Seek

Terminal-based semantic code search across all your local repos. CocoIndex handles incremental re-indexing automatically on every file save (no cron jobs).

Everything runs locally: no API keys, no cloud services, no cost.

---

How it works

1. Watch — CocoIndex monitors SEEK_PROJECTS_PATH for changes to source files
2. Chunk — changed files are split into overlapping 500-char chunks
3. Embed — each chunk is embedded with all-MiniLM-L6-v2 (runs on CPU, no GPU needed)
4. Store — embeddings land in PostgreSQL with pgvector
5. Search — your query is embedded with the same model, then nearest-neighbour search returns the top results

---

Setup

1. Install PostgreSQL + pgvector

Download PostgreSQL from https://www.postgresql.org/download/

After install, create the database and enable the extension:

CREATE DATABASE seek;
\c seek
CREATE EXTENSION vector;

2. Install Python dependencies

pip install -r requirements.txt

3. Configure the environment

cp .env.example .env
# Open .env and fill in your Postgres password

The only required variable is COCOINDEX_DATABASE_URL.

4. Index

cocoindex update index.py

This creates the schema and indexes everything under SEEK_PROJECTS_PATH (default C:/Projects).

5. Live watch mode

cocoindex update index.py -L

Keep this running in a terminal. Seek re-indexes any changed file within a few seconds of saving.

6. Search

# Interactive REPL
python search.py

# One-shot
python search.py "function that reads a file"

---

Search options

python search.py [query] [options]

positional:
  query                 natural-language search (omit for interactive mode)

options:
  -k N, --top-k N       number of results to show (default: 5)
  --ext py,go,ts        filter by file extension(s), comma-separated
  --path SUBSTR         only show results whose path contains SUBSTR
  --json                print results as JSON (pipe-friendly)

Examples

python search.py "authentication middleware" --top-k 10
python search.py "database connection pool" --ext py,go
python search.py "error handling" --path myrepo/
python search.py "jwt decode" --json | jq '.[].filename'

---

Configuration

All settings are read from .env. Defaults are shown in .env.example.

Variable	Default	Description
COCOINDEX_DATABASE_URL	(required)	PostgreSQL connection string
SEEK_PROJECTS_PATH	C:/Projects	Root directory to index
SEEK_EMBED_MODEL	sentence-transformers/all-MiniLM-L6-v2	Embedding model
SEEK_CHUNK_SIZE	500	Chunk size in characters
SEEK_CHUNK_OVERLAP	100	Overlap between chunks
SEEK_TOP_K	5	Default number of search results

If you change SEEK_EMBED_MODEL after indexing, drop the seekindex__code_embeddings table and re-run cocoindex update index.py to re-embed with the new model.

---

Running tests

pytest tests/ -v

---

File structure

Seek/
├── index.py          # CocoIndex flow — chunking, embedding, storage
├── search.py         # CLI search tool
├── config.py         # Settings (reads from .env)
├── requirements.txt  # Python dependencies
├── .env.example      # Config template
└── tests/
    └── test_search.py

---

Troubleshooting

COCOINDEX_DATABASE_URL is not set Copy .env.example to .env and fill in your Postgres password.

Could not connect to database Make sure PostgreSQL is running.

No results after indexing Check that SEEK_PROJECTS_PATH points to where your code lives.

Slow first search The sentence-transformers model is downloaded on first run (~90 MB). Subsequent runs load from cache.