ExoPriors Scry API: Unlocking 60 M Research Documents for AI Alignment

Why the ExoPriors Scry API matters now

In the fast‑moving world of AI alignment research, having instant access to a massive, curated corpus of papers, forum posts, tweets, and comments can be the difference between a breakthrough insight and a dead‑end hypothesis. The original ExoPriors Scry documentation announces a public read‑only key that removes the friction of sign‑ups while still offering powerful query capabilities. This article unpacks the API’s core functions, showcases practical use‑cases, and points you to UBOS resources that can help you integrate Scry into your own AI‑driven products.

Core capabilities of the ExoPriors Scry API

1️⃣ SQL over 60 M documents

The API exposes a PostgreSQL‑compatible query layer that lets you treat the entire alignment corpus as a relational table. You can filter by source (e.g., arxiv, lesswrong, twitter), document type (post, comment, paper), date ranges, and even custom metadata fields. Because the engine runs on a highly‑optimized columnar store, simple queries return results in seconds, while more complex joins finish within a minute under typical load.

2️⃣ Embedding storage & retrieval

Researchers can upload named embeddings (vectors) that are persisted on the server. These embeddings become first‑class citizens that you can later join with the alignment.entities table, enabling hybrid lexical‑semantic queries without moving large vector files locally. The Chroma DB integration on UBOS demonstrates a similar pattern for vector‑based workloads.

3️⃣ Semantic search across the whole corpus

By combining stored embeddings with the <=> (cosine distance) operator, you can rank results by semantic similarity. This is especially useful for “find all discussions related to mesa‑optimization after 2023” or “retrieve the most conceptually similar papers to a given abstract.” The API also supports BM25 lexical search, giving you the flexibility to start with keyword matches and then re‑rank with vectors.

Key features and real‑world usage examples

• Public read‑only API key – No registration required; simply include the bearer token exopriors_public_readonly_v1_2025 in your Authorization header.
• Schema discovery – A GET /v1/alignment/schema endpoint returns the full table definition, so you can programmatically generate type‑safe queries.
• Query estimation – Use /alignment/estimate to preview execution cost before running a heavy query.
• Materialized views – Pre‑aggregated views like mv_lesswrong_posts or mv_arxiv_papers accelerate common patterns such as “top‑scoring comments” or “recent arXiv papers with embeddings”.
• Hybrid search workflow – Start with a BM25 search, then join the result set with stored embeddings for semantic re‑ranking.

Example 1: Find recent alignment papers on “corrigibility”

curl -X POST https://api.exopriors.com/v1/alignment/query \
  -H "Authorization: Bearer exopriors_public_readonly_v1_2025" \
  -H "Content-Type: application/json" \
  -d '{"sql":"SELECT id, title, uri FROM alignment.search(''corrigibility'', kinds => ARRAY[''paper''], limit_n => 20)"}'

Example 2: Store a concept embedding and reuse it

curl -X POST https://api.exopriors.com/v1/alignment/embed \
  -H "Authorization: Bearer exopriors_public_readonly_v1_2025" \
  -H "Content-Type: application/json" \
  -d '{"text":"mesa optimization", "name":"p_8f3a1c2d_mesa_opt"}'

Once stored, you can retrieve the most similar documents with a single SQL statement:

SELECT e.id, e.title, e.uri
FROM alignment.embeddings emb
JOIN alignment.entities e ON e.id = emb.entity_id
WHERE emb.embedding IS NOT NULL
ORDER BY emb.embedding  @p_8f3a1c2d_mesa_opt
LIMIT 15;

Example 3: Combine lexical and semantic filters

WITH candidates AS (
  SELECT id FROM alignment.search('inner alignment', kinds => ARRAY['post'], limit_n => 200)
)
SELECT e.original_author, COUNT(*) AS cnt
FROM candidates c
JOIN alignment.entities e ON e.id = c.id
GROUP BY e.original_author
ORDER BY cnt DESC
LIMIT 10;

This pattern first narrows the universe with a keyword search, then aggregates on the author level—perfect for mapping thought leaders in a sub‑field.

Why AI alignment researchers should care

Alignment work thrives on interdisciplinary literature reviews, rapid hypothesis testing, and reproducibility. The Scry API delivers three strategic advantages:

1. Speed of discovery – Query the entire corpus in seconds, eliminating the need to manually scrape forums or maintain local mirrors.
2. Semantic depth – Vector embeddings capture nuanced meanings that keyword search alone misses, helping you surface novel connections across domains.
3. Reproducible pipelines – Because queries are expressed in standard SQL, you can version‑control your data‑access layer and share exact query strings with collaborators.

“Having a single API that unifies lexical and semantic search over a curated alignment corpus is a game‑changer for systematic literature reviews.” – AI alignment community member

Getting started in minutes

Follow these three steps to integrate the Scry API into your workflow:

Step 1 – Grab the public key

The public read‑only token is embedded in the documentation for convenience. Keep it in a secure environment variable (e.g., EXOPRIORS_TOKEN) to avoid accidental exposure in logs.

Step 2 – Test a simple query

curl -X POST https://api.exopriors.com/v1/alignment/query \
  -H "Authorization: Bearer $EXOPRIORS_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"sql":"SELECT uri, title FROM alignment.search(''alignment'', limit_n => 5)"}'

Step 3 – Build a reusable client

Most developers wrap the endpoint in a thin SDK. Below is a Python snippet using requests:

import os, requests, json

BASE_URL = "https://api.exopriors.com/v1/alignment"
HEADERS = {
    "Authorization": f"Bearer {os.getenv('EXOPRIORS_TOKEN')}",
    "Content-Type": "application/json"
}

def query(sql):
    resp = requests.post(f"{BASE_URL}/query", headers=HEADERS, data=json.dumps({"sql": sql}))
    resp.raise_for_status()
    return resp.json()

print(query("SELECT id, title FROM alignment.search('mesa optimization') LIMIT 3"))

Once you have a client, you can chain calls, cache results, or feed them directly into downstream LLM prompts.

Deep‑dive resources on UBOS

UBOS offers a suite of tools that complement the Scry API, making it easy to turn raw query results into production‑grade AI applications.

• Explore the UBOS platform overview for a low‑code environment that can ingest Scry results and expose them as REST endpoints.
• Accelerate prototype building with the Web app editor on UBOS, which lets you drag‑and‑drop UI components that visualize search results.
• Automate data pipelines using the Workflow automation studio, perfect for scheduling nightly corpus updates.
• Leverage pre‑built UBOS templates for quick start such as the “AI SEO Analyzer” or “AI Article Copywriter” to see how semantic search can power content generation.
• For startups looking to embed alignment data into their products, check out UBOS for startups and the UBOS solutions for SMBs.
• Enterprise teams can scale with the Enterprise AI platform by UBOS, which includes role‑based access control for private embedding namespaces.
• Boost your marketing automation with AI marketing agents that can query Scry for trend analysis and feed insights directly into campaign generators.
• Join the UBOS partner program to get co‑marketing credits when you build a public app that showcases Scry‑powered features.
• Need inspiration? Browse the UBOS portfolio examples for real‑world case studies.
• Finally, the UBOS pricing plans include a generous free tier that covers most research‑grade workloads.

Complementary integrations you might love

While the Scry API handles data retrieval, you often need downstream processing. UBOS’s ecosystem already supports popular AI services:

• OpenAI ChatGPT integration – feed Scry results directly into ChatGPT for summarization or hypothesis generation.
• ChatGPT and Telegram integration – create a bot that answers alignment‑related queries on the fly.
• ElevenLabs AI voice integration – turn retrieved abstracts into spoken briefings for busy researchers.