What is Alexandria?
Alexandria verifies claims against curated ground truth libraries. It extracts testable claims from text, checks them against the libraries you select, and reports consensus — never promising truth, only transparency.
Instead of asking an AI “is this true?”, Alexandria asks “what do these specific sources say?” You choose the sources. You interpret the results. The system provides the evidence.
Quick Start
Verifying your first piece of text takes three steps.
Select your ground truth libraries
Choose the sources you trust. System libraries like SciFact, FEVER, and PolitiFact are available by default. You can also subscribe to community libraries or upload your own.
Paste text to analyze
Enter any text — an article excerpt, a social media post, a research abstract. Alexandria will extract every testable claim and decompose them into atomic subclaims.
Review the results
Each claim is checked against your selected libraries. You'll see color-coded consensus labels showing what the evidence says, with supporting and contradicting sources listed.
Understanding Results
After analysis, each claim receives a consensus label based on what your selected libraries contain. These labels represent the weight of evidence, not a truth judgment.
Alexandria reports what your chosen libraries say — it never declares truth.
Libraries
Libraries are the foundation of Alexandria. Each library is a curated set of claims that serves as a reference point for verification. The more libraries you select, the more perspectives you get.
System Libraries
Pre-built, verified collections available to all users.
SciFact
45,911 biomedical claims from expert annotations and corpus sentences
FEVER
108,680 general knowledge claims from Wikipedia with support/refute labels
PolitiFact
11,502 political statements with truth ratings from pants-fire to true
Personal Libraries
Upload your own documents — PDFs, articles, research papers — and Alexandria converts them into personal libraries. These are private by default and can be used alongside system libraries for verification.
Subscribing to Libraries
Toggle which libraries appear in your ground truth selector from the Libraries page. Claim counts next to each library show how many reference points it provides — more claims generally means broader coverage.
Communities
Communities bring collaborative governance to knowledge curation. A community is a group of people who share epistemic commitments and collectively maintain one or more libraries.
Create a community
Start a community around a shared domain or set of values. Define what counts as evidence, what standards claims must meet, and who can contribute.
Roles and permissions
Five roles govern access: Owner (full control), Admin (manage members and settings), Moderator (review proposals), Contributor (submit proposals), and Member (read access).
Proposals
Suggest adding or removing claims from community libraries. Proposals are reviewed and voted on by moderators and admins before being applied.
Multi-library governance
A single community can govern multiple libraries — for example, a medical research community might maintain separate libraries for clinical trials, meta-analyses, and case studies.
Explore existing communities or create your own from the Communities page.
API Access
Access Alexandria programmatically to integrate claim verification into your own tools and workflows.
Create an API key
Go to Settings and generate an API key. Keys are shown only once — store them securely.
Key scopes
read
List and search libraries
verify
Run claim verification
write
Add claims to libraries
See the full endpoint reference in the API Documentation.
Philosophy
Alexandria is built on a set of principles about how technology should relate to truth.
Epistemic sovereignty
You choose your priors. Alexandria never imposes a single source of truth — you select which libraries govern verification, and the results reflect those choices.
Models are tools, not authorities
AI models extract and compare claims. They never decide what is true. Truth comes from the libraries you trust, not from model opinion.
Consensus from density
When multiple claims across multiple libraries point in the same direction, the signal is stronger. A single match is informative; corroboration across sources is compelling.
Part of Cosmos Institute
Alexandria is supported by the Cosmos Institute, whose mission is building AI for human flourishing through three pillars: truth-seeking, human autonomy, and decentralization.