feat: add streaming response support for ReActAgent and DeepAgent by vearne · Pull Request #7 · vearne/agentscope-go

vearne · 2026-05-20T03:14:59Z

Add ReplyStream() method to ReActAgent and DeepAgent for progressive response streaming
Implement thinkAndActStream() internal method using model.Stream() API
Support tool calling, interruption, and tracing in streaming mode
Add comprehensive unit tests with mock streaming model
Create 5 example programs demonstrating various streaming scenarios:
- agent_stream: basic streaming usage
- stream_with_tools: streaming with tool integration
- stream_comparison: performance comparison between streaming and non-streaming
- interactive_stream: interactive CLI chat with streaming
- deep_agent_stream: advanced DeepAgent features with streaming
Add complete documentation and usage guides

All tests pass with race detection enabled.

Summary by CodeRabbit

Release Notes

New Features
- Added streaming response capability to ReActAgent and DeepAgent for token-by-token incremental output
- Streaming now supports tool calls and mid-stream interruption handling
- Added interactive terminal-based chat example with real-time streaming responses
Documentation
- Added comprehensive streaming implementation guide with 5 usage scenarios
- Included performance comparisons, advanced configuration options, and FAQs

- Add ReplyStream() method to ReActAgent and DeepAgent for progressive response streaming - Implement thinkAndActStream() internal method using model.Stream() API - Support tool calling, interruption, and tracing in streaming mode - Add comprehensive unit tests with mock streaming model - Create 5 example programs demonstrating various streaming scenarios: * agent_stream: basic streaming usage * stream_with_tools: streaming with tool integration * stream_comparison: performance comparison between streaming and non-streaming * interactive_stream: interactive CLI chat with streaming * deep_agent_stream: advanced DeepAgent features with streaming - Add complete documentation and usage guides All tests pass with race detection enabled. Co-authored-by: Cursor <cursoragent@cursor.com>

coderabbitai · 2026-05-20T03:15:17Z

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 46125a1e-708a-4bdd-bae0-3e409b7c4111

📥 Commits

Reviewing files that changed from the base of the PR and between a93f30e and 232df3a.

📒 Files selected for processing (11)

STREAMING_IMPLEMENTATION_SUMMARY.md
examples/STREAMING_EXAMPLES.md
examples/agent_stream/main.go
examples/deep_agent_stream/main.go
examples/interactive_stream/main.go
examples/stream_comparison/main.go
examples/stream_with_tools/main.go
pkg/agent/deep.go
pkg/agent/deep_test.go
pkg/agent/react.go
pkg/agent/react_test.go

📝 Walkthrough

Walkthrough

AgentScope-Go streaming output implementation adds ReplyStream and thinkAndActStream methods to ReActAgent and DeepAgent, enabling real-time token-by-token response streaming over channels. Includes full test coverage, comprehensive documentation of the implementation and architecture, and five example programs demonstrating streaming across basic, tool-based, comparative, interactive, and advanced DeepAgent scenarios.

Changes

Core Streaming Implementation & Testing

Layer / File(s)	Summary
ReActAgent streaming entrypoint and iteration `pkg/agent/react.go`	`ReplyStream` runs agent loop in background and streams messages over a buffered channel; `thinkAndActStream` performs one iteration using `model.Stream`, forwarding partial messages and handling tool execution.
DeepAgent streaming entrypoint and iteration `pkg/agent/deep.go`	`ReplyStream` mirrors non-streaming Reply control flow but returns buffered channel; `thinkAndActStream` performs one iteration using `model.Stream`, streaming assistant content with optional tool approval and output offloading.
ReActAgent streaming tests `pkg/agent/react_test.go`	`mockStreamModel` enables streaming test harness; `TestReActAgent_StreamSimpleReply` validates incremental message emission and final content; `TestReActAgent_StreamWithToolUse` validates streaming through tool-use interaction.
DeepAgent streaming tests `pkg/agent/deep_test.go`	`TestDeepAgent_StreamSimpleReply` validates streamed messages and memory size; `TestDeepAgent_StreamWithToolUse` validates streaming through tool-use with correct message count.
Implementation documentation `STREAMING_IMPLEMENTATION_SUMMARY.md`	Comprehensive overview covering core capabilities, test coverage, example program catalog, usage patterns, performance characteristics, technical highlights, code quality metrics, and next steps.

