FrontierLens

English | 简体中文

FrontierLens is a local AI research reader powered by the public Sciverse Paper Schema API. It helps a reader discover papers, follow a guided reading path, inspect extracted entities and relations, traverse resolved citations, and return structured evidence to the source text.

The application runs locally with a Python backend and a React frontend.

Product tour

1. Choose a research question

Start with a natural-language research question or choose a prepared example. FrontierLens turns it into a bounded retrieval scope grounded in the live Sciverse Paper Schema corpus.

2. Enter the research workflow

FrontierLens plans retrieval terms, finds seed papers, expands their extracted structure, assembles the topic graph, and generates a grounded reading guide. The required model receives structured Paper Schema material rather than complete papers.

3. Browse the paper-card reading path

Papers are organized as a reading path rather than a flat result list. Each card surfaces the title, contribution, and concise context needed to decide what to read next.

4. Inspect the graph network

Switch to the graph view to compare paper nodes, extracted Entity categories, paper-internal Relations, resolved paper-to-paper citations, and explained related-paper suggestions within the same bounded topic.

5. Expand a paper and continue through related work

Each paper opens with a concise overview and a card-based knowledge structure. Entity groups and relation lines expose methods, contributions, experimental setups, findings, measures, and document structure without requiring the reader to scan the full paper first. Beyond the single-paper guide, resolved citations and explained related-paper links connect the paper to other paper topologies for systematic, associative learning.

6. Trace structured knowledge back to the source

Selecting an Entity or Evidence item opens the ordered paper text, jumps to its provenance location, and highlights the supporting paragraph. DOI and arXiv links remain available when the metadata provides them.

All screenshots show live Sciverse results captured from a locally running FrontierLens instance. Results may change as the Paper Schema corpus evolves.

What it does

• Builds a bounded topic graph from keyword seed papers.
• Enriches seed papers with a balanced, bounded set of extracted Entity nodes and internal Relations, rendered in data-driven semantic clusters.
• Keeps paper-internal Entity relations separate from paper-to-paper citations.
• Shows the complete reference list, including unresolved references.
• Opens resolved citation targets by schema_id.
• Loads a complete paper-scoped Entity and Relation graph on demand.
• Reads paper text in ordered paragraph batches through the public provenance API, without direct OpenSearch access or local paper storage.
• Searches high-value Evidence and traces Entity/Evidence selections back to exact source paragraphs, with a labeled paper-local fallback when no anchor exists.
• Uses a clearly labeled, title- and author-verified related publication version when the selected formal publication has no paragraph rows.
• Shows real, cancellable progress for topic exploration and paper opening.
• Provides adjustable and collapsible paper-guide and source-reader panes.
• Rewrites questions into concise English retrieval phrases with the required locally configured OpenAI-compatible model.
• Produces a grounded research guide using only paper IDs already returned by Sciverse. A model failure stops the task and is reported to the user.
• Switches the interface and generated Guide/Roadmap between Chinese and English while retaining source paper, Entity, Evidence, and provenance text.
• Preserves the FrontierLens three-state journey:
  • discovery map and research-question entry;
  • Research Guide with Learning Roadmap / Knowledge Subgraph;
  • Paper Reading Guide with Knowledge Structure / Paper Internal Graph.

The visible corpus scope is the current 1M+ AI conference papers with completed Paper Schema extraction. An empty result never means that the work is absent from the full scholarly literature.

Boundary

• This is a standalone source tree with no dependency on ScholarWeave at runtime.
• The backend calls https://api.sciverse.space/paper-schema/*.
• The browser calls only this project's /api/* backend.
• The Sciverse key remains in the local backend.
• No OpenSearch, S3, SQLite graph, internal index name, or raw audit endpoint is used.
• No paper, Entity, Relation, Citation, Evidence, or graph data is read from a local folder. There is no fixture or demo-data runtime mode.

Requirements

• Python 3.11 or newer
• uv
• Node.js 22 or newer
• npm
• A Sciverse API key
• An OpenAI-compatible model endpoint, model name, and API key

Install and run

git clone https://github.com/Shannon4Science/sciverse-frontier-lens.git
cd sciverse-frontier-lens

make setup

Start the backend:

make backend

In a second terminal, start the frontend:

make frontend

Open http://127.0.0.1:3040.

A fresh installation starts unconfigured and shows no research data. Open Connection settings, configure, save, and enable both the Sciverse connection and the OpenAI-compatible model connection. Exploration remains paused until both connections are enabled. All papers and graph facts are fetched from the live Sciverse API.

No .env file is required for normal use. Configure the Sciverse connection and the required model connection entirely from the web settings page.

The Sciverse and model connections are managed from the same page and stored together as plaintext in local.config.json in the directory where FrontierLens is started. The file is ignored by Git and written with owner-only permissions on supported systems. It contains credentials and connection settings only, never papers or graph data. If either key is missing or disabled, the application identifies the missing connection and directs the user to Local Settings instead of starting an incomplete exploration.

OpenAI-compatible model endpoints may use HTTP or HTTPS in this standalone local application. Prefer HTTPS because plain HTTP does not encrypt the model secret or prompts in transit.

Optional environment configuration

Environment variables are only an optional interface for scripted, headless, development, or managed startup. Normal local users should use the web settings page and can skip this section. To provide startup defaults, create .env from .env.example:

SCIVERSE_API_TOKEN=<server-side token>

.env is ignored by Git. Do not put the token in frontend environment variables, URLs, screenshots, logs, or committed files.

The Sciverse key and model secret are separate required credentials. Both are normally configured from Local Settings; environment variables are startup defaults for managed or scripted environments.

Run the read-only production check:

make prod-smoke

See docs/USER_GUIDE.md for the complete workflow.

Verification

make setup
make lint
make test
make build
npm --prefix frontend audit --omit=dev --audit-level=high

Documentation

Detailed API behavior and operations are documented in docs/API_AND_OPERATIONS.md. Architecture decisions are in docs/ARCHITECTURE.md. The verified data and model boundary is recorded in docs/DATA_SUPPORT_AUDIT.md.

Security reports should follow SECURITY.md. Contributions are described in CONTRIBUTING.md.

License

Apache License 2.0. See LICENSE.