import types
from google.adk.models.anthropic_llm import Claude
from google.adk.agents import Agent

# Enable extended thinking with explicit budget
agent = Agent(
    model=Claude(
        model="claude-opus-4-1@20250805",
        max_tokens=4096
    ),
    config=types.GenerateContentConfig(
        thinking_config=types.ThinkingConfig(
            include_thoughts=True,
            thinking_budget=2048  # Explicit budget (min 1024)
        )
    )
)

# Disable thinking
config=types.GenerateContentConfig(
    thinking_config=types.ThinkingConfig(
        include_thoughts=True,
        thinking_budget=0  # Disabled
    )
)

Interleaved Thinking with Tool Use

# Enable interleaved thinking for multi-step tool reasoning
agent = Agent(
    model=Claude(
        model="claude-opus-4-1@20250805",
        max_tokens=4096,
        enable_interleaved_thinking=True  # Beta: reason between tool calls
    ),
    config=types.GenerateContentConfig(
        thinking_config=types.ThinkingConfig(
            include_thoughts=True,
            thinking_budget=2048
        )
    ),
    tools=[calculator_tool, database_tool]
)

Accessing Thinking Blocks

# Streaming mode (explicit control)
async for response in model.generate_content_async(request, stream=True):
    if response.content:
        for part in response.content.parts:
            if part.thought:
                # Access thinking block with signature
                print(f"Thought: {part.text}")
                print(f"Signature: {part.thought_signature}")

Streaming Decision Logic

Streaming is controlled exclusively by the stream parameter:

stream=True: Uses streaming mode via messages.stream()
stream=False: Uses non-streaming mode via messages.create()

Both modes support thinking blocks when thinking parameter is provided.

Interleaved Thinking Details

When to Use:

Multi-step problems requiring reasoning between tool calls
Complex tool chains where intermediate results inform next steps
Scenarios where Claude needs to "think about" tool results before proceeding

How It Works:

Non-interleaved (default): Claude thinks once, makes all tool decisions upfront
Interleaved (beta): Claude can reason about tool results before deciding what to do next
Thinking blocks are automatically preserved when passing assistant messages back to API
Beta header only sent when BOTH enable_interleaved_thinking=True AND thinking is enabled

Requirements:

Claude 4 models (Opus 4, Sonnet 4, etc.)
Thinking config must be enabled (thinking_budget > 0)
Only supports automatic tool selection (tool_choice: auto)

Key Implementation Notes

Signature Preservation: Thinking blocks now preserve cryptographic signatures from Anthropic API
Thinking Block Ordering: Content blocks automatically reordered to place thinking blocks first (Anthropic requirement)
Non-Streaming Thinking: Non-streaming mode correctly passes thinking parameter to API
Budget Validation: Unlimited budget (-1) raises clear error message
Final Message: Streaming builds final response from content blocks (not accumulated strings) to preserve signatures
Beta Header: Conditionally set based on enable_interleaved_thinking flag and thinking config

pandemosth · 2025-10-02T08:15:44Z

Does this support tool use with interleaved thinking?
https://docs.claude.com/en/docs/build-with-claude/extended-thinking#tool-use-with-interleaved-thinking

pandemosth · 2025-10-02T08:21:30Z

+      budget = llm_request.config.thinking_config.thinking_budget
+      if budget:
+        if budget == -1:
+          # Automatic thinking budget - use recommended default of 10000 tokens


If Anthropic doesn't support automatic thinking budget we shouldn't fake it here. Throw an error instead as its important that the end user considers thinking budget for their use case as per https://docs.claude.com/en/docs/build-with-claude/extended-thinking#working-with-thinking-budgets

pandemosth · 2025-10-02T08:22:58Z

+    use_streaming = (
+        stream  # From runtime context (streaming_mode == SSE)
+        or thinking
+        != NOT_GIVEN  # Extended thinking requires streaming (Anthropic-specific)


Anthropic only supports extended thinking if streaming?

nope... only for larger thinking budgets. now revised.

