semsearch

semsearch command [options]

semsearch is the Typer-based administration CLI for Semsearch, an embedding-focused indexing and search engine aimed at aggregating and promoting indie blogs. The stack pairs a FastAPI web frontend with PostgreSQL + pgvector storage. The CLI handles schema setup, site registry, background ingestion, and operational status; the HTTP app serves search (see uvicorn semsearch.web.app:app).Configuration comes from environment / .env (see .env.example): database URL, embedding API endpoint and key, model name, and embedding dimension. Chunking and embedding settings are shared between CLI and web. Changing chunking algorithm, embedding dimension, or model requires a full re-index.Typical local layout: Postgres via docker compose, Python tools via uv run semsearch …. Production runs the same binary inside the app container (docker compose exec app … semsearch …).

init-db

Create database schema for the configured embedding model and dimension. Prints the model name, vector size, and redacted DSN when ready.

Register or update a feed-backed site. --feed is mandatory in practice (auto-discovery or explicit URL). --sitemap defaults to auto.

List configured sites.

Remove a site from the index configuration.

Long-running process that polls feeds, fetches pages, chunks text, embeds, and writes to pgvector. Uses an advisory lock so only one daemon runs at a time.

Report index statistics and list failed ingestion jobs.

Project status is proof-of-concept toward production. Requires a modern Python (upstream pins >=3.14), a running Postgres with pgvector, and a working embedding API key before indexing. Feed URLs are effectively required when adding sites. Not a general-purpose desktop search tool—it is the operator CLI for this specific blog-search service.

Semsearch is a small open-source project exploring agentic RAG-style embedding pipelines for independent blogs, with admin tooling under the semsearch console script.

curl(1), docker(1), uv(1)

Source code · Homepage