Streaming Usage Examples & User Guide

Layer / File(s)	Summary
Streaming user guide and quick start `examples/STREAMING_EXAMPLES.md`	Introduction to streaming examples, setup/run instructions, streaming vs non-streaming control flow explanation, use case recommendations, performance comparison, advanced agent configuration options, FAQ, and related documentation links.
Basic and comparative streaming examples `examples/agent_stream/main.go`, `examples/stream_comparison/main.go`	`agent_stream` demonstrates streaming from both ReActAgent and DeepAgent; `stream_comparison` runs non-streaming vs streaming side-by-side with elapsed time measurement.
Streaming with tool calls `examples/stream_with_tools/main.go`	Demonstrates streaming agent replies with registered weather and time tools across three scenarios: single weather query, single time query, and combined multi-tool query.
Advanced DeepAgent streaming examples `examples/deep_agent_stream/main.go`	Runs three demonstrations: basic DeepAgent response, DeepAgent with calculate tool, and context-compression scenario using `TruncatingCompressor` with max context token limits.
Interactive terminal streaming `examples/interactive_stream/main.go`	Read-eval loop over stdin that streams assistant responses, supporting quit/exit and clear commands; includes per-session message counting and summary output.

Sequence Diagram

sequenceDiagram
  participant Client
  participant ReplyStream as ReplyStream<br/>(goroutine)
  participant thinkAndActStream
  participant Model
  participant Channel as Channel<br/>(buffered)
  Client->>ReplyStream: ReplyStream(ctx, msg)
  ReplyStream->>ReplyStream: setup memory & hooks
  ReplyStream->>ReplyStream: spawn background goroutine
  activate ReplyStream
  loop Agent Iterations
    ReplyStream->>thinkAndActStream: invoke
    activate thinkAndActStream
    thinkAndActStream->>Model: Stream()
    activate Model
    loop Streamed Chunks
      Model->>thinkAndActStream: ChatResponse chunk
      thinkAndActStream->>Channel: emit partial *message.Msg
    end
    Model->>thinkAndActStream: stream complete
    deactivate Model
    thinkAndActStream->>thinkAndActStream: persist final assistant msg
    alt Tool Use Detected
      thinkAndActStream->>thinkAndActStream: execute tools
      thinkAndActStream->>Channel: emit tool result msg
    end
    deactivate thinkAndActStream
  end
  ReplyStream->>Channel: close channel
  deactivate ReplyStream
  Client->>Channel: receive <-chan *message.Msg

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~25 minutes

Possibly related PRs

vearne/agentscope-go#3: DeepAgent streaming implementation directly extends the core DeepAgent loop, memory management, compression, and tool handling introduced in this PR.
vearne/agentscope-go#4: Streaming implementations add mid-stream interruption handling that builds on the Interrupt/HandleInterrupt functionality from this PR.

Poem

🐰 Token by token, the stream flows free,
No more waiting for a full reply, you see!
From ReAct to Deep, now both can stream,
With tools and interrupts—a developer's dream! ✨

✨ Finishing Touches

📝 Generate docstrings

Create stacked PR
Commit on current branch

🧪 Generate unit tests (beta)

Create PR with unit tests
Commit unit tests in branch cursor/streaming-output-support

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

vearne merged commit c7fd6c6 into main May 20, 2026
1 check passed

vearne deleted the cursor/streaming-output-support branch May 20, 2026 06:44

Navigation Menu

feat: add streaming response support for ReActAgent and DeepAgent - #7

feat: add streaming response support for ReActAgent and DeepAgent#7
vearne merged 1 commit into
mainfrom
cursor/streaming-output-support

vearne commented May 20, 2026 •

edited by coderabbitai Bot

Loading

Uh oh!

Uh oh!

coderabbitai Bot commented May 20, 2026 •

edited

Loading

Review failed

Walkthrough

Changes

Sequence Diagram

Estimated code review effort

Possibly related PRs

Poem

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

vearne commented May 20, 2026 • edited by coderabbitai Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary by CodeRabbit

Release Notes

Uh oh!

Uh oh!

coderabbitai Bot commented May 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Review failed

Walkthrough

Changes

Sequence Diagram

Estimated code review effort

Possibly related PRs

Poem

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

vearne commented May 20, 2026 •

edited by coderabbitai Bot

Loading

coderabbitai Bot commented May 20, 2026 •

edited

Loading