troublemaker @youdotcom creator of https://xitter.cf/t.co/0Rqz1Id86I and https://xitter.cf/t.co/0ceGmq3QIF fascinated by the shape of the stone
internet
Joined June 2015
- Tweets730
- Following592
- Followers679
- Likes413
Sparker retweeted
we and many others in our industry know agentic work can't happen without live, external data from the web. this is why @youdotcom exists.
we decided to take it one step further: provide inference for a range of inexpensive, high performance open (and closed) weight models with our own web search and web fetch tools built into the system.
over the years, we've built a strong understanding of how agents use web search tools. now, you can reap the benefits of these learnings right out of the box when you run inference with us.
you.com/router
Introducing Inference Router: enterprise-grade inference on demand.
Kimi K3, GLM 5.3, DeepSeek V4.1 Flash, Qwen3.8 Flash, Muse Spark 1.3, Mercury 3 and more, available with an uptime SLA of at least 99.9%. US hosting and web search built in.
you.com/router
Sparker retweeted
Sometimes, improving web search means going beyond the search system itself.
It means training models to use search better, reason through queries, find the right information, and get to answers with fewer tool calls.
That’s what we explored with @nvidia and @CoreWeave, improving BrowseComp accuracy by 22.94% while reducing average tool calls from 168 to 115.
@youdotcom, @nvidia, and CoreWeave post-trained NVIDIA Nemotron 3.5 Lightning using SFT and RL Rollouts.
The results: BrowseComp accuracy rose from 36.97% to 45.45%, with 30% fewer tool calls per task—and we did it in a three-week sprint.
Read all the details and see what you can pull into your own post-training here: crwv.co/ut8yF
Sparker retweeted
MULTIPLAYER
The latest release of Joshu now can coordinate with other Joshus. Just ask your Joshu to make a team link, share it with other Joshus, and they can all connect to each other securely.
- Can be your coworkers, friends, even your spouse ;)
- Never try to find a document, email, or timeslot again: knowledge is shared between Joshus, but totally under your control
- Joshus do their own type of daily stand up, and figure out if anyone is waiting on something, and will proactively nudge their human owner
Try it, 14 days free. Open source as always, source code in next link!
Sparker retweeted
we put in a lot of work to make our products searchable, readable and interpretable by agents (and humans too).
a good precursor for how well our actual APIs work when plugged into your agents.
#1 of 556 on @getpostman's AI-ready API leaderboard.
In good company with @OpenRouter, and @vercel.
Agents can find us, read us, and ship with us.
Sparker retweeted
#1 of 556 on @getpostman's AI-ready API leaderboard.
In good company with @OpenRouter, and @vercel.
Agents can find us, read us, and ship with us.
Sparker retweeted
Introducing Inference Router: enterprise-grade inference on demand.
Kimi K3, GLM 5.3, DeepSeek V4.1 Flash, Qwen3.8 Flash, Muse Spark 1.3, Mercury 3 and more, available with an uptime SLA of at least 99.9%. US hosting and web search built in.
you.com/router
Sparker retweeted
We've always been about bringing intelligence to customers, every since @RichardSocher started the business and he first conceived of web search for agents and LLMs.
Today we launch a natural compliment, Inference Router!
Tokens, Web Search, or both.
We have been all in on the combination of open weight models, growing token overall use but maxing cost efficiency for awhile now.
We saw a real need for an enterprise grade offering, and so we built our own product. Full access to dozens of models from our many partners. Fully reliable. ZDR.
Its even better when you ground inference with web search! The best of both, one API. ✨
Introducing Inference Router: enterprise-grade inference on demand.
Kimi K3, GLM 5.3, DeepSeek V4.1 Flash, Qwen3.8 Flash, Muse Spark 1.3, Mercury 3 and more, available with an uptime SLA of at least 99.9%. US hosting and web search built in.
you.com/router
Sparker retweeted
Recursive just dropped Flint and it changes what a model release is.
It's an open-weight 40B mixture-of-experts coding model, and another AI system helped build it.
> architecture improvements
> low-bit training improvements
> specialised inference kernels
> speculative decoding gains
> built from scratch in under four months
Recursive's Eureka Machine found every one of those, and its work touched nearly every part of the stack, from data and training recipes to infrastructure.
We are in the era where AI does AI research instead of just running on it
🤝 Paid partnership
Introducing Flint, @Recursive_SI’s first model: a fast local coding model built with our automated research system, the Eureka Machine.
• Best-in-class fast local coding
• Up to 2x faster than comparable models: 125+ tokens/sec on a MacBook Pro, 150+ on RTX Spark
• Built from scratch in under 4 months
Private, unmetered intelligence on your own machine. Open weights available now on @huggingface, run it locally on your favorite framework or try it for free on @opencode or @pidotdev.
Sparker retweeted
Introducing Flint, @Recursive_SI’s first model: a fast local coding model built with our automated research system, the Eureka Machine.
• Best-in-class fast local coding
• Up to 2x faster than comparable models: 125+ tokens/sec on a MacBook Pro, 150+ on RTX Spark
• Built from scratch in under 4 months
Private, unmetered intelligence on your own machine. Open weights available now on @huggingface, run it locally on your favorite framework or try it for free on @opencode or @pidotdev.
