Please set --log_level to DEBUG, if you are interested in having those API request and responses in logs.
NOTE: Generally it is not recommended to have DEBUG log level for services that run in a production setting. It is our recommendation to only use DEBUG log level in a debug or development setting.
PiperOrigin-RevId: 785972338
Merge https://github.com/google/adk-python/pull/1959
### What
Fix misleading comment.
```diff
- # Make sure a malicious user can obtain a session and events not belonging to them
+ # Make sure a malicious user **cannot** obtain a session or events not belonging to them
```
### Why
The previous wording contradicted the assertion `assert len(session_mismatch.events) == 0`, which verifies that a malicious user **cannot** access another user’s session or events.
### Testing plan
Docs-only change.
COPYBARA_INTEGRATE_REVIEW=https://github.com/google/adk-python/pull/1959 from mthorme:fix-comment-session-mismatch b1f139af340bd240d40ed58f5dea3274c3a3bd83
PiperOrigin-RevId: 785908548
This change takes cares of SQL results containing values that are not json serializable (e.g. datetime, bignumeric) by converting them to their string representation.
PiperOrigin-RevId: 785719997
We update both adk web run eval endpoint and adk eval cli to use the LocalService. The old method is marked as deprecated and will be removed in later PRs.
PiperOrigin-RevId: 785612708
Fixes#423
Related to #1670
- This avoids the `GeneratorExit` error thrown, which would crash OTel metric collection and cause `Failed to detach context` error.
- This also allows all function calls are processed when exit_loop is called together with other tools in the same LLmResponse.
A sample agent for testing:
```
from google.adk import Agent
from google.adk.agents.loop_agent import LoopAgent
from google.adk.tools.exit_loop_tool import exit_loop
worker_1 = Agent(
name='worker_1',
description='Worker 1',
instruction="""\
Just say job #1 is done.
If job #1 is said to be done. Call exit_loop tool.""",
tools=[exit_loop],
)
worker_2 = Agent(
name='worker_2',
description='Worker 2',
instruction="""\
Just say job #2 is done.
If job #2 is said to be done. Call exit_loop tool.""",
tools=[exit_loop],
)
work_agent = LoopAgent(
name='work_agent',
description='Do all work.',
sub_agents=[worker_1, worker_2],
max_iterations=5,
)
root_agent = Agent(
model='gemini-2.0-flash',
name='hello_world_agent',
description='hello world agent that can roll a check prime',
instruction="""Hand off works to sub agents.""",
sub_agents=[work_agent],
)
```
PiperOrigin-RevId: 785538101
Merge https://github.com/google/adk-python/pull/1195
## Summary
Updated the Toolbox Agent documentation to address a critical missing dependency that prevents the agent from running successfully.
## Changes Made
- **Added missing dependency**: Documented that `toolbox-core` must be installed via `pip install toolbox-core`
- **Improved documentation structure**: Added clear section numbering and better organization
- **Enhanced readability**: Fixed grammar, capitalization, and formatting throughout
- **Added Prerequisites section**: Set clear expectations before installation begins
- **Clarified optional steps**: Made it clearer when database creation can be skipped
## Problem Solved
The original documentation was missing a crucial step - installing the `toolbox-core` package. Without this dependency, users encounter an `ImportError: No module named 'toolbox-core'` when trying to use the `ToolboxToolset` class in ADK. This fix ensures users can successfully set up and run the agent without encountering import errors.
## Testing
- Verified the installation steps work correctly with the added dependency
- Confirmed the agent runs successfully after following the updated documentation
COPYBARA_INTEGRATE_REVIEW=https://github.com/google/adk-python/pull/1195 from designcomputer:patch-1 b90c71fe95aa09a3dca069e91f14791f557ab2e3
PiperOrigin-RevId: 785487495
Merge https://github.com/google/adk-python/pull/1130
This enables the use of the `model-optimizer-*` family of models in vertex, as per the [documentation](https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/vertex-ai-model-optimizer#using-vertex-ai-model-optimizer).
