Show agent thinking and tool calls in the chat UI
維護者通常 1 天內回覆
還沒有人認領這個 Issue。
評估
- 難度
- 5/5
- 預估耗時
- 一週以上
- 新手友好度
- 22/100
- Issue 類型
- 功能
- 描述清晰度
- 需要釐清
- 活躍度
- 停滯
- 技術堆疊
- python, typescript
研究方向
Start with apps/app/src/components/tool-rendering.tsx and inspect the existing ToolReasoning and useDefaultRenderTool integration. Then read apps/agent/main.py to understand how agent events are emitted; the work is done when thinking and tool activity can be presented as a coherent, collapsible chat view with the proposed status details.
由索引模型根據 Issue 內容生成。
描述
Summary
Surface the agent's thinking process and tool call activity in the chat UI so users can see how the agent reasons, not just the final answer.
Current State
- Tool calls: The
ToolReasoningcomponent (apps/app/src/components/tool-rendering.tsx) already renders tool names, arguments, and status (executing → complete) viauseDefaultRenderTool. This is a good foundation but is minimal — it shows what tool was called, not why. - Thinking/reasoning: Not shown at all. The agent's intermediate thought process is invisible to the user. There is no UI for streaming thinking tokens or displaying step-by-step reasoning.
Proposal
1. Agent thinking display
Show the agent's reasoning as it works through a problem. This could be:
- Streaming thinking tokens — If using a model with extended thinking (e.g. Claude with
thinkingblocks), render these in a collapsible "Thinking…" section above the response. Show the thinking text streaming in real-time with a subtle visual treatment (muted text, indented, or in a distinct container). - Step-by-step reasoning — For multi-step agent tasks, show each reasoning step as the agent progresses (e.g., "Analyzing your request…", "Looking up existing todos…", "Organizing by priority…").
2. Enhanced tool call visualization
Build on the existing ToolReasoning component:
- Execution timeline — When multiple tools are called in sequence, show them as a vertical timeline so users see the agent's workflow.
- Tool results — Show a summary of what the tool returned (e.g., "Found 3 todos" or "Updated 2 items"), not just the arguments passed in.
- Duration — Show how long each tool call took.
- Grouped display — Collapse completed tool calls into a compact summary, expand only the active one.
3. Thinking + tools combined view
When the agent thinks, then calls a tool, then thinks again — show this as a coherent narrative:
🧠 Thinking: "The user wants to organize their todos by priority..."
🔧 get_todos → returned 5 todos
🧠 Thinking: "I'll group these into high/medium/low priority..."
🔧 manage_todos → updated 5 todos
✅ "I've organized your todos by priority!"
Implementation Notes
- CopilotKit's
useDefaultRenderToolalready provides the hook point for tool rendering - The agent backend (
apps/agent/main.py) would need to emit thinking/reasoning events — this could be done via LangGraph's streaming callbacks or by adding thinking state toAgentState - For models with native thinking (Claude extended thinking), CopilotKit may already surface thinking tokens — needs investigation
- The
ToolReasoningcomponent atapps/app/src/components/tool-rendering.tsxis the starting point for enhanced tool visualization
Design Considerations
- Thinking should be collapsible by default after completion — users who want transparency can expand it, but it shouldn't clutter the chat for those who don't
- Tool calls should feel like a progress indicator, not noise — compact when done, detailed when active
- The overall effect should make the agent feel more trustworthy and transparent without overwhelming the UI
- 主要語言
- TypeScript
- 星號
- 3.9k
- 分支
- 414
- 平均合併
- 4 小時
- 30 天內合併 PR
- 6
環境準備
- 沒有 Dockerfile 或 Docker Compose 檔案
- 沒有 Pull Request 範本
- 閱讀貢獻指南
從這裡開始
- 先讀完整個 Issue,再讀專案的貢獻指南。
- 在 Issue 下留言說明你要接手 —— 這能避免兩個人做同樣的事。
- Fork 儲存庫,在一個分支上完成修改。
- 送出 Pull Request,並在描述裡引用這個 Issue 編號。
CopilotKit/OpenIntelligentUI 的其他 Issue
-
Script dedup key truncation may cause collisions可能已有人在做 @saschabuehrle 於 216 天前認領。 未關閉bug
難度 2/5 1-3 小時 新手友好度 72/100
CopilotKit/OpenIntelligentUI#37 ·
維護者通常 1 天內回覆
-
enhancement
難度 2/5 1-3 小時 新手友好度 72/100
CopilotKit/OpenIntelligentUI#29 ·
維護者通常 1 天內回覆
-
Self-host with Docker Compose and optional server-held keys可能已有人在做 @kannan983 於 16 天前認領。 未關閉
難度 4/5 3-5 天 新手友好度 18/100
CopilotKit/OpenIntelligentUI#98 ·
維護者通常 1 天內回覆
-
Support OpenAI-compatible gateways (OpenRouter) via vendor/model names可能已有人在做 @kannan983 於 16 天前認領。 未關閉
難度 2/5 1-3 小時 新手友好度 25/100
CopilotKit/OpenIntelligentUI#97 ·
維護者通常 1 天內回覆
-
UI未關閉
難度 5/5 一週以上 新手友好度 8/100
CopilotKit/OpenIntelligentUI#96 ·
維護者通常 1 天內回覆
查看 CopilotKit/OpenIntelligentUI 的全部 Issue
相似的 Issue
-
難度 2/5 1-3 小時 新手友好度 72/100
VictoriaMetrics/victorialogs-datasource#747 ·
維護者通常 1 天內回覆
-
難度 2/5 1-3 小時 新手友好度 79/100
維護者通常 1 天內回覆
-
bug from-studio
難度 2/5 1-3 小時 新手友好度 62/100
esengine/DeepSeek-Reasonix#14608 ·
維護者通常 1 天內回覆
-
triage
難度 2/5 1-3 小時 新手友好度 75/100
getarcaneapp/arcane#4590 ·
維護者通常 1 天內回覆
-
難度 2/5 1-3 小時 新手友好度 77/100
Clouditera/VulnHunter#57 ·