Sparker retweeted
Cursor, meet You.
4 skills plus an MCP server with 4 tools for search, contents, research, and finance, now in the @cursor_ai Marketplace.
Your agent asks. You answers. Cursor writes the code.
cursor.com/marketplace/ydc
Sparker retweeted
Your agent team needs current sources, not another tab for you to babysit.
@youdotcom is now in Paperclip. Give your researcher web search and page extraction, hand a sourced brief to your writer, and review the result. All within the same project.
We're partnering with @papercliping to give every agent in your org the live web.
One connector in Paperclip’s catalog, every runtime covered: Claude, Codex, Gemini, Cursor, OpenClaw, Hermes.
Agents stop working from an old map, and every search lands in Paperclip’s audit log.
Start on the free profile: no key, 100 queries a day.
Sparker retweeted
@youdotcom is now an integration in @simdotai!
Give your agents access to what’s happening beyond your internal tools.
With @youdotcom, agents can search the live web, read pages, and research questions with sources—from daily news briefs and product comparisons to deeper research on a topic.
Bring those findings into the rest of your workflow: update a document, send a summary to your team, or add fresh information to a knowledge base.
Link in comments.
Sparker retweeted
I really enjoyed wiring together @youdotcom's powerful APIs, @typesafeai 's Jev, and @n8n_io to create a useful daily news digest workflow, custom-tailored to your specific interests. I hope you find it useful too! Check it out here: blog.n8n.io/spotlight-you-co…
Filtering is the expensive half of a news digest. Every article needs a judgment call, and a judgment call has always meant an LLM call.
Not anymore. $0.00004 per article.
We teamed up with @n8n_io and built a weekday AI brief using @typesafeai's Jev. Now live on n8n blog as part of their Verified Node Spotlight. 🧵
Sparker retweeted
You.com (@youdotcom) is now available as a connector in Merge Agent Handler!
Agents can search the live web and pull current results into their work, from this morning's news to a competitor's updated pricing page.
Every search runs through the same auth, permissions and logging as the rest of your agent's tools.
Check it out: merge.dev/merge-agent-handle…
Sparker retweeted
Humanity has problems that technology can solve well, @youdotcom and @Recursive_SI co-founder and CEO @RichardSocher tells @inafried.
“I believe that when we can automate research, first for AI but eventually also for all other kinds of sciences, humanity will massively benefit from us." #AxiosLive
Sparker retweeted
WATCH: @inafried and @MadisonMills22 host conversations with @Box CEO @levie, @CredoAI CEO @navrinasingh , and @youdotcom CEO @RichardSocher on digital sovereignty in the AI era.
xitter.cf/i/broadcasts/1AKEmvZEl…
Sparker retweeted
Had my first ever VC meeting for tryskillissue.com and my big takeaway was that I need to find a better way to word what I'm building. Everyone I pitch this to is initially confused.
Right now I say "I'm building a marketplace for Consumer AI agent builders to find tools, skills, etc"
Also, I started building this with consumer AI agent builders in mind, but am realizing that the tool/skill creators actually care about this more.
Founders who dealt with this prior; how did you do it?
Sparker retweeted
what can an ai agent buy over x402 today? per call:
→ web search on @exaailabs and @youdotcom: $0.005-0.007
→ smart money and wallet intel on @nansen_ai and @arkham: $0.01-0.20
→ wallet portfolios and contract data on @zerion and @blockscout: from $0.002
→ onchain metrics on @glassnode and @alliumlabs : $0.01-0.05
→ dex prices on @coingecko and @CoinMarketCap : $0.01
→ images and video on stablestudio by @merit_systems: $0.13 an image
→ gift cards on @bitrefill at face value, flight search on stabletravel: $0.02
→ llm calls on @BlockRunAI, 87 models: under a cent for a short prompt
the agent pays first and earns after. that's what credit is for
agents with a clean record can now pay less for the same api call.
x402 sellers price by priors record. repay more, pay less. default once and you're refused.
this is where credit starts compounding for machines.
@openclaw @elizaOS @OpenRouter @Paxos @a16zcrypto @virtuals_io @RobinhoodApp
Sparker retweeted
Web search is now part of how open-weight models get trained.
We teamed up with @nvidia and @CoreWeave to put @youdotcom search inside the RL loop for Nemotron 3.5 Lightning.
BrowseComp 36.97% → 45.45%. 30% fewer tool calls.
Smarter model, less searching. More to come with this crew!
In reinforcement learning, inference is part of the training loop. Every checkpoint used to mean a redeploy, and the trainer waited.
CoreWeave RL Rollouts load the new weights into a live deployment without touching in flight requests, about 15x faster than a redeploy cycle.
@nvidia and @youdotcom used it to post train Nemotron 3.5 Lightning for web search and lifted BrowseComp accuracy from 36.97% to 45.45% while cutting tool calls by 30.24%.
Full breakdown here: crwv.co/utdYc
Sparker retweeted
Which companies will have a #1 AI model by December 31?
Live answer now with Knowledge on @youdotcom.
One parameter. No connector list to maintain.
Try it here - you.com/platform/playground/…