damithsenanayake · 2025-10-02T12:40:22Z

Does this support tool use with interleaved thinking? https://docs.claude.com/en/docs/build-with-claude/extended-thinking#tool-use-with-interleaved-thinking

not yet: on it

PiperOrigin-RevId: 817267158

PiperOrigin-RevId: 817278990

… on tool confirmation requests PiperOrigin-RevId: 817373403

… case Agent developers can now create an eval set and add eval cases through command line itself. Adding an eval case is limited only to specifying conversation scenarios. Sample comamnds: - Create an eval set: adk eval_set create \ contributing/samples/hello_world \ set_01 - Add an eval case with scenario file Content of scenarios.json file: '{"scenarios": [{"starting_prompt": "hello", "conversation_plan": "world"}]}' adk eval_set add_eval_case \ contributing/samples/hello_world \ set_01 \ --scenarios scenarios.json PiperOrigin-RevId: 817456117

The added section provides details for the community call on Oct 15, 2025, including the agenda and links to join and add to calendars. PiperOrigin-RevId: 817457276

PiperOrigin-RevId: 817468275

This CL updates the "What's new" section to include Resumability, ReflectRetryToolPlugin, Context compaction, and Search tool support. It also moves "Agent Config" and "Tool Confirmation" from "What's new" to "Key Features". PiperOrigin-RevId: 817469210

This change removes the `evaluate`, `_evaluate_row`, `are_tools_equal`, `_remove_tool_outputs`, `_report_failures`, and `_print_results` static methods from `TrajectoryEvaluator`, along with their corresponding unit tests. These methods were previously marked as deprecated. PiperOrigin-RevId: 817477494

…ertexAiSearchTool PiperOrigin-RevId: 817493869

Scores with values 0 (or 0.0) were also getting excluded. PiperOrigin-RevId: 817658059

…ssue google#3131 changed the LiteLLM content conversion so Part.file_data.file_uri (like the gs://…) becomes a file object with file_id, making sure GCS-backed files reach LiteLLM proxies instead of being dropped add unit tests covering both _get_content and _content_to_message_param paths for file URIs PiperOrigin-RevId: 817658432

…l whether it's in progress updates (thought) or the final response PiperOrigin-RevId: 817682171

…gent PiperOrigin-RevId: 817766247

Directs agent to avoid deleting existing content PiperOrigin-RevId: 817823999

… for each recommended change PiperOrigin-RevId: 817831087

…earchTool and VertexAiSearchTool PiperOrigin-RevId: 818053371

…ineSearchTool For google#3146 PiperOrigin-RevId: 818458080

PiperOrigin-RevId: 818473733

This change removes the `run_evals` function and its helper `_get_evaluator` from `cli_eval.py`, as they were marked as deprecated. Corresponding test mocks and patches in `test_fast_api.py` are also removed. PiperOrigin-RevId: 818719422

PiperOrigin-RevId: 818765464

This is so we don't need to worry about side effect of Loop in all agent type. Custom agent should do the same if there exists loop inside. PiperOrigin-RevId: 818766305

PiperOrigin-RevId: 818781277

Update plugin manager and built-in plugins to prioritize CallbackContext. Keep InvocationContext access for legacy plugins with adapter. Change callback docs/tests to cover the new context. PiperOrigin-RevId: 818798087

…onstructing a2a request PiperOrigin-RevId: 818813897

Update plugin manager and built-in plugins to prioritize CallbackContext. Keep InvocationContext access for legacy plugins with adapter. Change callback docs/tests to cover the new context. PiperOrigin-RevId: 818822267

…oogle#1683 - add a shared --structured_logs flag to adk web and adk api_server so users can opt into JSON-formatted output - introduce CloudTraceJSONFormatter that emits structured entries and attaches current Cloud Trace/Span IDs when an OpenTelemetry context is active - update CLI logging setup to clear duplicate stdout handlers when Cloud Logging is enabled and to reconfigure existing handlers (like from Uvicorn) so they also pick up the structured format and requested log level With the flag disabled the CLIs keep their existing text logs; when enabled, the services now produce Cloud Logging–friendly JSON that can be correlated with distributed traces. PiperOrigin-RevId: 818823818

LiteLLM providers can extract the MIME type from the data URI. Removing the separate `format` field avoids redundancy and potential issues with backends that may reject requests containing this field. Close google#2017 Co-authored-by: George Weale <gweale@google.com> PiperOrigin-RevId: 828014286

Co-authored-by: George Weale <gweale@google.com> PiperOrigin-RevId: 828022792

…lbacks demo Related: google#2292 Co-authored-by: Shan Cao <caoshan@google.com> PiperOrigin-RevId: 828024955

