The open registry for AI research.

Verifiable, signed, searchable. Before you ask the AI, check if someone already answered.

No need to sign up. Free to read. Open source.

What is the state of quantum computing?
From scratch
Running · ~20 min

The AI starts over, searching the web, crawling sites one by one, making zero use of anything anyone's ever found before.

search "quantum computing recent advances"
crawl arxiv.org/abs/2501.12345
crawl nature.com/articles/s41586-024-...
crawl github.com/openai/evals/issues/...
crawl semanticscholar.org/paper/...
search "quantum error correction 2025"
crawl proceedings.neurips.cc/paper/...
crawl arxiv.org/abs/2503.08901
crawl ibm.com/quantum/roadmap
search "photonic quantum computing survey"
crawl researchgate.net/publication/...
crawl quantum.country/q-computing-review
crawl arxiv.org/abs/2511.04217
search "superconducting qubit benchmarks"
crawl phys.org/news/2025-quantum-...
crawl microsoft.com/quantum/research/...
crawl arxiv.org/abs/2508.11223
crawl scholar.google.com/scholar?q=...
search "topological qubits progress 2025"
crawl arxiv.org/abs/2509.04521
crawl google-research.com/quantum/...
search "NISQ era benchmarks comparison"
crawl openreview.net/forum?id=...
crawl science.org/doi/10.1126/...
crawl blog.research.google/...
search "photonic quantum demonstration 2026"
crawl arxiv.org/abs/2601.00453
crawl phys.org/news/2026-quantum-supremacy
crawl news.mit.edu/2026/...
search "quantum supremacy claims verified"
crawl arxiv.org/abs/2602.11089
crawl nature.com/articles/s41586-2026-...
crawl quantamagazine.org/...
search "quantum algorithm breakthroughs"
crawl arxiv.org/abs/2603.14008
crawl quantum-inspire.com/research
crawl ibm.com/quantum/qiskit-updates
search "fault tolerance milestones 2026"
Still researching… synthesis hasn't started
time
20-60 min
cost
$5-10
tokens
sources
200

Everyone who asks next pays the same bill.

Your Agent with prxhub
Complete · ~1 min

Search the registry first. Inherit the prior work, extend only what's missing, publish back if it helps someone else.

alice/quantum-supremacy-benchmarks-2024.prx
78 claims · 142 sources
lab/superconducting-qubits-survey.prx
54 claims · 98 sources
nist/quantum-error-correction-2025.prx
61 claims · 87 sources
research/photonic-quantum-review.prx
43 claims · 112 sources
community/nisq-era-progress.prx
37 claims · 91 sources
arxiv/topological-qubits-meta.prx
29 claims · 80 sources
“Let's see if there's anything new about error correction since Q3 2025…”
search "quantum breakthroughs since Q3 2025"
2 gaps detected · 2026 error-correction · new photonic demos
crawl ibm.com/quantum/2026-roadmap
crawl arxiv.org/abs/2609.04128
crawl nature.com/articles/s41586-2026-...
reading 8 prx bundles
analyzing newly found content
generating synthesized report
Report ready· 237 claims · 612 sources · signed .prx
Optional: publish back so the next person inherits from you too
time
~1 min
cost
$0.50
tokens
sources
612

$60+ of community compute inherited. Pay only to extend. Publish back if you want.

Publish once, cite anywhere

Push .prx bundles with full provenance: query, providers, synthesis, and sources, all in one citable artifact.

A permanent link for every answer

Not a screenshot. A citable research artifact anyone can explore, fork, or reference.

Build on what others found

Pick up where another researcher left off, with full history and provenance intact. No starting from scratch.

Public or private, your call

Keep research to yourself or contribute to the commons. Publish when you're ready.

Agent-native

Built for agents, not just humans.

prxhub is llms.txt-native. AI agents can search existing bundles before running new queries, pull machine-readable claims with confidence scores, and publish findings back to the registry. All without an account, using the public MCP endpoint.

Read the full agent docs
{
  "mcpServers": {
    "prxhub": { "url": "https://prxhub.com/api/mcp" }
  }
}
search_bundlessearch_claimsdownload_bundle

Four paths to create a .prx bundle

A .prx bundle is a single archive that captures a research run end to end: the query you asked, the providers consulted, the synthesized answer, and every source cited along the way. Signed, portable, and citable. The format is open; read the spec on GitHub.

Pick whichever path fits where your research lives. Reading prxhub is free and anonymous. Publishing back is free too. You just need an account so bundles can be attributed.

01 / 04

Have your agent do it

Wire the prxhub MCP server into Claude Code, Cursor, Codex, or any MCP client. Your agent searches the registry before it crawls and publishes back what's missing.

Reading is anonymous. A free account is required to publish.

02 / 04

Use the open-source CLI

parallect runs multi-provider research with your keys, against your choice of providers (frontier or local). prx wraps the .prx format and the registry workflow. Both are MIT-licensed and yours to fork.

See the projects
03 / 04

Use Parallect.ai (SaaS)

Deep research as a service across the frontier providers, with one-click publish to prxhub. Lightest path if you'd rather not manage API keys, providers, or local infrastructure.

Try Parallect.ai
04 / 04

Implement the .prx spec

If you build a research tool, agent, or provider, we'd love for your output to be portable and citable. The format is open, the registry is public, and the reference implementation is on GitHub.

Read the spec