Unflatten
Unflatten says it enriches the world through epistemic infrastructure. Its principal product, Scry, gives people and agents read-only SQL, typed search, vector search and recursive/conjunctive query access over corpora built from public internet content.
At a Glance
- Individual researchers and noncommercial users
- AI agents and developers using MCP clients
- AI-safety and effective-altruism organizations
- Academic researchers
- +1 more
AI Tools by Unflatten
(1)
Scry
SQL Query Engine for AI Agents
Discussions
No discussions yet
Be the first to start a discussion about Unflatten
Latest News
Announcing Scry: a research tool over arXiv, EA Forum and other sources with SQL plus vector algebra
Scry launched as public read-only SQL-over-HTTPS/MCP access to public internet research corpora
Agent-first research tool project posted to Manifund; creator described the product, goals, corpus expansion plan and five-month funding need
Updated Terms of Service identify Unflatten as the operating name of ExoPriors, Inc. and describe Scry's paid access, provenance and acceptable-use model
Products & Services
An open-alpha research platform and MCP server providing read-only SQL, typed search, vector search, recursive/conjunctive queries, compositional embeddings, reranking and alerts over public-web corpora. It is available through the MCP endpoint https://mcp.scry.io and an HTTP API.
Open-source research/software for LLM sorting: pairwise comparisons that produce readings on a shared ratio scale, with attention to consistency, error and cost.
Research into eliciting more stable cardinal latent beliefs from language models through pairwise ratio comparisons and compositional inference-time techniques.
Market Position
Unflatten positions Scry as an agent-first, open-alpha alternative to conventional web search and answer/deep-research products: instead of returning a small set of links, it lets an agent run bounded SQL, joins, recursive queries, vector composition and reranking over source-native records while retaining provenance. The Scry site contrasts its benchmark result with Google's DeepSearchQA/Gemini Deep Research result and emphasizes exhaustive querying and negative evidence.
Leadership
Founders
Xyra Sinclair
Founder and lead of Unflatten/ExoPriors. Sinclair attended West Point from 2017–2019, trained as an Army deep-sea diver and left West Point in 2019 to build epistemic infrastructure; they describe systems-programming and epistemic-tool work during 2020–2023 and incorporated ExoPriors in 2024. Their personal site also notes first place among 4,800 teams in the 2019 Interdisciplinary Contest in Modeling.
Executive Team
Xyra Sinclair
Founder and lead
Founder of ExoPriors/Unflatten and creator of Scry; former West Point cadet and Army deep-sea diver, with systems-programming and epistemic-tool engineering experience.
Founding Story
Sinclair describes leaving West Point in 2019 to build epistemic infrastructure, then incorporating ExoPriors in 2024. After Claude Opus 4.5's release in 2025, Sinclair rented a dedicated server and focused on making important internet content queryable by agents; Scry was built and launched as an agent-first research system using SQL, embeddings and vector composition.
Business Model
Revenue Model
Scry combines subsidized noncommercial access with prepaid commercial query usage, monthly patron/team plans, and custom source builds or dataset engagements. The Terms say paid usage draws from prepaid credits and commercial engagements fund the public-benefit infrastructure.
Pricing Tiers
$5 of signup credit, no card; API key and MCP; non-commercial use.
Non-commercial; the $100 monthly payment becomes balance and rolls over.
Commercial use, dedicated capacity and custom source builds.
x402 access without an account; the declared time box is charged.
Target Markets
- Individual researchers and noncommercial users
- AI agents and developers using MCP clients
- AI-safety and effective-altruism organizations
- Academic researchers
- Commercial teams needing large evaluations, custom source builds or dataset engagements
- Agent-assisted research over large public-web corpora
- Exhaustive or negative-evidence searches that are difficult to perform with link lists
- AI-safety, effective-altruism and existential-risk research
- Academic and literature research over arXiv and papers
- Tracking concepts, attitudes or attributes over time with vector composition
- Standing monitoring queries and nuanced alerts