To use this, ensure your location is set to `global` and pass a model optimizer model to an agent:
```python
root_agent = Agent(
model="model-optimizer-exp-04-09",
name="fast_and_slow_agent",
instruction="Answer any question the user gives you - easy or hard.",
generate_content_config=types.GenerateContentConfig(
temperature=0.01,
model_selection_config=ModelSelectionConfig(
feature_selection_preference=FeatureSelectionPreference.BALANCED
# Options: PRIORITIZE_QUALITY, BALANCED, PRIORITIZE_COST
)
),
)
```
COPYBARA_INTEGRATE_REVIEW=https://github.com/google/adk-python/pull/1130 from calvingiles:feat-model-optimizer 1a76bfa22420edb07d83415dcea6dd0114084e8e
PiperOrigin-RevId: 784921913
Now the LangchainTool can wrap:
* Langchain StructuredTool (sync and async).
* Langchain @Tool (sync and async).
This enhance the flexibility for user and enables async functionalities.
PiperOrigin-RevId: 784728061
**Highlights:**
- **Callback Chaining:** Now supports a list of before and after callbacks, as in the non-live version. Callbacks are executed in order, stopping when one returns a response.
- **Unit Tests:** Added new unit tests for before and after callbacks, async and sync versions, callback chains and mixed callbacks.
- **Sample Agent:** Introduced a new example agent with multiple callbacks showing various behaviors: audit, security, validation, and enhancement. This provides practical usage examples.
PiperOrigin-RevId: 783884562
This change adds activity start and end signals to the LiveRequestQueue,
allowing clients to manually control the start and end of user input in
streaming sessions when automatic voice activity detection is disabled.
The LiveRequestQueue allows users to send messages to the model with the following semantics:
- `content`: sends turn-by-turn content.
- `blob`: sends a media blob for realtime streaming (e.g., audio).
- `activity_start`: indicates the beginning of an activity.
- `activity_end`: indicates the end of an activity.
- `close`: closes the connection.
GeminiLLMConnection has been updated to send the new activity signals to the backend.
This change is a necessary to support clients (e.g. voice assistants) that do not want to use automatic voice activity detection. In this case, the client will be responsible to send the `activity_start` signal when the user starts talking, and `activity_end` when the user finishes talking.
To test the change:
run_config = RunConfig(
realtime_input_config=types.RealtimeInputConfig(
automatic_activity_detection=types.AutomaticActivityDetection(
disabled=True,
),
)
)
import threading # Add this import
def thread_target():
# Define the async operations to run in the background.
async def background_task():
live_request_queue.send_activity_start()
# live_request_queue.send_content(
# content=types.Content(
# role='user',
# parts=[types.Part.from_text(text="hi, what's the time?")],
# )
# )
await asyncio.sleep(3)
live_request_queue.send_activity_end()
PiperOrigin-RevId: 783882447
1. credential service may be accessed by callbacks
2. plan to add load_credential and save_credential method in CallbackContext (see cl/782158513) given customer has requirement to access credential service themselves. (see https://github.com/google/adk-python/issues/1816)
It's backward compatible given CallbackContext is parent class of ToolContext
PiperOrigin-RevId: 783480378
This is to support model path like :
projects/265104255505/locations/us-central1/publishers/google/models/gemini-2.0-flash-001"
PiperOrigin-RevId: 783413351
This version of the EvalSetsManager is intended to support two main behaviors
1) The agent developer wants to bring in their own eval set file, which is usually the case with `adk eval` cli. Once their eval sets are uploaded into this version of the eval sets manager, the EvalSetManager could be handed over to the Eval system for running evals.
2) As a part of AgentEvaluator testing, we expect developers to supply Eval cases in json files. The in-memory version of the EvalSetsManager will help us run those test cases using LocalEvalService.
PiperOrigin-RevId: 783198788
When get session is being called on a session with more than 1 page of events (100+), the response of the subsequent listevents calls fails due to missing parsing logic. This change fixes the processing of the listevents responses.
