Metadata-Version: 2.1
Name: semfind
Version: 0.1.3
Summary: Semantic grep for the terminal — search files by meaning, not pattern
Author: PaperBoardOfficial
License: MIT
Project-URL: Homepage, https://github.com/PaperBoardOfficial/semfind
Keywords: semantic-search,grep,embeddings,cli
Classifier: Development Status :: 3 - Alpha
Classifier: Environment :: Console
Classifier: Intended Audience :: Developers
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.9
Classifier: Programming Language :: Python :: 3.10
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Topic :: Text Processing :: General
Classifier: Topic :: Utilities
Requires-Python: >=3.9
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: fastembed>=0.7.0
Requires-Dist: faiss-cpu>=1.7.0
Requires-Dist: numpy>=1.24.0

# semfind

Semantic grep for the terminal. Search files by meaning, not pattern.

`grep` finds exact text matches. `semfind` finds lines that mean the same thing. Search your logs, notes, docs, or any text file using natural language — no regex needed.

## Why semfind?

Traditional grep fails when you don't know the exact wording. If your log says "container build failed due to missing environment variables" but you search for "deployment issue", grep finds nothing. semfind finds it instantly because it understands meaning.

**Built for AI agents.** Tools like [OpenClaw](https://github.com/openclaw/openclaw) and other AI agents need lightweight semantic search over local files — searching memory, history, and context without spinning up a full vector database. semfind is a single CLI command with auto-caching that agents can call directly from the shell.

**Also great for humans.** Search your markdown notes, project logs, documentation, or any text files by what you mean, not what you remember typing.

### Key features

- **No API keys** — runs 100% locally using [fastembed](https://github.com/qdrant/fastembed) (BAAI/bge-small-en-v1.5) + FAISS
- **Auto-caching** — indexes files on first search, caches embeddings, auto-invalidates when files change
- **Fast** — ~2s cold start, 14ms cached queries, 252MB RAM
- **Grep-like output** — colored results with file, line number, and similarity score
- **Zero config** — just `pip install semfind` and go

## Install

```bash
pip install semfind
```

## Usage

```bash
# Search a file
semfind "deployment issue" logs.md

# Search multiple files, top 3 results
semfind "permission error" memory/*.md -k 3

# Show 2 lines of context around each match
semfind "database migration" notes.md -n 2

# Force re-index (ignore cache)
semfind "query" file.md --reindex

# Set minimum similarity threshold
semfind "auth bug" *.md -m 0.5
```

### Output

```
memory/HISTORY.md:9: [2026-01-15 10:30] DEPLOYMENT: Fixed docker build...  (0.796)
memory/HISTORY.md:3: [2026-01-17 09:15] FILE_PERMS: Agent couldn't...     (0.689)
```

## How it works

1. On first search, each file's non-empty lines are embedded and cached in `~/.cache/semfind/`
2. Cache is keyed by file content hash — changes auto-invalidate
3. Your query is embedded and compared via FAISS inner-product search
4. Results are printed grep-style with similarity scores

## Options

| Flag | Description | Default |
|------|-------------|---------|
| `-k, --top-k` | Number of results | 5 |
| `-n, --context` | Context lines before/after | 0 |
| `-m, --max-distance` | Minimum similarity score | none |
| `--reindex` | Force re-embed | false |
| `--model` | Embedding model | BAAI/bge-small-en-v1.5 |
| `--no-cache` | Skip cache | false |
| `--version` | Print version | |

## For OpenClaw and AI agents

semfind is built as a CLI tool specifically so that AI agents like [OpenClaw](https://github.com/openclaw/openclaw) can install and use it on demand.

### When to use semfind

Use `grep` first — it's instant and has zero overhead. But when grep fails because you don't know the exact wording, install semfind and search by meaning instead.

```bash
# grep found nothing useful
grep "deployment issue" memory/*.md  # no results

# semfind understands meaning
semfind "deployment issue" memory/*.md -k 3
# memory/HISTORY.md:9: [2026-01-15] Fixed docker build with missing env vars  (0.796)
```

### Why a CLI tool instead of a built-in agent tool?

A built-in tool would load the embedding model into memory on every agent session — even when semantic search isn't needed. That's ~250MB of RAM wasted on most runs.

As a CLI tool, semfind only loads the model when you actually call it. The process starts, runs the search, and exits — freeing all memory immediately. Agents that never need semantic search pay zero cost.

### Resource usage

| | First run | Cached runs |
|---|---|---|
| **Model download** | ~65MB download (~10-30s depending on connection) | Skipped (cached in `/tmp/fastembed_cache/`) |
| **Model disk storage** | ~65MB | Same |
| **RAM while running** | ~250MB (model + embeddings) | ~250MB |
| **RAM after exit** | 0 (process ends) | 0 |
| **Query latency** | ~2s (model load + embed + search) | ~14ms (embed + search) |
| **Embedding cache** | Written to `~/.cache/semfind/` | Read from cache, auto-invalidates on file change |

### Example agent use cases

- **Memory search** — searching history/memory files for relevant past context
- **Document retrieval** — finding relevant docs before answering user questions
- **Log analysis** — searching logs by describing the problem rather than knowing exact error strings

```bash
# Agent searching its memory
semfind "user asked about authentication" memory/*.md -k 3

# Searching project docs for context
semfind "how to configure database" docs/*.md -k 5

# Finding relevant logs without knowing exact error strings
semfind "something went wrong with file permissions" logs/*.md -k 3
```

## License

MIT
