The July 2025 DeepSearch Upgrade: Remedy your Context Rot

Our July 2025 release of the Valyu DeepSearch API is here and it’s laser focused on one goal: banishing context rot from your AI’s context windows. We’re rolling out 7 updates to deliver only the information your models need, exactly when they need it.

From ultra‑clean HTML to Markdown parsing and region‑specific relevance, to 40 percent faster queries and rock‑solid financial ticker lookups, these enhancements yield leaner, quieter context windows, sharper relevance, and full configurability whether you need a brief snippet or a deep‑dive dossier.

Here’s what we are shipping this month:

1. Cleaner Web Parsing: HTML → Markdown

Say goodbye to noise that trips up your model’s attention: sidebar ads, pop‑ups, tracking scripts, even rogue navigation bars. With our new dynamic HTML-to-Markdown parser, all unnecessary content is stripped out at parse time. It learns for any website what content on a page is relevant to the user and what you get is the distilled essence of a page in tidy Markdown. No distractions, no noise.

Before (excerpt from an ArXiv Abstract full of irrelevant information):

# Computer Science > Computation and Language
# Title:Attention Is All You Need
Computation and Language (cs.CL); Machine Learning (cs.LG)

arXiv:1706.03762 [cs.CL]

(or arXiv:1706.03762v7 [cs.CL] for this version)

https://doi.org/10.48550/arXiv.1706.03762 Focus to learn more arXiv-issued DOI via DataCite

# Submission history
HTML (experimental)
References & Citations
Semantic Scholar
DBLP - CS Bibliography
## BibTeX formatted citation
# Bibliographic and Citation Tools
# Code, Data and Media Associated with this Article
# arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.

New dynamic parsing (Now only extracts relevant content from the page):

[Submitted on 12 Jun 2017 (v1), last revised 2 Aug 2023 (this version, v7)]

# Title:Attention Is All You Need
Authors:Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, Illia Polosukhin

Abstract:The dominant sequence transduction models are based on complex recurrent or convolutional neural networks in an encoder-decoder configuration. The best performing models also connect the encoder and decoder through an attention mechanism. We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions entirely. Experiments on two machine translation tasks show these models to be superior in quality while being more parallelizable and requiring significantly less time to train. Our model achieves 28.4 BLEU on the WMT 2014 English-to-German translation task, improving over the existing best results, including ensembles by over 2 BLEU. On the WMT 2014 English-to-French translation task, our model establishes a new single-model state-of-the-art BLEU score of 41.8 after training for 3.5 days on eight GPUs, a small fraction of the training costs of the best models from the literature. We show that the Transformer generalizes well to other tasks by applying it successfully to English constituency parsing both with large and limited training data.

That leaner and more relevant context means fewer spurious tokens competing for your model’s attention, so it never “forgets” what matters.

2. Region-Specific Search

Stop one‑size‑fits‑all indexing from watering down your local insights. With the new country_code parameter, you can bias results geographically. Wether it is UK news get prioritised for London‑based users, Australian weather for Sydney applications, and so on:

response = valyu.search(
  "renewable energy policy updates",
  country_code="GB",        # prioritize UK sources
  max_num_results=5
)

All your queries honour regional context, giving you results that matter locally.

3. Exclude Unwanted Sources

Control your corpus by blocking entire domains, paths, or datasets with excluded_sources. No more irrelevant outlets or unwanted sources muddying your results.

data = valyu.search(
  "quantum computing breakthroughs",
  excluded_sources=[
    "<http://example-blog.com/old-posts/>",
    "wikipedia.org"
  ]
)

DeepSearch will omit any result matching those patterns. No more firefighting the fallout from off‑topic content.

4. Enhanced Ticker Lookup for Financial Data

We’ve overhauled our equity lookup so that lesser-known or OTC tickers resolve flawlessly. No more guesswork outside the S&P 500.

data = valyu.search(
  "BJ holdings stock over the last 7 years"
)

5. Customisable Response Length

Sometimes you need a tweet‑length snippet; other times, you need a deep dive. With fully configurable response_length , you can now decide how much text you need. Set an exact character length, or use presets:'short' (25 000 chars), 'medium' (50 000 chars), 'large' (100 000 chars), or 'max' for the full content.

# Return roughly 30k characters:
response = valyu.search(
  "history of machine learning",
  response_length=30000
)

# Use preset sizes:
response = valyu.search(..., response_length="short")

By tailoring the volume of content, you avoid overloading your prompt window and keep context freshness high.

6. 40% Faster Web Search

Under the hood optimisations and better caching have slashed average web search query times by 40%. Faster results for real-time agents, and happier users.

7. Refreshed Platform UI

Our dashboard, playground, and datasets pages have a fresh coat of paint.

The new dashboard for the Valyu Platform

What’s Next: Deeper Financial Insights

We’re rolling out expanded financial datasets in the next few weeks: SEC Filings, company fundmentals, and more market signals. Stay tuned for updates your Agents can’t (and shouldn’t) trade without.

Sunsetting V1 API

The V1 API was officially deprecated as of June 26, 2025. Please ensure you’ve migrated to V2 to avoid any interruptions. All existing V1 functionality remains available and improved in V2.

Getting Started

No client upgrades are needed. All these features live in the DeepSearch API and are available immediately with your existing API key. Explore our docs and let us know what you build.

Questions or feedback? Reach out at founders@valyu.network —we love hearing from you!