PiperOrigin-RevId: 783166959
1) raise explicit error if the response event contains responses against multiple function call events
2) merge all function responses for the corresponding function call event
PiperOrigin-RevId: 782154577
According to a2a protocol task artifact is a different concept from adk artifact.
if a task is completed the final result should be in task artifact.
PiperOrigin-RevId: 782154265
This change:
- Introduces the LocalEvalService Class.
- Implements only the "perform_inference" method. Evaluate method will be implemented in the next CL.
- Adds required test coverage.
PiperOrigin-RevId: 781781954
This change integrates the plugin system with ADK. PluginManager is attached to the invocation context similar to session/artifact/memory.
It includes integrations with following ADK internal callbacks:
* App callbacks: Integrated in the BaseRunner class, in run_async and run_live
* On Message callbacks: Integrated in the BaseRunner class, triggers on run_async.
* Agent callbacks: Integrated in the BaseAgent class. Leveraging the existing *callback functions
* Model callbacks: Integrating in the base_llm_flow.
* Tool callbacks: Integrated in functions.py, wrapped around the code for agent tool_callbacks
Sample code to use plugins:
```python
# Add plugins to Runner
runner = Runner(
app_name="my-app",
agent=root_agent,
artifact_service=artifact_service,
session_service=session_service,
memory_service=memory_service,
plugins=[
MySamplePlugin(),
LoggingPlugin(),
],
)
```
PiperOrigin-RevId: 781746586
Plugin is a collection of callbacks (lifecycle hooks), including callbacks for Agent, Tools, Model and App. It helps developers to add customizations and change behaviors of their agent application easily
This is the first change that introduces only the base interfaces based on the doc.
plugins folder is generated with `create_symlink.sh plugins`
PiperOrigin-RevId: 781745044
Including basic fields in configs, the from_config() methods and the JSON schema. Other fields will be added in following PRs.
PiperOrigin-RevId: 781660569
Even though InMemoryMemoryService is intended only for testing and local development, we eliminate a potential source of bugs during prototyping by providing a thread-safe InMemoryMemoryService.
PiperOrigin-RevId: 781554006
- Allow run_async to break on partial events instead of raising ValueError
- Generate aggregated streaming content regardless of finish_reason
- Add error_code and error_message to final streaming responses if model response is interrupted.
PiperOrigin-RevId: 781377328
This commit includes a number of new tests for live streaming with function calls. These tests cover various scenarios:
- Single function calls
- Multiple function calls
- Parallel function calls
- Function calls with errors
- Synchronous function calls
- Simple streaming tools
- Video streaming tools
- Stopping a streaming tool
- Multiple streaming tools simultaneously
The tests use mock models and custom runners to simulate the interaction between the agent, model, and tools. They verify that function calls are correctly generated, executed, and that the expected data is returned.
PiperOrigin-RevId: 781318483
# Step 3: Check if the commit count is greater than 1
# This step uses the output from the previous step to decide whether to pass or fail.
- name:Check Commit Count
# This step only runs if the 'commit_count' output from the 'count_commits' step is greater than 1.
if:steps.count_commits.outputs.commit_count > 1
# If the condition is met, the workflow will exit with a failure status.
run:|
echo "This pull request has ${{ steps.count_commits.outputs.commit_count }} commits."
echo "Please squash them into a single commit before merging."
echo "You can use git rebase -i HEAD~N"
echo "...where N is the number of commits you want to squash together. The PR check conveniently tells you this number! For example, if the check says you have 3 commits, you would run: git rebase -i HEAD~3."
echo "Because you have rewritten the commit history, you must use the --force flag to update the pull request: git push --force"
exit 1
# Step 4: Success message
# This step runs if the commit count is not greater than 1 (i.e., it's 1).
- name:Success
if:steps.count_commits.outputs.commit_count <= 1
run:|
echo "This pull request has a single commit. Great job!"
