Utkarsh Srivastava

You can now create stateful web research agents with the Parallel Task API. Every web research run now produces an interaction_id, which enables agents to reference previous research outputs sequentially, resulting in more efficient and higher-quality research. Try interactions in the new developer platform showcase: platform.parallel.ai/play/interacti…

Claude Code's built-in Web Search is a single-hop search. It breaks down for batch operations and multi-source synthesis. That's why I just ran a test with the new Parallel CLI in Claude Code for bulk research enrichment. I had a CSV of 12 academic papers on emotional activation in LLMs. Only 3 columns filled (title, URL, year), needed 8 more (authors, affiliation, methodology, models tested, citation count, etc.). Claude Code invoked parallel-cli enrich run. The CLI parsed my CSV, created a task group of 12 independent web research agents server-side, and ran them all concurrently. Each agent visited the paper URL, cross-referenced sources, and returned structured JSON. Each row gets processed by a Processor (Parallel's execution engine tiers). My agent decided to use core-fast, which is cross-referenced, moderate complexity. Claude then polled with parallel-cli enrich poll, transformed the JSON back to CSV with inline Python, and wrote both a .csv and .md table. I added routing rules to my global CLAUDE[.]md so it picks the right tool automatically: - Web search / research: Use parallel-web-search or deep research - Data enrichment: Use parallel-data-enrichment - Web scraping: Use parallel-web-extract - Do NOT use built-in WebSearch/WebFetch when a Parallel Skill can handle it The context window economics seem like the real gain here. If I'd done this with WebSearch + WebFetch with a rough calculation: - Search for paper 1 → ~500 tokens for results - Fetch arxiv page → ~2,500-5,000 tokens - Fetch Semantic Scholar → ~2,500 tokens - Fetch author page → ~2,500 tokens - Repeat 12× = ~100,000+ tokens just for raw page content Plus my Claude Code would need to reason over each page to extract the 8 fields and it'd do it sequentially, probably 5-10 minutes. With Parallel, all research happens on their infrastructure. Only the final structured JSON (~3,000 tokens total for all 12 papers) enters my context. I'm not affiliated with them; I just recently started to really like their web agents. Congrats on the CLI launch, @p0 and @paraga.

Parallel is built for a world where agents are the primary users of the web. Agents can now autonomously pay per API call for web access via Parallel using MPP. parallelmpp.dev

MCPs waste precious context. Modern agent harnesses love CLIs as tools. The Parallel CLI gives your agent web super powers at the best price performance out there.

It’s a special privilege to be able to partner with companies building on the frontier. Because of this deep partnership with the team at @harvey, AI agents can now search and access content on the web that hasn’t been indexed by search engines built for humans.

BU meets Parallel Web Systems! We're collaborating with @p0 for our web search on BU app! Read our blogpost! 👇

Google nailed search for humans. Parallel search is purpose-built for agents. Vertex AI now offers both.

Proud that @p0 is the first and only non-Google web search provider to be launched as an option for grounding with Vertex AI. The boost for agents using our search is substantial, and it has been great partnering with Vertex to bring this increased accuracy to their customers.

Our customers rave about our deep research product. It's nice to see that feedback reflected in the benchmark numbers as well!

@Trace_Cohen @travers00 @CBinsights @OpenAI @GeminiApp @p0 You could try the google apps history feature as it might have stored that data in flight. You can also use "Paste Special" to get the values only for a CSV download.

@travers00 @CBinsights @OpenAI @GeminiApp @p0 Hey thanks def fun/simple to work with but broke again... It's not downloadable via sheets into CSV bc it's an API call, then it broke bc I somehow ran through all $80 so all the data is gone again and there is no way I'm paying more to do it again. What do you recommend?

I spent hours today using the @CBinsights unicorn list data to research all US (NY/SF) startups last funding date to do some fun research experiments on. I tried @OpenAI and @GeminiApp normal chat + deep research but they have to batch it manually bc the data set is in the hundreds, which broke multiple times and lost everything... very frustrating. Then I just came across @paraga @p0 @googlesheets integration like an hour ago and tried it out. Installed it, took a few min to realize how to install it, side bar, API set up and free $80 usage and it basically one shotted 700+ in about 20min one by one!

Could your workflow use a best-in-class web agent or search? We have you covered with the Parallel n8n node.

Parallel Search is now integrated with @langchain! docs.langchain.com/oss/python/int… Here are some examples we’ve created to help you get started:

We've integrated the @p0 Monitor API into @graphlit and @zine_ai, and it's wild how useful this already is. Here I created a monitor for seed fundraising announcements in the US, and it's been ingesting these web pages automatically for the last day. In Zine, I can chat with them to explore in more detail, or even use Parallel Deep Research to get really in-depth about a company and market.

@hanqiao69 This benchmark 🔥

We built FindAll at Parallel because there simply wasn’t a tool that works. This will change how you plan, research, and make decisions. Got a FindAll query? Drop what you built + your thoughts on FindAll👇 platform.parallel.ai

@arnabdotorg @MikeCafarella @AlonHalevy @sirrice @yaaang @p0 @sampullara 's dream as well.

Love that the dream of Yahoo's YQL and Google Webtables (@MikeCafarella, @AlonHalevy, @sirrice, Daisy, and @yaaang) lives on with @p0 FindAll!

Spent the afternoon integrating @p0 into our client's stack. Their API design philosophy is what every platform should aspire to: predictable response patterns, comprehensive error handling, and webhook events that actually make sense. The kind of DX that makes you want to build on it.

Monitor the web directly from Slack. Or integrate it into your app through our new Monitor API

Moving the web from a human pull to an agentic push model. Really excited for this launch!

Excited to be building with this stellar team! Let's go!

@paraga Excited to be building with this stellar team! Let's go!