<500KAI agentCinematicB2D (developers)ExplainerSeries C+Feature launchDemoUSVertical AI
TLVC Rating
3.60
Hook
Editing / Creativity
Copy
Sentiment of launch
Distribution strategy
Community Rating
No ratings yet
to rate this launch.
Exa Snapshot is a time-indexed web search API aimed at AI researchers, quant teams, and agent developers who need to query the internet as it existed on a specific past date. Powered by over 400 billion webpage snapshots spanning 2 decades, given a date, Exa returns results from the web as it was at that point in time. Developers plug it in by setting snapshotAsOf in the /search and /contents API endpoints , so an existing Exa integration can be pinned to any moment in history without rewriting agent code.
The launch matters because temporal contamination has become a real problem for anyone evaluating models against the live web. If a training run takes place in September on tasks written in June, some solutions may already be on the web in a paper, a GitHub pull request, or a blog post, and agents with search tools will find and submit the answer, making it hard to tell a genuinely solved task from one copied from the web. Snapshot addresses that directly, and it opens the same kind of point-in-time methodology that quant researchers already rely on for prices. Point-in-time datasets exist for stock prices and fundamentals but nothing similar existed for the web, so trading signals derived from web page insights would take months of careful data collection to test.
Exa is the AI-native search company behind the underlying index and retrieval stack, started by Will Bryk and Jeffrey Wang from their Harvard dorm room in 2021 with a focus on building search infrastructure specifically for agents rather than humans. Snapshot ships as a research preview alongside the existing Search, Contents, and Agent APIs, and is already being used for RL training at labs, backtesting prediction models, and exploring the pre-AI web.
Comments (9)
Sign in to join the discussion.
P
Priya Vatsan26d ago
A time machine framing is cute but the real unlock is backtesting prompts against a pre-ChatGPT internet. Every model eval paper from the last 18 months just got a footnote it can't answer.
Y
Yuki Tanabe26d ago
naive question but what counts as a snapshot here, is it a full DOM capture or just the rendered text at time T? asking because those two answers imply very different products.
C
Chidi Okafor26d ago
ok wait, 400B snapshots is either the coolest data moat of the year or a storage bill that keeps someone up at night. probably both.
M
Mikko M.26d ago
The launch video pacing is solid but that voiceover mix is buried under the synth pad. Duck the music 4dB when the founder speaks and it slaps.
R
renata builds26d ago
watching this while I procrastinate on my own launch (3 weeks out, promise). the tweet hook is clean, no fluff, straight to the demo. stealing this structure.
D
Dmitri Volkov26d ago
Cool index. Curious how you handle robots.txt retroactively and takedown requests from sites that don't want their 2016 selves searchable. That's the actual ship problem.
A
Amara Nwosu26d ago
No pricing on the page which usually means it's either enterprise-only or you haven't decided. Please don't do per-query billing on a research tool, it kills exploration every time.
L
Leila Haddad26d ago
per-query would be brutal for RL workloads. subscription with a generous quota or bust, otherwise labs just spin up their own crawlers again.
S
Sanjay P.26d ago
hot take: this is the last serious web index built primarily for humans to query. everything downstream of this is an agent hitting it 10k times a minute.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
ExaAI
The world's data for AI agents
The world's data for AI agents
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
Exa Snapshot is a time-indexed web search API aimed at AI researchers, quant teams, and agent developers who need to query the internet as it existed on a specific past date. Powered by over 400 billion webpage snapshots spanning 2 decades, given a date, Exa returns results from the web as it was at that point in time. Developers plug it in by setting snapshotAsOf in the /search and /contents API endpoints , so an existing Exa integration can be pinned to any moment in history without rewriting agent code. The launch matters because temporal contamination has become a real problem for anyone evaluating models against the live web. If a training run takes place in September on tasks written in June, some solutions may already be on the web in a paper, a GitHub pull request, or a blog post, and agents with search tools will find and submit the answer, making it hard to tell a genuinely solved task from one copied from the web. Snapshot addresses that directly, and it opens the same kind of point-in-time methodology that quant researchers already rely on for prices. Point-in-time datasets exist for stock prices and fundamentals but nothing similar existed for the web, so trading signals derived from web page insights would take months of careful data collection to test. Exa is the AI-native search company behind the underlying index and retrieval stack, started by Will Bryk and Jeffrey Wang from their Harvard dorm room in 2021 with a focus on building search infrastructure specifically for agents rather than humans. Snapshot ships as a research preview alongside the existing Search, Contents, and Agent APIs, and is already being used for RL training at labs, backtesting prediction models, and exploring the pre-AI web.
Comments (9)
A time machine framing is cute but the real unlock is backtesting prompts against a pre-ChatGPT internet. Every model eval paper from the last 18 months just got a footnote it can't answer.
naive question but what counts as a snapshot here, is it a full DOM capture or just the rendered text at time T? asking because those two answers imply very different products.
ok wait, 400B snapshots is either the coolest data moat of the year or a storage bill that keeps someone up at night. probably both.
The launch video pacing is solid but that voiceover mix is buried under the synth pad. Duck the music 4dB when the founder speaks and it slaps.
watching this while I procrastinate on my own launch (3 weeks out, promise). the tweet hook is clean, no fluff, straight to the demo. stealing this structure.
Cool index. Curious how you handle robots.txt retroactively and takedown requests from sites that don't want their 2016 selves searchable. That's the actual ship problem.
No pricing on the page which usually means it's either enterprise-only or you haven't decided. Please don't do per-query billing on a research tool, it kills exploration every time.
per-query would be brutal for RL workloads. subscription with a generous quota or bust, otherwise labs just spin up their own crawlers again.
hot take: this is the last serious web index built primarily for humans to query. everything downstream of this is an agent hitting it 10k times a minute.