@@ -10,6 +10,15 @@ The Agent Development Kit (ADK) is an open-source, code-first Python toolkit for
Please refer to [ADK Project Overview and Architecture](https://github.com/google/adk-python/blob/main/contributing/adk_project_overview_and_architecture.md) for details.
### ADK Live (Bidi-streaming)
- ADK live feature can be accessed from runner.run_live(...) and corresponding FAST api endpoint.
- ADK live feature is built on top of [Gemini Live API](https://cloud.google.com/vertex-ai/generative-ai/docs/live-api). We integrate Gemini Live API through [GenAI SDK](https://github.com/googleapis/python-genai).
- ADK live related configs are in [run_config.py](https://github.com/google/adk-python/blob/main/src/google/adk/agents/run_config.py).
- ADK live under multi-agent scenario: we convert the audio into text. This text will be passed to next agent as context.
- Most logics are in [base_llm_flow.py](https://github.com/google/adk-python/blob/main/src/google/adk/flows/llm_flows/base_llm_flow.py) and [gemini_llm_connection.py](https://github.com/google/adk-python/blob/main/src/google/adk/models/gemini_llm_connection.py).
- Tests are in [tests/unittests/streaming](https://github.com/google/adk-python/tree/main/tests/unittests/streaming).
## ADK: Style Guides
### Python Style Guide
@@ -200,4 +209,11 @@ The following changes are considered breaking and necessitate a MAJOR version
(e.g., the switch from snake_case to camelCase for all JSON payloads).
- Dependency Removal: Removing support for a previously integrated third-party
library or tool type.
library or tool type.
## Commit Message Format
- Please use [conventional commits](https://www.conventionalcommits.org/en/v1.0.0/)
format.
- If it's not a breaking change, please add #non-breaking tag. If it's a
* Add ability to send state change with message [3f9f773](https://github.com/google/adk-python/commit/3f9f773d9b5fcca343e32f76f6d5677b7cf4c327)
* [Eval] Support for persisting eval run results [bab3be2](https://github.com/google/adk-python/commit/bab3be2cf31dc9afd00bcce70103bdaa5460f1a3)
* Introduce [Plugin]: Plugin is simply a class that packages these individual callback functions together for a broader purpose[162228d](https://github.com/google/adk-python/commit/162228d208dca39550a75221030edf9876bf8e3a)
### Bug Fixes
* Create correct object for image and video content in litellm [bf7745f](https://github.com/google/adk-python/commit/bf7745f42811de3c9c80ec0998001ae50960dafc)
* Support project-based gemini model path for BuiltInCodeExecutor and all built-in tools [a5d6f1e](https://github.com/google/adk-python/commit/a5d6f1e52ee36d84f94693086f74e4ca2d0bed65)
* Add instruction in long running tool description to avoid being invoked again by model [62a6119](https://github.com/google/adk-python/commit/62a611956f8907e0580955adb23dfb6d7799bf4f)
* [A2A] Import A2A well known path from A2A sdk [a6716a5](https://github.com/google/adk-python/commit/a6716a55140f63834ae4e3507b38786da9fdbee2)
* Fix the long running function response event merge logic [134ec0d](https://github.com/google/adk-python/commit/134ec0d71e8de4cf9bcbe370c7e739e7ada123f3)
* [A2A] Return final task result in task artifact instead of status message [a8fcc1b](https://github.com/google/adk-python/commit/a8fcc1b8ab0d47eccf6612a6eb8be021bff5ed3a)
* Make InMemoryMemoryService thread-safe [10197db](https://github.com/google/adk-python/commit/10197db0d752defc5976d1f276c7b5405a94c75b)
### Improvements
* Improve partial event handling and streaming aggregation [584c8c6](https://github.com/google/adk-python/commit/584c8c6d91308e62285c94629f020f2746e88f6f)
### Documentation
* Update agent transfer related doc string and comments [b1fa383](https://github.com/google/adk-python/commit/b1fa383e739d923399b3a23ca10435c0fba3460b)
* Update doc string for GcsArtifactService [498ce90](https://github.com/google/adk-python/commit/498ce906dd9b323b6277bc8118e1bcc68c38c1b5)
This sample demonstrates how to use a **remote Agent-to-Agent (A2A) agent as the root agent** in the Agent Development Kit (ADK). This is a simplified approach where the main agent is actually a remote A2A service, also showcasing how to run remote agents using uvicorn command.