This lets users to specify `drop_params` when initializing `LiteLlm`, which will be forwarded to LiteLLM's `acompletion` or `completion` calls Close google#1718 Co-authored-by: George Weale <gweale@google.com> PiperOrigin-RevId: 828058105

Co-authored-by: Yifan Wang <wanyif@google.com> PiperOrigin-RevId: 828061540

…LLM usage Closes google#3049 Co-authored-by: Eliza Huang <heliza@google.com> PiperOrigin-RevId: 828091671

Populate the usage_metadata field for live events with the metadata provided by the Gemini live API. Co-authored-by: Kathy Wu <wukathy@google.com> PiperOrigin-RevId: 828124232

PiperOrigin-RevId: 828126582

This change introduces a new section in the README.md to highlight the `adk-python-community` GitHub repository, describing it as a place for community-contributed tools and integrations. Co-authored-by: Hangfei Lin <hangfei@google.com> PiperOrigin-RevId: 828155205

Users were getting spammed with this log even though their tools didn't require authentication. To fix, reduce the log level to DEBUG so that it doesn't show up by default. Co-authored-by: Kathy Wu <wukathy@google.com> PiperOrigin-RevId: 828161281

Add support for MCP prompts via the McpInstructionProvider class, which can be specified as an agent's instruction. Co-authored-by: Kathy Wu <wukathy@google.com> PiperOrigin-RevId: 828166051

This change introduces BigQueryLoggerConfig to allow customization of the BigQueryAgentAnalyticsPlugin. Users can now enable/disable the plugin, specify event type allowlists and denylists, and provide a custom function to format or redact the content field before logging to BigQuery. The content logged for model and tool errors has also been enhanced. PiperOrigin-RevId: 828172241

…Engine Co-authored-by: Yeesian Ng <ysian@google.com> PiperOrigin-RevId: 828178479

Merge google#2651 ### Summary Correct a misspelling in the build configuration: - "swtich" → "switch" in `pyproject.toml`. ### Rationale This is a spelling fix only. It improves readability and avoids potential confusion in configuration. There is no impact on runtime behavior, tests, or public APIs. ### Notes - Follows Conventional Commits style for build/config changes (`build:`). - CLA status should be green via the Google CLA bot. COPYBARA_INTEGRATE_REVIEW=google#2651 from marsboy02:docs/fix-type-pyproject b78c014 PiperOrigin-RevId: 828221776

Merge google#3381 ### Link to Issue or Description of Change **1. Link to an existing issue (if applicable):** - Closes: google#3363 - This PR sets a max column width for the table printed in detailed output of agent evaluations. **Problem:** The detailed output of agent evaluations is not readable due to rows in the table getting wrapped. This happens when there are long text values in cells. <img width="1904" height="717" alt="508807185-9e8fe1c3-d04a-43dd-acf9-0befaa1b247d" src="https://github.com/user-attachments/assets/61526ad2-8a9e-4c18-83e2-51a3b9b32d2b" /> **Solution:** Existing code uses `tabulate` python package to format the table. We can set a maximum column width using `maxcolwidths` parameter. I have set it to `25`. After the fix: <img width="1882" height="711" alt="508810179-b91c5bca-fb43-480b-90ff-bca2e909417c" src="https://github.com/user-attachments/assets/b653f825-719e-4101-9acb-e28a52694cf8" /> ### Testing Plan I have manually tested if the output is properly displayed after changes. Please let me know if any unit tests can be added for this. **Unit Tests:** - [ ] I have added or updated unit tests for my change. - [x] All unit tests pass locally. <img width="1627" height="39" alt="image" src="https://github.com/user-attachments/assets/59a70619-3669-4113-8ab7-dcff130ee241" /> **Manual End-to-End (E2E) Tests:** 1. Create a simple agent using adk (preferably an agent that outputs a long text). 2. Create an evalset for this agent. 3. Run the evalset with `print_detailed_results` option and check if the output is properly displayed. If you want a quick setup for testing this, I have a sample repo with an agent and an evalset [here](https://github.com/nimanthadilz/adk-test/tree/reproduce-print-detailed-results). You will have to manually build & install the fixed adk version to test it. ### Checklist - [x] I have read the [CONTRIBUTING.md](https://github.com/google/adk-python/blob/main/CONTRIBUTING.md) document. - [x] I have performed a self-review of my own code. - [x] I have commented my code, particularly in hard-to-understand areas. - [ ] I have added tests that prove my fix is effective or that my feature works. - [x] New and existing unit tests pass locally with my changes. - [x] I have manually tested my changes end-to-end. - [x] Any dependent changes have been merged and published in downstream modules. COPYBARA_INTEGRATE_REVIEW=google#3381 from nimanthadilz:fix-eval-output-rows-wrapping-issue f6d4012 PiperOrigin-RevId: 828265715

