Sourcegraph, Greptile, Zoekt: What Is Left of AI Code Search in 2026

说明与用途

使用 Zoekt 的轻量级可复现工作流程，可快速索引大型仓库并执行快速正则查询。若需更深层次的跨仓库洞察，可集成 Sourcegraph Deep Search 或免费 Sourcebot 等付费代理。

问题

Engineers need to locate code across multiple repositories and services, but IDE assistants only see the current checkout, limiting scope for multi-service bugs.

结果

A lightweight, reproducible workflow using Zoekt allows teams to quickly index large repos and answer regex queries locally. If deeper cross‑repo context is needed, paid agents like Sourcegraph Deep Search or free self‑hosted Sourcebot can be integrated into coding agents.

工作流

• 1. Clone the target repository (e.g., kubernetes/kubernetes) shallowly. 2. Install Zoekt binaries. 3. Run zoekt-git-index to build the index. 4. Query the index with a regex and file filter. 5. Evaluate results: if regex answers the question, no paid AI layer is needed. 6. If deeper behavior tracing is required, use Sourcegraph Deep Search or Ask Sourcebot. 7. Wire the chosen MCP server into your coding agent (Claude Code, Cursor, VS Code) for future queries.

证据

I indexed a shallow clone of kubernetes/kubernetes, 31,300 files, with Zoekt on a laptop. 6.03 seconds to index, 193ms for a filtered regex query across the whole thing.

来源

Sourcegraph’s 2026 pitch is Deep Search, an agent that investigates across every repo in your org, follows code and its commit history, and cites the files and diffs behind each answer.

来源