refactor: Consolidate the logic of the agent and remove redundancy#93
Merged
refactor: Consolidate the logic of the agent and remove redundancy#93
Conversation
I think I should convert this to a more event-driven based architecture. It might be a breaking change but it would be clearer and easier to extend. - Updated the version of the inference-gateway/adk module from v0.11.2 to v0.13.1 in multiple go.mod and go.sum files across examples and server directories. - Refactored the AgentResponse struct to remove AdditionalMessages and directly append messages to task history. - Modified the Run method in OpenAICompatibleAgentImpl to handle task history updates more efficiently. - Streamlined input_required tool handling in agent_streamable.go and agent_toolbox.go, ensuring proper message flow and task state management. - Updated tests to reflect changes in the agent response structure and behavior. Signed-off-by: Eden Reich <eden.reich@gmail.com>
…reaming tasks Signed-off-by: Eden Reich <eden.reich@gmail.com>
Signed-off-by: Eden Reich <eden.reich@gmail.com>
…ved clarity Signed-off-by: Eden Reich <eden.reich@gmail.com>
…handler - Updated the AIStreamingTaskHandler to utilize RunWithStream for handling AI responses in a streaming manner. - Accumulated response parts from streaming events and created a final response message. - Enhanced task status management to reflect completion and history updates. refactor: Improve Dockerfile build processes for client and server examples - Modified Dockerfiles for ai-powered, default-handlers, artifacts-filesystem, and minimal examples to streamline build commands and paths. - Ensured proper binary naming and reduced build context complexity. chore: Update go.mod and go.sum files across examples - Replaced the adk module version to v0.13.1 in multiple go.mod files for consistency. - Cleaned up go.sum files by removing outdated entries. fix: Adjust Docker Compose configurations for client and server services - Updated Docker Compose files to correctly reference Dockerfile paths and build contexts for client and server services in various examples. Signed-off-by: Eden Reich <eden.reich@gmail.com>
Contributor
|
Thank you for adding the much needed tests!!! I'll probably just have to re-do most of my stuff (https://github.com/Joshswooft/adk/tree/support-callbacks) once this is all merged as sorting out all the merge conflicts and new examples (again) is a lot of effort. |
Contributor
Author
|
@Joshswooft I see, I had to refactor this, sorry for the conflicts with the files, I can help you resolve these if you push the changes. I think now it is easier to add the callbacks functionality, when there are clear events in the system. I've removed one of the method from the OpenAI-compatible agent (Run Method) which was redundant. Also running some tests it feels more performant right now with SSE internally and we lose less data due to connectivity. Previously the Run was adding the messages to the storage at the end of all agent iterations. Looking forward to see your changes 👍 |
…e build process and add agent arguments Signed-off-by: Eden Reich <eden.reich@gmail.com>
… instead of remote version Signed-off-by: Eden Reich <eden.reich@gmail.com>
…le for a client and server Reason - simply easier to maintain and creating more examples as the project grows ensuring consistency. - Created new Dockerfile.client and Dockerfile.server in the examples directory to standardize the build process for client and server applications. - Removed individual Dockerfiles from various example directories, consolidating them into the new standardized Dockerfiles. - Updated docker-compose.yaml files across multiple examples to reference the new Dockerfile locations and include necessary build arguments (LD_FLAGS, EXAMPLE_PATH, COMPONENT). - Ensured that build-time metadata is correctly passed to the Go build process for both client and server applications. Signed-off-by: Eden Reich <eden.reich@gmail.com>
… for client and server containers Signed-off-by: Eden Reich <eden.reich@gmail.com>
…th appropriate configurations Signed-off-by: Eden Reich <eden.reich@gmail.com>
- Added markdownlint-cli2 package to the manifest for linting markdown files. - Created a markdownlint configuration file to enforce line length and disable specific rules. - Introduced a new task in Taskfile.yml to run markdownlint on example markdown files. - Updated multiple example README files to improve formatting, add troubleshooting sections, and enhance clarity with a Table of Contents. - Included troubleshooting instructions for using the A2A Debugger in various example READMEs. Signed-off-by: Eden Reich <eden.reich@gmail.com>
Signed-off-by: Eden Reich <eden.reich@gmail.com>
Built-in is a bad wording, because those tools are being registered (not a built-in tools that I'm planning for future versions) - built-in tools would be tools that are in the default toolbox. Signed-off-by: Eden Reich <eden.reich@gmail.com>
Signed-off-by: Eden Reich <eden.reich@gmail.com>
…er in docker-compose.yaml Signed-off-by: Eden Reich <eden.reich@gmail.com>
…adiness Also expose the inference gateway locally in case the person wants to run the example locally without docker. Signed-off-by: Eden Reich <eden.reich@gmail.com>
…ent waits for server readiness Signed-off-by: Eden Reich <eden.reich@gmail.com>
…e.yaml Signed-off-by: Eden Reich <eden.reich@gmail.com>
Signed-off-by: Eden Reich <eden.reich@gmail.com>
… task listing Signed-off-by: Eden Reich <eden.reich@gmail.com>
