Skip to content

codebase_readme.md

Latest commit

 

History

History
77 lines (46 loc) · 3.15 KB

codebase_readme.md

File metadata and controls

77 lines (46 loc) · 3.15 KB
Raw

CodeArts Code Index Comparison (Kafka 3.9.0, manual walkthrough)

Language: English ｜ 简体中文

Video (7 min 13 s, 1920×1200, ~104 MB, click to download): codebases_demo.mp4

What This Demo Shows

  • A controlled comparison: Ask the same question about Kafka 3.9.0 with the code index off and on.
  • Different retrieval paths: With the index off, an explore SubAgent reads project files; with it on, the run can use CodeSemanticSearch for semantic retrieval.
  • The observed time difference: In this walkthrough, the indexed run took 1 min 41 s, compared with 10 min 46 s without the index—about 1/6.4 as long.

1. Download a suitable project — in this case the open-source Apache Kafka repository

2. Open CodeArts Agent, select the Kafka path as the project, and check that the local index toggle is off

CodeArts Agent with the kafka-3.9.0 workspace open, Local Index off, Cloud Index showing No index

3. Open the right-hand sidebar, sign in, pick a model, and enter the prompt:

Where is the authentication module of this project? Analyze it.

Wait for the run to finish and check the result.

With the index off, the model spawns an explore SubAgent to read project files

As the screenshot shows, it starts a SubAgent to read the project's files.

Recorded run time: 10 min 46 s:

Answer with the index off; the UI shows Completed 10min 46s

4. Turn the index on and wait for it to be built

Cloud Index build at 100% in CodeArts Agent

The kafka-3.9.0 repo being parsed under Repository Index in the CodeArts console

5. Once the index is ready, ask again with the same prompt and watch the result

Where is the authentication module of this project? Analyze it.

Note: a plain question does not always trigger an index lookup. When that happens, adjust the prompt slightly:

Where is the authentication module of this project? Analyze it. Use CodeSemanticSearch first.

Normally you will see CodeSemanticSearch called at the very start of the thinking process, instead of an explore SubAgent reading project files:

CodeBase: CodeSemanticSearch called at the start of thinking (Thinking 22s), instead of an explore SubAgent reading project files

Wait for the run to finish and check the result.

Answer with the index on; the UI shows Completed 1min 41s and the answer cites CodeSemanticSearch

Results

Index Path actually taken Time
Off explore SubAgent reads project files one by one 10 min 46 s
On CodeSemanticSearch semantic retrieval at the start of thinking 1 min 41 s

Same question. With the index on, the run took about 1/6.4 of the time (646 s → 101 s).