SaberTools — the research platform

SaberTools is an AI-powered research intelligence platform. It researches any topic using Claude plus 18 live public APIs, organizes the results into a structured, queryable knowledge base, analyzes sources deeply, and exports everything as portable UKS packets.

The platform is the producer in the ecosystem: it turns a research question into graded, licensed, structured knowledge that conforms to the open UKS standard — and that can be published to the Registry.

➡️ The app lives at saber.tools. This page explains what it does.

What it does

Capability	Description
Research agent	Claude web search + domain-routed live API calls. Depth: Quick / Medium / Deep.
18 live connectors	Health, bibliographic, scientific, government, finance, and open-web sources — auto-selected per domain. See Connectors.
Source management	Every source carries an evidence grade, clinical status, credibility score, summary, key findings, tags, and license. Filter + full-text search.
Deep analysis	Claude batch-extracts structured domain data — dosing, mechanisms, interactions, lab markers, and more — accumulated on each source.
Gap analysis	Claude identifies coverage gaps, missing topics, and priority follow-up searches.
Knowledge graph	Force-directed, interactive visualization of entities and relationships.
Page scraper	Jina Reader (primary) / direct fetch (fallback) for arbitrary URLs.
Import	Paste any Markdown/document → Claude extracts the sources it mentions.
Export	UKS v3.0 JSON — full KB, curated KB, analyzed entities, or downstream-app formats.

Evidence is first-class

SaberTools never collapses what matters. Every source preserves:

• Evidence grade — GRADE-aligned (high → very_low / n_a).
• Clinical status — validated, emerging, contested, commercial, informational, n_a (never a boolean).
• License — a license_label and a rights_url on every record, never collapsed to a flag.

This is the same vocabulary the UKS standard enforces, so anything you research is publishable and portable by construction.

From research to a UKS packet

1. Create a project with a domain (health, bibliographic, finance, …).
2. Research a topic — the agent routes to the right connectors and grades what it finds.
3. Analyze & curate — deep analysis, gap analysis, bookmark to the knowledge base.
4. Export as a UKS packet → validate it, serve it over MCP, or publish it to the registry.

Under the hood

• Frontend: React + Vite.
• Backend: Node/Express + Prisma + PostgreSQL (pgvector) + Redis + BullMQ.
• Two auth systems: JWT for the web app (/api/*); scoped API keys (st_live_*) for external apps (/service/v1/*). See the API reference.
• AI provider keys stay server-side — never exposed to the browser.

→ Next: the 18 connectors · the UKS standard · the API