## Overview
The A2A Root sample consists of:
- **Root Agent** (`agent.py`): A remote A2A agent proxy as root agent that talks to a remote a2a agent running on a separate server
- **Remote Hello World Agent** (`remote_a2a/hello_world/agent.py`): The actual agent implementation that handles dice rolling and prime number checking running on remote server
## Architecture
```
┌─────────────────┐ ┌────────────────────┐
│ Root Agent │───▶│ Remote Hello │
│ (RemoteA2aAgent)│ │ World Agent │
│ (localhost:8000)│ │ (localhost:8001) │
└─────────────────┘ └────────────────────┘
```
## Key Features
### 1. **Remote A2A as Root Agent**
- The `root_agent` is a `RemoteA2aAgent` that connects to a remote A2A service
- Demonstrates how to use remote agents as the primary agent instead of local agents
- Shows the flexibility of the A2A architecture for distributed agent deployment
### 2. **Uvicorn Server Deployment**
- The remote agent is served using uvicorn, a lightweight ASGI server
- Demonstrates a simple way to deploy A2A agents without using the ADK CLI
- Shows how to expose A2A agents as standalone web services
### 3. **Agent Functionality**
- **Dice Rolling**: Can roll dice with configurable number of sides
- **Prime Number Checking**: Can check if numbers are prime
- **State Management**: Maintains roll history in tool context
- **Parallel Tool Execution**: Can use multiple tools in parallel
### 4. **Simple Deployment Pattern**
- Uses the `to_a2a()` utility to convert a standard ADK agent to an A2A service
- Minimal configuration required for remote agent deployment
elsef"{', '.join(str(num)fornuminprimes)} are prime numbers."
)
root_agent=Agent(
model='gemini-2.0-flash',
name='hello_world_agent',
description=(
'hello world agent that can roll a dice of 8 sides and check prime'
' numbers.'
),
instruction="""
You roll dice and answer questions about the outcome of the dice rolls.
You can roll dice of different sizes.
You can use multiple tools in parallel by calling functions in parallel(in one request and in one round).
It is ok to discuss previous dice roles, and comment on the dice rolls.
When you are asked to roll a die, you must call the roll_die tool with the number of sides. Be sure to pass in an integer. Do not pass in a string.
You should never roll a die on your own.
When checking prime numbers, call the check_prime tool with a list of integers. Be sure to pass in a list of integers. You should never pass in a string.
You should not check prime numbers before calling the tool.
When you are asked to roll a die and check prime numbers, you should always make the following two function calls:
1. You should first call the roll_die tool to get a roll. Wait for the function response before calling the check_prime tool.
2. After you get the function response from roll_die tool, you should call the check_prime tool with the roll_die result.
2.1 If user asks you to check primes based on previous rolls, make sure you include the previous rolls in the list.
3. When you respond, you must include the roll_die result from step 1.
You should always perform the previous 3 steps when asking for a roll and checking prime numbers.
You should not rely on the previous history on prime results.
The ADK Answering Agent is a Python-based agent designed to help answer questions in GitHub discussions for the `google/adk-python` repository. It uses a large language model to analyze open discussions, retrieve information from document store, generate response, and post a comment in the github discussion.
This agent can be operated in three distinct modes:
- An interactive mode for local use.
- A batch script mode for oncall use.
- A fully automated GitHub Actions workflow (TBD).
---
## Interactive Mode
This mode allows you to run the agent locally to review its recommendations in real-time before any changes are made to your repository's issues.
### Features
* **Web Interface**: The agent's interactive mode can be rendered in a web browser using the ADK's `adk web` command.
* **User Approval**: In interactive mode, the agent is instructed to ask for your confirmation before posting a comment to a GitHub issue.
