Source Filtering

Control exactly which sources your search uses. Focus on trusted domains, target specific datasets, exclude unreliable sources, or soft-rank toward preferred domains.

Three controls

Parameter	Type	Effect
`included_sources`	string[]	Only search within these sources (hard filter)
`excluded_sources`	string[]	Remove these sources from results (hard filter)
`source_biases`	object (source → int)	Soft-rank from `-5` (demote) to `+5` (boost), no hard filter

If both included_sources and excluded_sources are set, included_sources wins. Use hard filters for strict control; use source_biases to nudge ranking while still letting any source appear if highly relevant.

What you can pass

Every source field accepts any mix of these formats:

Format	Example	What it does
Domain	`"arxiv.org"`	Includes/excludes the entire domain
Base URL	`"https://docs.aws.amazon.com"`	Includes/excludes the entire site
Specific path	`"nasa.gov/news"`	Targets only that path
Dataset id	`"valyu/valyu-arxiv"`	A Valyu proprietary dataset
Preset	`"academic"`	Expands to a curated bundle for a domain (see below)
Collection	`"collection:my-sources"`	Expands to your saved collection
Web keyword	`"web"`	Includes general web search alongside any datasets you list

Paths are exact: "valyu.ai/blog" affects only that path. To include a whole domain, use just the domain name.

Presets

Presets are curated bundles of vetted sources. Pass a preset name and it expands to the right datasets and domains for you - no need to remember individual ids.

Preset	Covers
`academic`	Research papers and preprints (arXiv, PubMed, bioRxiv, …)
`finance`	Markets, filings, and financial data
`patent`	Patent records (USPTO, EPO)
`transportation`	Transportation and logistics data
`politics`	Government and parliamentary sources
`legal`	Case law and legislation
`health`	Clinical trials, drug labels, and health data
`genomics`	Genomics and bioinformatics databases
`chemistry`	Chemical and drug-discovery databases (ChEMBL, Open Targets, PubChem)
`physics`	Physics open data
`medical`	Cross-domain life-sciences bundle (health + chemistry + genomics + medical literature)
`cybersecurity`	Vulnerability intelligence (CISA KEV, NVD, EPSS, MITRE ATT&CK)
`environment`	Weather, air quality, disasters, and natural events
`automotive`	Vehicle records, recalls, complaints, and investigations
`compliance`	Sanctions and watchlists
`pulse`	Attention and adoption signals

Presets are disjoint - a dataset lives in exactly one. academic is papers and preprints only: clinical trials, drug labels, WHO and NIH grants are in health, and ChEMBL, Open Targets and PubChem are in chemistry. medical is the one deliberate cross-domain bundle. Some presets resolve to datasets that are DeepResearch-only and return no results on /search - check the access_modes field on GET /v1/datasources before relying on one.

from valyu import Valyu

valyu = Valyu()
response = valyu.search(
    "mRNA vaccine thermostability",
    included_sources=["academic", "health"],  # combine presets
)

import { Valyu } from "valyu-js";

const valyu = new Valyu();
const response = await valyu.search("mRNA vaccine thermostability", {
  includedSources: ["academic", "health"],
});

curl -X POST https://api.valyu.ai/v1/search \
  -H "x-api-key: your-valyu-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "mRNA vaccine thermostability",
    "included_sources": ["academic", "health"]
  }'

Premium presets (finance, patent, genomics, chemistry, and others that resolve to proprietary datasets) require a subscription. academic and web work on any plan, and a plan also lowers your cost per credit.

Reuse the same source combinations often? Save them as a Collection and reference by name.

Source biases

source_biases influences ranking without hard filtering. Biased sources can still appear (or drop) based on relevance - values just nudge the order. Range: -5 (strong demotion) to +5 (strong boost); 0 has no effect.

response = valyu.search(
    "climate change policy impact",
    source_biases={
        "epa.gov": 5,
        "nasa.gov": 4,
        "noaa.gov": 3,
        "example.com": -4,
    },
)

const response = await valyu.search("climate change policy impact", {
  sourceBiases: {
    "epa.gov": 5,
    "nasa.gov": 4,
    "noaa.gov": 3,
    "example.com": -4,
  },
});

curl -X POST https://api.valyu.ai/v1/search \
  -H "x-api-key: your-valyu-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "climate change policy impact",
    "source_biases": { "epa.gov": 5, "nasa.gov": 4, "noaa.gov": 3, "example.com": -4 }
  }'

You can combine source_biases with included_sources / excluded_sources in the same request for fine-grained control.

More examples

Target academic datasets by id

Python

valyu.search(
    "quantum computing error correction",
    included_sources=[
        "valyu/valyu-arxiv",
        "valyu/valyu-pubmed",
        "valyu/valyu-biorxiv",
        "valyu/valyu-medrxiv",
        "valyu/valyu-chemrxiv",
    ],
)

Restrict to official documentation

Python

valyu.search(
    "React server components best practices",
    included_sources=[
        "https://react.dev/",
        "https://nextjs.org/docs",
        "https://docs.aws.amazon.com/",
        "developer.mozilla.org",
    ],
)

Exclude unreliable sources

Python

valyu.search(
    "artificial intelligence safety research",
    excluded_sources=["example.com", "example.org", "example.net"],
)

Medical research

Python

valyu.search(
    "immunotherapy cancer treatment efficacy",
    included_sources=[
        "valyu/valyu-pubmed",
        "valyu/valyu-clinical-trials",
        "valyu/valyu-drug-labels",
        "valyu/valyu-medrxiv",
    ],
)

Financial research with government sources

Python

valyu.search(
    "cryptocurrency regulation impact banking sector",
    included_sources=[
        "federalreserve.gov",
        "sec.gov",
        "treasury.gov",
        "imf.org",
    ],
    max_num_results=15,
)

Collections

Save reusable source bundles

Data sources

Every dataset id and preset

Getting Started

Guides & Best Practices

Use Cases

Core Concepts

Compare

Account & Pricing

Other

Source Filtering

Three controls

What you can pass

Presets

Source biases

More examples

Collections

Data sources

​Three controls

​What you can pass

​Presets

​Source biases

​More examples

​Related

Collections

Data sources

Three controls

What you can pass

Presets

Source biases

More examples

Related