PiperOrigin-RevId: 828293079

Co-authored-by: Xuan Yang <xygoogle@google.com> PiperOrigin-RevId: 828313025

Co-authored-by: Yeesian Ng <ysian@google.com> PiperOrigin-RevId: 828460860

The sample agent now uses updated model names for Gemini Live, including a new Vertex model as the default and a new AI Studio model option. Co-authored-by: Hangfei Lin <hangfei@google.com> PiperOrigin-RevId: 828515811

The blob content is often large and binary, which makes the logs unreadable and can cause excessive logging. Co-authored-by: Hangfei Lin <hangfei@google.com> PiperOrigin-RevId: 828523413

Co-authored-by: Xuan Yang <xygoogle@google.com> PiperOrigin-RevId: 828533243

PiperOrigin-RevId: 828538350

This allows state to be passing across agents Co-authored-by: Shangjie Chen <deanchen@google.com> PiperOrigin-RevId: 828557989

Co-authored-by: Xuan Yang <xygoogle@google.com> PiperOrigin-RevId: 828560608

PiperOrigin-RevId: 828579343

Merge google#3407 This PR corrects misspellings identified by the [check-spelling action](https://github.com/marketplace/actions/check-spelling) Note: while I use tooling to identify errors, the tooling doesn't _actually_ provide the corrections, I'm picking them on my own. I'm a human, and I may make mistakes. ### Testing Plan The misspellings have been reported at https://github.com/jsoref/adk-python/actions/runs/19056081305/attempts/1#summary-54426435973 The action reports that the changes in this PR would make it happy: https://github.com/jsoref/adk-python/actions/runs/19056081446/attempts/1#summary-54426436321 **Unit Tests:** - [ ] I have added or updated unit tests for my change. - [ ] All unit tests pass locally. _Please include a summary of passed `pytest` results._ **Manual End-to-End (E2E) Tests:** _Please provide instructions on how to manually test your changes, including any necessary setup or configuration. Please provide logs or screenshots to help reviewers better understand the fix._ ### Checklist - [x] I have read the [CONTRIBUTING.md](https://github.com/google/adk-python/blob/main/CONTRIBUTING.md) document. - [x] I have performed a self-review of my own code. - [ ] I have commented my code, particularly in hard-to-understand areas. - [ ] I have added tests that prove my fix is effective or that my feature works. - [ ] New and existing unit tests pass locally with my changes. - [ ] I have manually tested my changes end-to-end. - [ ] Any dependent changes have been merged and published in downstream modules. ### Additional context - google#3382 (comment) COPYBARA_INTEGRATE_REVIEW=google#3407 from jsoref:spelling-issue-template ce8febc PiperOrigin-RevId: 828610865

Replace the full JSON schema dump with a compact text summary of key AgentConfig components like LlmAgent, ToolConfig, and GenerateContentConfig Co-authored-by: George Weale <gweale@google.com> PiperOrigin-RevId: 828627911

- add FileArtifactService that persists artifacts to the local filesystem - adjust BaseArtifactService and exports so callers can wire in the filebacked implementation Co-authored-by: George Weale <gweale@google.com> PiperOrigin-RevId: 828629298

pandemosth reviewed Oct 2, 2025

View reviewed changes

Jacksunwei and others added 26 commits October 9, 2025 11:24

docs: Update CHANGELOG with [847df16](google@847df16)

