The technical backbone here is pretty impressive. Instead of just wrapping a traditional search engine, they've built an index optimized for agentic workflows. If you are building an LLM agent, you know that every millisecond of latency in the retrieval step compounds during a multi-step reasoning chain. Keenable is hitting p95 latencies under 250ms from us-east, which is critical if you want your agent to feel responsive rather than hanging for ten seconds between every thought.
A SQL-like interface for the open web #
One of the most interesting features for anyone working on complex AI workflows is their SQL-like interface. Most search APIs return a messy blob of JSON or raw HTML that requires a secondary LLM pass just to clean up. Keenable allows you to interact with the web using structured queries, which makes it much easier to perform targeted data extraction. This is a massive win for building reliable RAG (Retrieval-Augmented Generation) pipelines where you need specific facts rather than just "relevant" pages.
Here is how the core value proposition breaks down for developers:
Latency: p95 < 250ms (crucial for agentic loops)Scale: 100B+ page indexInterface: SQL-like querying for structured extractionCost: Focused on high-volume, low-cost deploymentFree Tier: 100,000 requests per month
Stop "benchmaxxing" and look at NEEDLE #
The team is also being surprisingly transparent about their performance. Instead of just claiming they are better, they open-sourced their internal benchmarking suite called NEEDLE. This isn't your typical static benchmark that uses old datasets; it’s a live benchmark that compares Keenable against other major search APIs using fresh, agent-centric queries. If you are skeptical of the marketing claims, you can actually go and verify the results yourself through their live testing suite.
The pedigree behind this is also worth noting. The founder spent seven years at Amazon as a scientist working on web grounding for Alexa/AGI, and the co-founder previously led search at Yandex. They aren't just hobbyists; they are building from the perspective of people who have spent a decade trying to make machines understand the structure of the internet.
For anyone currently struggling with the "noise" problem in agentic search or the sheer cost of high-frequency API calls, this looks like a serious contender for a production-grade deployment.
https://keenable.ai
https://keenableai.github.io/needle
Amazon is torching rare texts to fuel its AI training 7d ago
Amazon is using rare books to train its AI models 8d ago
AI books now make up 20% of Amazon's self-publishing catalog 10d ago
Anthropic aiming for a 2 trillion dollar IPO by October is 12d ago
Amazon order confirmation emails are basically useless now 12d ago
Next Is AI actually burning the planet down or is that just hype? →
All Replies （0） #
No replies yet — be the first!