Research-First Agent Shell
for Scientific Rigor.

Source-backed research, with specialist agents when useful.

Built on Pi Runtime

GitHub

Need just the skills? Install the skills-only bundle.

Hypatia CLI

Specialists / called when useful

Five roles. Clear boundaries.

Hypatia can delegate to these specialists when a workflow benefits from it. Narrow requests stay lead-owned.

ROLE / AUDIT

Auditor

Checks source support for claims and removes what the evidence cannot support.

Checks URLs and paper identity, then ties claims, results, and tables to inspectable evidence.

An index of capabilities, not a required workflow order.

Research workflows

Find the next useful command.

Run as slash commands in the REPL or as hypatia <workflow> from your shell.

Explore

01

Find, read, and connect sources

/deepresearch

Plan-first investigation across papers, web, and code, with parallel evidence scouts when the topic needs them

/lit

Literature review from primary sources with consensus mapping

/compare

Side-by-side source comparison with agreement and conflict matrix

/summarize

Summarize papers, reports, READMEs, local artifacts, and PDFs with bounded-window reading for long sources

Assess

02

Review and reproduce work

/review

Internal research critique with severity scores and a revision plan

/audit

Paper-to-code mismatch audit for reproducibility claims

/replicate

Replication plan with environment choice before any execution

Create & test

03

Turn evidence into next steps

/recipe

Ranked ML training recipes grounded in papers, datasets, docs, and code

/draft

Polished paper-style draft with inline citations from findings

/autoresearch

Bounded loop: hypothesize, experiment, measure, repeat

Sources

Where Hypatia looks for evidence.

Literature databases
Semantic Scholar, OpenAlex, arXiv, PubMed, Europe PMC, bioRxiv/medRxiv, Crossref, CORE, and DOAB with stable identifiers
alphaXiv
Paper search, Q&A, code reading, and annotations
Hugging Face Hub
Dataset schema, repo file, and small text-file inspection
Web research
Search, page fetching, and PDF extraction; Exa works without a key
Documents
Local PDF and office-document parsing, search, and page screenshots

Built on Pi, with alphaXiv and scholarly database integrations. Selected workflows can produce cited briefs with provenance sidecars.