0aede9f

PiperOrigin-RevId: 817267158

feat: Support ContentUnion as static instruction

cc24d61

PiperOrigin-RevId: 817278990

feat: Support resuming a parallel agent with multiple branches paused…

9939e0b

… on tool confirmation requests PiperOrigin-RevId: 817373403

chore: Announce the first ADK Community Call in the README

731bb90

The added section provides details for the community call on Oct 15, 2025, including the agenda and links to join and add to calendars. PiperOrigin-RevId: 817457276

fix: Create context cache only when prefix matches with previous request

9e0b1fb

PiperOrigin-RevId: 817468275

feat: Add bypass_multi_tools_limit option to GoogleSearchTool and V…

9a6b850

…ertexAiSearchTool PiperOrigin-RevId: 817493869

fix: Only exclude scores that are None

998264a

Scores with values 0 (or 0.0) were also getting excluded. PiperOrigin-RevId: 817658059

fix: Better handling the A2A streaming tasks so calling Agent can tel…

bddc70b

…l whether it's in progress updates (thought) or the final response PiperOrigin-RevId: 817682171

feat: Support resuming from a paused invocation starting from a sub-a…

59670d2

…gent PiperOrigin-RevId: 817766247

fix: Add a guideline to avoid content deletion

16b030b

Directs agent to avoid deleting existing content PiperOrigin-RevId: 817823999

fix: Add more clear instruction to the doc updater agent about one PR…

b21d0a5

… for each recommended change PiperOrigin-RevId: 817831087

feat: Set default for bypass_multi_tools_limit to False for GoogleS…

6da7274

…earchTool and VertexAiSearchTool PiperOrigin-RevId: 818053371

feat: Add the support for returning struct_data.uri in DiscoveryEng…

e63180c

…ineSearchTool For google#3146 PiperOrigin-RevId: 818458080

feat: Add new methods in the artifact service interface

e212ff5

PiperOrigin-RevId: 818473733

chore: Adjust the instruction about default model

214986e

PiperOrigin-RevId: 818765464

chore: Delegate the agent state reset logic to LoopAgent

bb1ea74

This is so we don't need to worry about side effect of Loop in all agent type. Custom agent should do the same if there exists loop inside. PiperOrigin-RevId: 818766305

chore: Correct the callback signatures

fa84bcb

PiperOrigin-RevId: 818781277

feat: migrate invocation_context to callback_context

e2072af

Update plugin manager and built-in plugins to prioritize CallbackContext. Keep InvocationContext access for legacy plugins with adapter. Change callback docs/tests to cover the new context. PiperOrigin-RevId: 818798087

fix: correctly populate context_id in remote_a2a_agent library when c…

2158b3c

…onstructing a2a request PiperOrigin-RevId: 818813897

feat: migrate invocation_context to callback_context

df05ed6

Update plugin manager and built-in plugins to prioritize CallbackContext. Keep InvocationContext access for legacy plugins with adapter. Change callback docs/tests to cover the new context. PiperOrigin-RevId: 818822267

GWeale and others added 30 commits November 4, 2025 09:44

docs: Refine ADK triaging agent labeling guidelines and response format

e25beb4

Co-authored-by: George Weale <gweale@google.com> PiperOrigin-RevId: 828022792

chore: Add model tracking to LiteLlm and introduce a LiteLLM with fal…

d4c63fc

…lbacks demo Related: google#2292 Co-authored-by: Shan Cao <caoshan@google.com> PiperOrigin-RevId: 828024955

fix: Pass drop_params to LiteLLM completion API

ce91a8e

This lets users to specify `drop_params` when initializing `LiteLlm`, which will be forwarded to LiteLLM's `acompletion` or `completion` calls Close google#1718 Co-authored-by: George Weale <gweale@google.com> PiperOrigin-RevId: 828058105

chore: update adk web for adk web builder: google/adk-web#248

abdc2bb

Co-authored-by: Yifan Wang <wanyif@google.com> PiperOrigin-RevId: 828061540

feat: Add support for extracting cache-related token counts from Lite…

4f85e86

…LLM usage Closes google#3049 Co-authored-by: Eliza Huang <heliza@google.com> PiperOrigin-RevId: 828091671

