Shebe

Shebe is a local BM25 code-search utility for developers and AI coding agents working across repositories. It suits people who keep source code on their own machines or self-hosted systems and need fast textual lookup. The trade-off is straightforward: strong exact-term search and agent tooling, but no understanding of related terminology.

Overview

Shebe indexes repositories that have been cloned locally, then ranks full-text matches with BM25. It needs no embeddings, GPU, or cloud service, making it a sensible choice when source and search activity must remain on local infrastructure. The local-only workflow is also a boundary: remote repository URLs cannot be indexed directly.

Queries can use phrases, Boolean operators, or literal matching. The project's published benchmarks report about 2 ms query latency and indexing rates from 1,928 to 11,210 files per second. Those figures suit frequent navigation through large codebases, but textual speed does not provide conceptual coverage. A query for “login” will not find code that uses only “authenticate.”

Key features

• Ranked repository search: BM25 ranking helps surface the most relevant occurrences of known words and symbols. Phrase, Boolean, and literal modes provide useful control for precise lookups. Shebe has no regex search and no semantic similarity, so it is a poor fit for broad pattern matching or terminology that varies across a codebase.
• MCP tools for agents: Fourteen MCP tools cover searching, repository indexing, file reading, and reference finding. Coding agents get a focused interface to a local checkout rather than a cloud dependency, while developers can use the same indexed material for investigation.
• Editor integration: The installation material documents MCP setup and a Zed configuration, and directs users to the organization for VS Code and other editor extensions. That makes Shebe more practical in editor-centered workflows than a standalone command-line index.
• Code navigation: Navigation and reference-finding functions complement textual search. The project positions Shebe alongside structural tools such as Serena MCP, so teams can combine exact text retrieval with separate structural analysis instead of expecting one index to do both.
• Local deployment: Shebe is self-hosted and runs locally under the Apache-2.0 license. This reduces reliance on external processing for repositories, although users remain responsible for operating and maintaining the local installation.

Pricing

Shebe has one free plan at no cost. It provides local code search and is Apache-2.0 licensed, with no free trial because the product itself is free. There are no paid tiers, seat charges, or usage quotas stated for this plan. It is therefore the natural choice for individual developers, open-source projects, and teams able to run their own infrastructure; organizations wanting a hosted service or commercial support model should look elsewhere.

Platforms

Shebe is offered as an API and runs on Linux, macOS, Windows, and self-hosted deployments. Prebuilt binaries are listed for Linux x86_64 and macOS on both Apple Silicon and Intel. Linux ARM64 is planned rather than listed as a current binary target. Windows users must build from source, adding setup work compared with the supported prebuilt platforms.

Who it's for

Choose Shebe if you are a developer or coding-agent workflow owner searching large, locally cloned repositories and you value local processing, open licensing, and MCP access. It is particularly suitable for exact-term investigations, symbol lookups, and agent-assisted file and reference retrieval. It is not the right primary search layer for teams that need remote repository indexing, semantic discovery, regex-based searches, or a turnkey Windows installation.

Pros and cons

Pros

• Local by design: No cloud service, embeddings, or GPU are required, which fits privacy-conscious and self-hosted environments.
• Fast textual lookup: Published benchmarks cite 2 ms queries and indexing rates up to 11,210 files per second.
• Useful agent surface: Fourteen MCP tools combine search with indexing, file reading, and reference finding.
• Open and free: The Apache-2.0 license and 0.00 USD plan remove subscription and seat-cost barriers.

Cons

• No semantic matching: Different words for the same idea, such as “login” and “authenticate,” are not connected.
• Local clones required: Remote repository URLs are unsupported, so repositories must first be present on disk.
• Windows build requirement: Windows users need to compile from source rather than install a listed prebuilt binary.
• No regex search: Pattern-oriented searches need another tool.

Alternatives

Pick ast-grep instead when an extension-based, cross-platform option is preferable. Code Search is a stronger choice for unlimited users, find-and-replace across repositories, and a CLI/MCP server with SCIP code intelligence. Choose sqry for a locally run MIT-licensed open-source alternative. OpenGrok fits teams wanting an open-source, self-hosted web-oriented project. GNU GLOBAL is another free software option for Linux, macOS, and self-hosted environments. LXR Cross Referencer suits users seeking a self-installed web cross-referencer. Semantic Designs Source Code Search Engine is worth considering for its evaluation edition covering 10,000 lines across several languages.

Browse all code search tools for more options.

Verdict

Shebe is a strong recommendation for developers and AI agents that need fast, local, ranked text search over cloned repositories, especially when MCP integration and Apache-2.0 licensing matter. Look elsewhere if semantic discovery, remote indexing, regex, or a ready-made Windows install is central to your workflow.