Project summary

At scry.io, I make the internet programmatically searchable. I maintain a web index (270 TB and growing), and instead of doing what traditional search engines do, which is have a black-box function between essentially (keywords, strings, dates, boolean_operators) -> top_k_results that is applied to their search index, I give the user SQLProgram -> relation<row> and essentially raw access to the underlying data. This makes it trivial to do idea history mining like answering "has mechanistic interpretability writing gotten more optimistic over time" or "Which forum and mailing-list communities used a technical phrase first, and which comments carried it forward?" A SQL query over this index can actually dynamically compute and adjust search terms and what it is tabulating as it traverses billions of documents.

What are this project's goals? How will you achieve them?

I want to have a very strong index of all the remotely important signal relevant to the far future going well, and I want thousands of people taking the intelligence explosion seriously to have a research tool they prefer using on a daily basis. I particularly value optimizing for the needs of existential security grantmakers, which includes technical decisions like focusing on delivering cross-corpus author snapshots, and premier LLM structured judgement campaign infra that make it easy to probe and discover consistent model priors over online content like grant proposals, instead of being mislead by shallow, contingent, uncharitable, and uninterpretable framing effects and noise. A nice property of my focus on the analytical network process with this, is it's conduciveness to open-ended, unplanned collaborations, arguing their case purely through model structured judgements, with the mechanism being largely self-tuning and parameter-free.

I achieve this by talking with my users, deeply leveraging terminal coding agents, and soon, putting my APIs through more external benchmarks to build up social clarity about capabilities. (We had to end early because of cost, but with an older version we were about on track to be 4 points ahead of Perplexity in one).

How will this funding be used?

$20k retroactive for unpaid full-time work Dec 2025–Jul 2026

$2500/month x 12 months = $30k runway for datacenter colocation costs for hosting Scry for ~EA.

$5k monthly cloud compute costs x 3 months = $15k cloud compute runway.

$2k x 12 months AI subscriptions = $24k for foundational engineering, addressing user needs, and software maintenance.

4 x 60 TB NVMe drives at $6k each = $24k for 240 TB more of performant storage.

2 x RTX 5090s at $5k each = $10k for more embedding, OCR, and rerank functionality.

$5k x 12 months stipend = $60k. I have a good chance of success with cracking B2B sales while staying on the pure long-term path of delivering deeper infra for ~EA, so the modest stipend reflects that.

$183k total.

$20k minimum. Below full funding I allocate at my own discretion--most likely the retroactive compensation first, then pooling toward colocation. My true funding ceiling is above $183k, because mirrored servers and contractors buy superlinear returns on how fast this community can react during the intelligence explosion.

Who is on your team? What's your track record on similar projects?

I've been working solo and have already shipped much of this functionality. The API is live, reliable, and users can run long-running programs over currently 70B records. It is growing over 1B a day, with under 15 min freshness for many prioritized sources. People have paid for it repeatedly even on the far slower earlier infrastructure. The past requirement and user friction to adoption--needing to use a terminal agent like Claude Code--are being addressed with affordances like a traditional search box, a visually-friendly and question-adaptive SQL editor, MCP for use on apps like ChatGPT and Claude on mobile, and a full sandboxed agent available in-browser.

What are the most likely causes and outcomes if this project fails?

The hardware and software stack is going to be very powerful to people--agents having SQL over majority of the intellectual internet is just a very canonically powerful technology. The question is who are we going to be able to optimize for more. If this doesn't get funded through Manifund, I have to gatekeep a lot more sources for data sales and survival in the markets, and charge significantly more and price a number of people out, possibly optimizing exclusively for B2B. With Manifund funding, the question becomes more like, does this tool seem to be counterfactually steering multiples of it's funded amount in philanthropic capital allocations, and the most likely failure mode if not, is we fail to be sufficiently intelligent with building up ergonomic workflows and use of lexical and structured judgement methods for heavily guiding the identification and prioritization of giving opportunities, with the grantmakers we collaborate with.

If the project fails, people should still continue to have programmatic internet search, with machine access decided if by nothing else, by free-floating congestion pricing.

How much money have you raised in the last 12 months, and from where?

I have raised $5k from one angel investor. The substantial money that went to the minimum viable server hardware was unconditionally gifted (this service would cost over $45k/month on AWS to maintain). I believe in timeless cooperation, and being caught by intelligent actors, so I've spent the vast majority of my assets (2 month's runway remaining) in growing this compute substrate and preparing it ahead of time, for EA and pivotal decision making as the world heats up for AI progress.