* **Question & Answer**: You can ask ADK related questions, and the agent will provide answers based on its knowledge on ADK.
### Running in Interactive Mode
To run the agent in interactive mode, first set the required environment variables. Then, execute the following command in your terminal:
```bash
adk web
```
This will start a local server and provide a URL to access the agent's web interface in your browser.
---
## Batch Script Mode
The `answer_discussions.py` is created for ADK oncall team to batch process discussions.
### Features
* **Batch Process**: Taken either a number as the count of the recent discussions or a list of discussion numbers, the script will invoke the agent to answer all the specified discussions in one single run.
### Running in Interactive Mode
To run the agent in batch script mode, first set the required environment variables. Then, execute the following command in your terminal:
Or `python -m adk_answering_agent.answer_discussions --recent 10` to answer the 10 most recent updated discussions.
---
## GitHub Workflow Mode
The `main.py` is reserved for the Github Workflow. The detailed setup for the automatic workflow is TBD.
---
## Update the Knowledge Base
The `upload_docs_to_vertex_ai_search.py` is a script to upload ADK related docs to Vertex AI Search datastore to update the knowledge base. It can be executed with the following command in your terminal:
```bash
exportPYTHONPATH=contributing/samples # If not already exported
The following environment variables are required for the agent to connect to the necessary services.
*`GITHUB_TOKEN=YOUR_GITHUB_TOKEN`: **(Required)** A GitHub Personal Access Token with `issues:write` permissions. Needed for both interactive and workflow modes.
*`GOOGLE_GENAI_USE_VERTEXAI=TRUE`: **(Required)** Use Google Vertex AI for the authentication.
*`GOOGLE_CLOUD_PROJECT=YOUR_PROJECT_ID`: **(Required)** The Google Cloud project ID.
*`GOOGLE_CLOUD_LOCATION=LOCATION`: **(Required)** The Google Cloud region.
*`VERTEXAI_DATASTORE_ID=YOUR_DATASTORE_ID`: **(Required)** The full Vertex AI datastore ID for the document store (i.e. knowledge base), with the format of `projects/{project_number}/locations/{location}/collections/{collection}/dataStores/{datastore_id}`.
*`OWNER`: The GitHub organization or username that owns the repository (e.g., `google`). Needed for both modes.
*`REPO`: The name of the GitHub repository (e.g., `adk-python`). Needed for both modes.
*`INTERACTIVE`: Controls the agent's interaction mode. For the automated workflow, this is set to `0`. For interactive mode, it should be set to `1` or left unset.
The following environment variables are required to upload the docs to update the knowledge base.
*`GCS_BUCKET_NAME=YOUR_GCS_BUCKET_NAME`: **(Required)** The name of the GCS bucket to store the documents.
*`ADK_DOCS_ROOT_PATH=YOUR_ADK_DOCS_ROOT_PATH`: **(Required)** Path to the root of the downloaded adk-docs repo.
*`ADK_PYTHON_ROOT_PATH=YOUR_ADK_PYTHON_ROOT_PATH`: **(Required)** Path to the root of the downloaded adk-python repo.
For local execution in interactive mode, you can place these variables in a `.env` file in the project's root directory. For the GitHub workflow, they should be configured as repository secrets.
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Ankur Sharma
Xiang (Sean) Zhou
Yifan Wang
Max Thormé
Shangjie Chen
Google Team Member
Jacky Wu
Ariz Chang
Jianfeng Zeng
Liang Wu
Wei Sun (Jack)
Xuan Yang
Vasilii Novikov
Dana K. Williams
Calvin Giles
Hangfei Lin
Copybara-Service
seanzhou1023
Thiago Salvatore
Thiago Salvatore
Alejandro Cruzado-Ruiz
Zdravko Shehov
Frank Liu
Holt Skinner
Che Liu
Yanfei Chen
Bilawal Hameed
hironow
Marcello Maugeri
Mateusz Czubak
David Schmidt
Jeffrey Mak
kmak