Parasail × You.com Web Search

▾

🌐 You.com Web Search Powers Open-Weight Models

Open-weight models know the world up to their training cutoff · You.com Web Search fills the gap with live web intel

Models like GPT-4 and Claude are proprietary; you access them only through the vendor's API. Open-weight models (GLM, Kimi, MiniMax, DeepSeek, Qwen) have publicly released weights, so inference providers like Parasail can host and serve them through an OpenAI-compatible API, with full model choice at a fraction of the cost. But they share one hard constraint: a knowledge cutoff. Training ends on a fixed date, and everything after it is a blank. You.com Web Search is what fills that gap, fetching live web results at query time so the model can reason about the world as it is today, not as it was when training stopped.

AVERAGE KNOWLEDGE GAP

~12 months

Kimi K2.7 Code

Moonshot · Jan 2025

DeepSeek V4 Pro

DeepSeek · May 2025

GLM 5.2

Zhipu AI · Nov 2025

MiniMax M3

MiniMax · Dec 2025

Jul '24 Jan '25 May '25 avg Dec '25 Jul '26

🔎

You.com Web Search bridges the gap

Instead of relying on a model's frozen training snapshot, this demo fetches live web results at query time and injects them into the prompt context. The model reasons over today's news, not December 2025's.

❌ Without You.com Web Search

Model draws from training data only. Events after the cutoff produce hallucinations or "I don't have that information."

✅ With You.com Web Search

You.com fetches live results at query time; the model reasons over current news and recent events — not a stale snapshot.

▸

You.com APIs

Search API · Research API · live web intel · cited answers

Search API

GET ydc-index.io/v1/search · $0.005 / call

Returns live news + web results in a single call — no scraping, no stale index. Used in this demo to ground every LLM step in fresh company signals. Add livecrawl=web to get full page markdown, not just snippets.

import requests

resp = requests.get(
    "https://ydc-index.io/v1/search",
    headers={"X-API-Key": YDC_API_KEY},
    params={
        "query":            "Salesforce AI news 2026",
        "num_web_results":   5,
        "livecrawl":         "web",       # optional — full page markdown
        "livecrawl_formats": "markdown",
    },
    timeout=20,
)
data = resp.json()
news = data["results"]["news"]  # [{title, description, url, page_age}, ...]
web  = data["results"]["web"]   # [{title, url, snippets, contents.markdown}, ...]

▸Show example response

Tier	Cost	Notes
Search (web + news)	$5.00 / 1k calls	$0.005 per call
Livecrawl	$1.00 / 1k URLs	First 10 URLs per call free

A typical grounded Q&A triggers 2–3 searches = ~$0.015. That's roughly 65 questions for $1.

Research API

POST api.you.com/v1/research · from $0.012 / call

Goes beyond raw results — runs multiple searches autonomously, reads through sources, and synthesizes a cited, natural-language answer. You get a markdown response with inline citations [[1, 2]] plus the source list. Depth is controlled by the research_effort tier.

import requests

resp = requests.post(
    "https://api.you.com/v1/research",
    headers={"X-API-Key": YDC_API_KEY},
    json={
        "input":           "What are the latest AI infrastructure trends in enterprise sales?",
        "research_effort": "standard",  # lite | standard | deep | exhaustive
    },
)
data    = resp.json()
answer  = data["content"]  # markdown with inline citations [[1]]
sources = data["sources"]  # [{url, title, snippet}, ...]

Effort tier	Cost / call	Latency	Best for
Lite	$0.012	<10s	Quick lookups, high volume
Standard	$0.050	10–30s	Most production use
Deep	$0.100	<2 min	Complex, multi-faceted questions
Exhaustive	$0.450	<5 min	Comprehensive analysis

▸

Parasail — Serverless LLM Inference

OpenAI-compatible API · open-weight models · per-1M token rates

Initialise the client

import os
from openai import OpenAI

# Point the standard OpenAI SDK at Parasail — no other changes needed
client = OpenAI(
    api_key=os.environ["PARASAIL_API_KEY"],
    base_url="https://api.parasail.io/v1",
)

Discover available models

Loading…

Serverless pricing (per 1M tokens)

The table below covers the 5 models used in this demo. Parasail hosts many more — run the discovery snippet above to see everything available on your key.

Model	Provider	Input $/1M	Output $/1M
Loading…

You.com Web Search: $5.00 / 1,000 queries ($0.005 per call). Livecrawl: first 10 URLs included; excess at $1 / 1,000 URLs.

A full pipeline run (search + brief + email) costs roughly $0.003–$0.005 total — about 300 accounts per dollar.

▾

Config

Search loop limits · token caps · timeouts · system prompt

Loading…

▾

Project Structure

Key files and what each one does

File	What it does
`server.py`	HTTP server — routes requests, streams SSE events to the UI
`index.html`	Single-page frontend — all three tabs, no build step
`search_tool.py`	You.com Search API wrapper — tool schema, system prompt, and `execute_search()`
`base_agent.py`	Tool-use loop — `OpenAICompatibleAgent` drives the search → LLM → search cycle
`models.py`	Model registry — display names, model IDs, and per-token pricing
`costs.py`	Cost calculation — token + search call pricing formatted for the UI
`agents/parasail_agent.py`	Parasail-specific agent — client setup, model defaults, wires into `OpenAICompatibleAgent`
`agents/pipeline_agent.py`	Three-stage pipeline — `enrich()`, `brief_stream()`, `email_stream()`
`.env.example`	Environment variable template — copy to `.env` and add your API keys

You.com Web Search × Parasail

Any question. Live web context. Open-weight LLM.

Ready to search

A research assistant that searches the web and remembers what you discussed

From a company name to a personalized pitch, in seconds

How It Works