sweet-search CLI
Local code search for Claude Code and Codex: six fast tools in place of grep and Read, with an index that updates itself.
- 120 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- `init` changes agent settings in `.claude/`: it replaces the main agent and subagents.
- A background process reads the working tree and downloads models to disk.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/sweet-searchDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- `init` changes agent settings in `.claude/`: it replaces the main agent and subagents.
- A background process reads the working tree and downloads models to disk.
This is third-party code. Review the repository files before installing.
What it does
A Rust-engine CLI that builds a local repository index and gives the agent six tools: `ss-search`, `ss-grep`, `ss-find`, `ss-semantic`, `ss-trace` and `ss-read`. A question like "where do we validate JWT tokens" goes through semantic search instead of file-by-file scanning. `sweet-search init` wires the tools into Claude Code, Codex or opencode, and `sweet-search index` builds the index. Models run on-device: no keys, no cloud, no Docker. A background reconcile process tracks the working tree, uncommitted edits included. By the author's benchmark on 200 tasks, agent runs cost 5–33% less with the same solved count.
Who it is for. Developers working in large repositories with Claude Code or Codex who want to spend fewer tokens on code search.
Good fit when
- An agent spends many requests scanning files in a large monorepo.
- You need to find code by the meaning of a question rather than an exact word.
- Source code must not be sent to a cloud index.
- One setup must be wired into Claude Code and Codex at once.
Not a fit when
- You need Windows without WSL2: there is no native Windows build yet.
- You do not want to change the main agent and subagents of Claude Code.
- The repository is small and plain `grep` is enough.
Example request
Find where this project validates JWT tokens and show the call chain up to the request handler.Limitations
Requires Node.js 22 or newer. macOS and Linux (x64, arm64) are first class, Windows via WSL2. Models take a few hundred MB on CPU, +1.2 GB with GPU and about 3.2 GB for CoreML on M3 and newer. By default `init` for Claude Code replaces the main agent and the `general-purpose` and `Plan` subagents; `--no-lean-harness` keeps the native ones. Savings figures come from the author's own tasks: verify on yours.
How to disable. Run `sweet-search uninstall` in the repository, or `sweet-search uninstall --all` to remove rules, agents, models and the CLI. Preview with `--dry-run`.
FAQ
Which agents are supported?
Claude Code by default, `--codex`, `--opencode`, plus `AGENTS.md`, `GEMINI.md` and a Cursor rule via `--agents`, `--gemini`, `--cursor`.
Where are models stored?
In `~/.cache/sweet-search/models/`, and they run on-device only.
How does the index stay fresh?
After `sweet-search index`, a background reconcile process tracks the working tree, uncommitted changes included.
How do I uninstall?
`sweet-search uninstall` for the current repo or `sweet-search uninstall --all` for everything.
Which platforms are supported?
macOS and Linux (x64 and arm64) are first class, Windows runs via WSL2. Other CPUs use a slower WASM fallback.
Discussion
No questions or reviews yet.
Related
CLI toolscodesearch searches code across several repositories with hybrid retrieval and runs fully offline, without a GPU or Docker.
A fast local indexed code search in Rust for AI assistants that plugs into Claude Code, OpenCode and Codex.
A Rust MCP server and CLI with a call graph and hybrid semantic code search for Claude Code and Codex
A Rust CLI that indexes code symbols locally so agents search by structure instead of grep and spend 40-50% fewer tokens.
A local semantic code index: the agent searches by meaning, then follows hits into definitions, callers and dependency paths
Local code search for agents: finds the files and fragments that matter and keeps context within a set token budget
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| sweet-search (this card) | no | 120 | ||
| Claude Context | yes | API key | 12,690 | indexes code through an external vector database; sweet-search runs fully local and provides six tools |
| grepai | no | 1,950 | semantic code search from the terminal; this one adds agent rules and a search-tuned system prompt |
README in short
The Competition section compares honestly with peers and names trade-offs. Reports include engine speed and full-corpus MRR measurements.