# SailorSearch

> A search API designed for agents, RAG pipelines, and deep research — it returns clean markdown with enrichment, token compression, and source-aware outputs, so agents get dense, attributable context instead of raw HTML.

- Canonical: https://hasib.tech/work/sailor-search
- Role: Backend Engineer
- Year: 2025—26
- Category: AI Search
- Stack: FastAPI, PostgreSQL, Redis, Celery, RAG
- Output: Clean, source-aware markdown
- Designed for: Agents & deep research
- sailorsearch.dev: https://sailorsearch.dev

SailorSearch is a search API built for a new kind of consumer: not a human
scanning links, but an agent assembling context under a token budget. That
single constraint shapes everything about how it returns results.

## Search as an agent interface

I designed the API around three ideas. **Clean markdown** — pages arrive as
structured, readable markdown with enrichment instead of boilerplate-laden
HTML. **Token compression** — results are condensed toward what an agent
actually needs, because every wasted token in a context window is paid for in
quality and cost downstream. **Source-aware outputs** — every claim stays
attributable to where it came from, which is what makes the result safe to
cite in RAG and deep-research workloads.

## Built to scale

SailorSearch runs on FastAPI with PostgreSQL and Redis, using Celery for the
crawl, extract, and compress stages so they scale independently of the query
path — fast responses up front, heavy work in the background.