refhub

// structured_research_memory for humans, tools, and agents

refhub is a modern reference management platform for organizing academic publications, building citation networks, and sharing research collections.

it is built for the command-line generation: browser-first, keyboard-friendly, dark by default, structured enough for agents and scripts, and blunt about the difference between user workflows and automation workflows.

refhub is now a multi-repo system with a clear split between product, api, cli, extension, and agent integration work.

• refhub.io is the frontend product: vaults, papers, profiles, settings, help, and api-key management
• .netlify is the backend/api layer: /api/v1, scoped keys, semantic scholar routes, pdf/item routes, import/export surfaces
• refhub-cli and refhub-skill are the automation layer for humans, scripts, and agents
• refhub-extensions handles browser capture
• refhub-marketplace packages first-party claude code plugin distribution

near-term work is focused on cleaner vault workflows, stronger permission semantics, better import/export/search, and safer agentic operations.

refhub is not a prettier paper list. the target is a research memory system with explicit structure:

• vault-based publication management
• public and private research collections
• tags, metadata, relations, profiles, and settings
• pdf and google drive attachment workflows
• versioned api access for trusted automation
• cli, browser extension, and agent skill surfaces

in plain english: usable in the browser, programmable without regret.

install the refhub skill once — works across tools.

claude code

claude plugin marketplace add refhub-io/refhub-marketplace
claude plugin install refhub-skill@refhub-marketplace

gemini cli · opencode · codex cli

# gemini
mkdir -p ~/.gemini/skills/refhub-skill && curl -o ~/.gemini/skills/refhub-skill/SKILL.md \
  https://raw.githubusercontent.com/refhub-io/refhub-skill/main/SKILL.md

# opencode
mkdir -p ~/.config/opencode/skills/refhub-skill && curl -o ~/.config/opencode/skills/refhub-skill/SKILL.md \
  https://raw.githubusercontent.com/refhub-io/refhub-skill/main/SKILL.md

# codex
mkdir -p ~/.codex/skills/refhub-skill && curl -o ~/.codex/skills/refhub-skill/SKILL.md \
  https://raw.githubusercontent.com/refhub-io/refhub-skill/main/SKILL.md

cursor · windsurf · others

curl -O https://raw.githubusercontent.com/refhub-io/refhub-skill/main/AGENTS.md

if a concept matters, it should exist explicitly in the model and api.

tags, relations, permissions, vault roles, scopes, and item metadata should be real things — not hidden conventions.

agents and scripts should get narrow, inspectable access:

• scoped api keys
• vault restrictions
• honest errors
• auditability
• eventually dry-runs, safer bulk operations, and sync surfaces

normal user auth is for people. api keys are for runtime automation. account setup and connected services stay in the web app.

refhub is for researchers and builders who are comfortable with dense interfaces, keyboards, and actual structure.

• dark by default
• keyboard-first
• monospace where data matters
• // as the signature pattern
• low on marketing glaze
• closer to a good terminal tool than a lifestyle app

keep the split clean:

• frontend concerns belong in refhub.io
• backend/api concerns belong in .netlify
• cli ergonomics belong in refhub-cli
• extension concerns belong in refhub-extensions
• integration abstractions should follow the real backend, not outrun it

good contributions improve data structure, permission clarity, workflow efficiency, integration safety, or research usability.