Search.sh
Skill.mdAPICLIDashboard
Search Engine for AI Agents

Structured search
for machines

API-first search engine that returns structured JSON with answers, citations, confidence scores, and sources. Built for AI agents, pipelines, and automation.

For AI Agents

Point your agent to the skill file for full API docs, streaming examples, and usage patterns:

/skill.md|/llms.txt
REST API

One endpoint,
structured results

POST a query, get back JSON with an AI-synthesized answer, ranked sources, citations with source attribution, confidence scores, and related queries. Supports fast and deep search modes, streaming via SSE, and page content extraction.

POST /searchSearch with AI answer
GET /search?q=Quick search via GET
POST /extractExtract page content
GET /healthHealth check
bash
curl -X POST http://localhost:3100/search \
  -H "Content-Type: application/json" \
  -d '{
    "query": "best vector databases",
    "mode": "deep",
    "extract_content": true
  }'
json
{
  "query": "best vector databases",
  "answer": "The top vector databases...",
  "sources": [{ "title": "...", "url": "..." }],
  "citations": [{ "claim": "...", "source_url": "..." }],
  "confidence": 0.87,
  "related_queries": ["..."],
  "metadata": {
    "model": "google/gemini-3.1-flash-lite-preview",
    "duration_ms": 4320,
    "mode": "deep"
  }
}
Capabilities
Fast Search

Single query, AI answer with sources and citations. Optimized for speed — typically under 5 seconds.

Deep Search

Generates sub-queries, searches in parallel, extracts page content, cross-references sources.

Streaming

Server-sent events for real-time results. Stream sources, answer tokens, citations as they arrive.

Page Extraction

Extract clean content from any URL. Removes noise, returns structured text for processing.

Citations

Every claim attributed to a source. Confidence scoring based on source quality and agreement.

CLI Tool

Full-featured command line interface. Pipe JSON output to other tools. Built for automation.

For AI Agents

Discover via
skill.md

AI agents discover Search.sh through standard skill files. Point your agent to our skill.md for complete API documentation, streaming examples, code snippets in Python and JavaScript, and usage patterns for chaining searches.

Read skill.mdllms.txt
markdown
# Add to your agent's tools/skills:
https://search.sh/skill.md

# Or fetch it:
curl -s https://search.sh/skill.md

# Install locally:
mkdir -p ~/.agent/skills/search-sh
curl -s https://search.sh/skill.md \
  > ~/.agent/skills/search-sh/SKILL.md
python
import requests, json

res = requests.post(
    "https://search.sh/api/search",
    json={
        "query": "best practices for RAG",
        "mode": "deep",
        "stream": True,
    },
    stream=True,
)

for line in res.iter_lines():
    chunk = json.loads(
        line.decode().removeprefix("data: ")
    )
    if chunk["type"] == "answer_delta":
        print(chunk["data"], end="")
Command Line

Search from your terminal

bash
# Install
npm install -g @rogaai/search-sh

# Fast search
search-sh search "what is kubernetes"

# Deep research
search-sh search "best databases 2026" --mode deep

# JSON output (pipe to jq, other tools)
search-sh search "node.js frameworks" --json

# Sources only, no AI
search-sh search "react vs vue" --sources-only

# Extract page content
search-sh extract https://example.com