KeenableBeta Fund × AWS × OpenAI · Oct 23

Every agent here alreadysearches the web.

It comes built into most models and most harnesses.

So why should yoursuse Keenable?

Seven reasons. Thirty seconds.

Keenable
KeenableNEEDLE · the open benchmark for agent search

#1 for speed.#1 for quality.

Out of 23 search APIs, measured on the same queries. Updated this morning.

An agent searches 10 times for one task
Keenable1.4 s
Typical search API11.5 s
Search built into chat models5.7 min
100 MS1 S10 S1 MIN← FASTERBETTER →BUILT INTO CHAT MODELSKEENABLE · 141 MS

NEEDLE, keenableai.github.io/needle, Oct 23, 08:16 UTC · Keenable in realtime mode · p50 latency, log scale · a blink takes 100–400 ms, Harvard BioNumbers

KeenableVoice · in Pipecat as KeenableWebSearch

People reply in 200 ms.

A Keenable search fits inside that pause. A typical one doesn't.

Stivers et al., PNAS 2009: mean reply gap 208 ms · NEEDLE p50, realtime mode, Oct 23, 08:16 UTC · built into chat models: 34 s, off this slide

KeenableOur own index

Our ownindex.

100B+ documents, crawled and ranked by us.

Fresh · a real result, Oct 16

Elorian's first AI model wants to reason with images

techcrunch.com
PUBLISHED16:01:37
IN OUR INDEX16:03:58
GAP2 min 21 s

keenable.ai · every result carries its acquired time, and its published time when known (UTC)

KeenablePoint-in-time search
NOT WRITTEN YET2025OCT2026APRJULNOWwhat's the newest AI modelquery_time = 2025-06-01Search2025-06-01TECHCOMMUNITY.MICROSOFT.COM · DEC 27 2024Meta's next generation model, Llama 3.1 405BHUGGINGFACE.CO · FEB 12 2025Models – Hugging FaceAICHIEF.COM · APR 18 2025OpenAI unveils new AI model capable of imageunderstandingQUERY_TIME
query_time

Search the webas it was.

Pick a date. Get the index as it stood then.

  • Replay an agent run, result for result.
  • Evals that don't drift.
  • What was knowable before the news broke.
KeenableWhat to build today

Fast enough for agentsthat see, hear and act.

See

Point a camera at a sign or a label. Read it, then search what it says.

Hear

Voice agents whose web search fits inside the 200 ms pause between turns.

Enrich

Run ten searches at once. Fill 500 rows in the background.

Watch

Check sources on a schedule. Act on whatever was acquired since the last run.

Stivers et al., PNAS 2009: mean reply gap 208 ms across 10 languages

KeenableTwo tools
Look
search_web_pages

Ranked results from our own index, each one dated.

Read
fetch_page_content

Any page as clean markdown. Or ask it a question and get just the answer.

KeenableLIVE

Set up in60 seconds.

$ brew install keenableai/tap/keenable-cli
$ keenable login
$ keenable configure-mcp --all
$ claude  # or codex, cursor, opencode
Keenableapp.keenable.ai · Other → BETA200

$200 for every builder here.

$200
CodeBETA200
KEENABLE CREDIT · OCT 23
Scan: code + setup prompts
KeenableClaim your $200 · about 20 seconds

Claimit.

  1. 01Sign in at app.keenable.ai
  2. 02Pick your tool. Copy the setup line.
  3. 03Pick Other. Type BETA200.
KeenableDemos 5:00 PM
$1,000+

in Keenable credits for the best build. What we're looking for:

KeenableOne more winner · on X

Post iton X.

Share what you built and tag @KeenableAI by the end of today. We'll pick one more winner from X.

x.com/KeenableAIScan to follow
Your name@you

Built an agent that reads a sign from a photo and looks it up live with @KeenableAI. Made today at the Beta × AWS × OpenAI hackathon.

· 9,412 Views

14 23 186 9
Keenablekeenable.ai/careers

We're hiring.

Join the team that already built search twice.

Andrey Styskin
Andrey StyskinCo-founder & CEOex-CEO of Search and Ads at Yandex.
Director of Web Infra at Amazon AGI.
Matthias Petri
Matthias PetriCo-founder & Chief Scientistex-Principal Applied Scientist at Amazon AGI.
Built the web grounding service behind Alexa.
Scan: open roles
Keenable
Your code · $200 in credits
BETA200

app.keenable.ai → Other → BETA200

docs.keenable.ai
Come find George

Scankeenable.ai
← → · F full screen · R replay