Signed-off-by: Eden Reich <eden.reich@gmail.com>
Signed-off-by: Eden Reich <eden.reich@gmail.com>
… docker-compose files Signed-off-by: Eden Reich <eden.reich@gmail.com>
…ng and add TaskID and ContextID to messages in task handler Signed-off-by: Eden Reich <eden.reich@gmail.com>
Signed-off-by: Eden Reich <eden.reich@gmail.com>
…rver Signed-off-by: Eden Reich <eden.reich@gmail.com>
…rmissions Signed-off-by: Eden Reich <eden.reich@gmail.com>
Signed-off-by: Eden Reich <eden.reich@gmail.com>
…using build-time variables instead Signed-off-by: Eden Reich <eden.reich@gmail.com>
…racking Signed-off-by: Eden Reich <eden.reich@gmail.com>
Contributor
Author
|
I found out a huge flaw and inconsistency with the events, the task is streaming each delta to the client, which is not correct according to the protocol there is a TaskStatusUpdateEvent which should only emit when task status as changed - way less events. Also I don't like the current event system it's not consistent enough some events are used with custom event names, I will transition all the events to be CloudEvents so they are more compatible and easier to consume. I'll probably also introduce another event to see deltas in case some clients are interested to also see the deltas (although according to the protocol those are internal to the agent, those deltas could be useful for client applications that want to show some visual effects of the internal work of the agents). TODOs
|
- Removed the extractDeltaText and extractTextFromParts functions from the client. - Updated the client to directly handle TaskStatusUpdateEvent and extract text from TextPart. - Modified the server's MockAgent to send initial task status events and stream responses as CloudEvents. - Changed the StreamableTaskHandler interface to return CloudEvents instead of custom StreamEvent types. - Updated task handler to process CloudEvents directly, simplifying event handling logic. - Adjusted tests to reflect changes in message structure and event types. - Added new event type for task status changes to improve clarity in event handling. Signed-off-by: Eden Reich <eden.reich@gmail.com>
…on events Signed-off-by: Eden Reich <eden.reich@gmail.com>
- Added go-envconfig dependency to multiple examples for improved configuration handling. - Refactored main functions in various clients and servers to load configurations from environment variables. - Enhanced logging to provide better context and information based on the environment. - Updated health check and task submission logic to utilize the new configuration structure. - Improved code readability and maintainability by consolidating configuration and logging setup. Signed-off-by: Eden Reich <eden.reich@gmail.com>
…D, and TaskID Signed-off-by: Eden Reich <eden.reich@gmail.com>
Signed-off-by: Eden Reich <eden.reich@gmail.com>
Signed-off-by: Eden Reich <eden.reich@gmail.com>
…mproved clarity and type safety Signed-off-by: Eden Reich <eden.reich@gmail.com>
ig-semantic-release-bot bot
added a commit
that referenced
this pull request
Oct 11, 2025
## [0.14.0](v0.13.1...v0.14.0) (2025-10-11) ### ✨ Features * Add the callbacks ready to be used by the agent ([#94](#94)) ([84632f3](84632f3)) * **types:** Refactor Part deserialization to use concrete types ([#104](#104)) ([22d192e](22d192e)), closes [#102](#102) ### ♻️ Improvements * Consolidate the logic of the agent and remove redundancy ([#93](#93)) ([54e5e71](54e5e71)) * **docker-compose:** Remove port mappings for server service to avoid confusion ([3df7b08](3df7b08)) * **server:** Remove handler duplication between A2AServerImpl and DefaultA2AProtocolHandler ([#101](#101)) ([75b9c20](75b9c20)) ### 🐛 Bug Fixes * Improve validation logic for task handler configuration ([#98](#98)) ([34d6354](34d6354)) ### 👷 CI * Add Prettier setup and formatting steps for Go and markdown files ([#105](#105)) ([f2a1994](f2a1994)) ### 📚 Documentation * **examples:** Add input-required flow examples ([#96](#96)) ([17764f3](17764f3)) ### 🔧 Miscellaneous * **deps:** Update claude-code to 2.0.8 and install gh cli version 2.81.0 in flox environment ([4d6e41a](4d6e41a))
Contributor
|
🎉 This PR is included in version 0.14.0 🎉 The release is available on GitHub release Your semantic-release bot 📦🚀 |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Consolidate the logic of the agent and remove redundant code.
I'm deprecating the Run function in favor of using exclusively the RunWithStream to allow a more event-driven architecture.
The Run method is simple but it's not resilient when for example a connection drops or times-out - there is a lost of responses, better would be to stick to SSE in the backend, the A2A client that consumes the message can still decides whether they would like to receive SSE or Async Polling for task's updates.
The events are all cloud events for better standardisation and it's easier to add more events as pleased.
Also improve the examples, bump most examples to version 0.13.1 and use some examples with direct replace directive for using the latest features.
For those who used Run() in their custom tasks handler this method is going to be deleted so please always use pinned versions or modify the code to adopt the new event-driven architecture.
Fixed a non-critical issue, where context id and task id optional fields on the messages were not populated and were remained null.
TODOs
Regression tests - check that all examples still works as intended
Cleanup the input required logic
Add a2a-debugger to all examples for better visibility