Merge branch 'main' into claude-thoughts-stream

2435dcc

feat: Add token usage to live events for bidi streaming

6e5c0eb

Populate the usage_metadata field for live events with the metadata provided by the Gemini live API. Co-authored-by: Kathy Wu <wukathy@google.com> PiperOrigin-RevId: 828124232

fix: update the contribution analysis tool to use original write mode

54db3d4

PiperOrigin-RevId: 828126582

feat: Support MCP prompts

88032cf

Add support for MCP prompts via the McpInstructionProvider class, which can be specified as an agent's instruction. Co-authored-by: Kathy Wu <wukathy@google.com> PiperOrigin-RevId: 828166051

feat: Add support for Vertex AI Express Mode when deploying to Agent …

d4b2a8b

…Engine Co-authored-by: Yeesian Ng <ysian@google.com> PiperOrigin-RevId: 828178479

feat: Refactor gepa sample code and clean-up user demo colab

63353b2

PiperOrigin-RevId: 828293079

chore: Bumps version to v1.18.0 and updates CHANGELOG.md

e01e54e

Co-authored-by: Xuan Yang <xygoogle@google.com> PiperOrigin-RevId: 828313025

fix: Update dependency version constraints to be based on PyPI versions

0b1784e

Co-authored-by: Yeesian Ng <ysian@google.com> PiperOrigin-RevId: 828460860

chore: Update Gemini Live model names in live bidi streaming sample

aa77834

The sample agent now uses updated model names for Gemini Live, including a new Vertex model as the default and a new AI Studio model option. Co-authored-by: Hangfei Lin <hangfei@google.com> PiperOrigin-RevId: 828515811

chore: Stop logging the full content of LLM blobs

0826755

The blob content is often large and binary, which makes the logs unreadable and can cause excessive logging. Co-authored-by: Hangfei Lin <hangfei@google.com> PiperOrigin-RevId: 828523413

chore: Update v1.18.0 wheel and CHANGELOG.md

d8d7774

Co-authored-by: Xuan Yang <xygoogle@google.com> PiperOrigin-RevId: 828533243

fix: Add None check for event in remote_a2a_agent.py

744f94f

PiperOrigin-RevId: 828538350

chore: Returns agent state regardless if ctx.is_resumable

d6b928b

This allows state to be passing across agents Co-authored-by: Shangjie Chen <deanchen@google.com> PiperOrigin-RevId: 828557989

feat: Add on_model_error_callback in LlmAgent

9ec38c0

Co-authored-by: Xuan Yang <xygoogle@google.com> PiperOrigin-RevId: 828560608

feat: Improve gepa tau-bench colab for external use

e02f177

PiperOrigin-RevId: 828579343

feat: add file-backed artifact service

99ca6aa

- add FileArtifactService that persists artifacts to the local filesystem - adjust BaseArtifactService and exports so callers can wire in the filebacked implementation Co-authored-by: George Weale <gweale@google.com> PiperOrigin-RevId: 828629298

Merge branch 'main' into claude-thoughts-stream

9a92aa5

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add Anthropic Extended Thinking and Streaming Support#1

Add Anthropic Extended Thinking and Streaming Support#1
damithsenanayake wants to merge 321 commits into
mainfrom
claude-thoughts-stream

damithsenanayake commented Oct 2, 2025 •

edited

Loading

Uh oh!

pandemosth commented Oct 2, 2025

Uh oh!

pandemosth Oct 2, 2025

Uh oh!

damithsenanayake Oct 2, 2025

Uh oh!

pandemosth Oct 2, 2025

Uh oh!

damithsenanayake Oct 2, 2025

Uh oh!

damithsenanayake commented Oct 2, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

20 participants

Conversation

damithsenanayake commented Oct 2, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Key Features

Extended Thinking Support

Interleaved Thinking (Beta)

Improved Streaming

Backward Compatibility

Implementation Details

Core Changes

Test Coverage

Code Quality

Testing Results

Breaking Changes

Usage Examples

Basic Extended Thinking

Interleaved Thinking with Tool Use

Accessing Thinking Blocks

Streaming Decision Logic

Interleaved Thinking Details

Key Implementation Notes

Uh oh!

pandemosth commented Oct 2, 2025

Uh oh!

pandemosth Oct 2, 2025

Choose a reason for hiding this comment

Uh oh!

damithsenanayake Oct 2, 2025

Choose a reason for hiding this comment

Uh oh!

pandemosth Oct 2, 2025

Choose a reason for hiding this comment

Uh oh!

damithsenanayake Oct 2, 2025

Choose a reason for hiding this comment

Uh oh!

damithsenanayake commented Oct 2, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

20 participants

damithsenanayake commented Oct 2, 2025 •

edited

Loading