How is Idea Network different from search and RAG?

Search and RAG mainly answer what content to retrieve. Idea Network also organizes how published work connects and evolves, where the current evidence boundary lies, and what state should persist across research turns.

Why can this environment not be fully internalized by a model?

A model can learn how to search and compare, but new papers, citation relations, unknown boundaries and human judgment trajectories keep changing. Keeping those dynamic facts and states outside the model allows the environment to update faster than a new training cycle.

Are Idea Network and Idea Interflowing the same product?

No. Idea Network is the agent-callable knowledge environment. Idea Interflowing is the pre-experiment research-decision SaaS built on top of it, including Creator and Supervisor.

Can I use Idea Network now?

Yes. Idea Network is live at https://wentor.ai/network.

Home/Idea Network

Knowledge Environment · MCP · Stateful Research

Idea NetworkResearch knowledge environment

Let agents see how the literature connects.Start experiments without betting blind.

Idea Network is not another literature-search interface. It is an agent-native research knowledge environment that uses ontology to organize papers, authors, concepts, citation relations and temporal change, connects them to source evidence and research state, and exposes those capabilities to tool-using research agents through MCP.

Open Idea Network ↗ See the technical layers →

RelationsPapers, authors, concepts and citations

EvidenceSource evidence and provenance paths

StateResearch sessions and exploration trajectories

What It Organizes

Search retrieves content.
Idea Network organizes how the work connects.

Research agents need more than a longer list of papers. They need an external environment they can traverse, inspect for provenance, and resume without losing the paths already explored.

01 / RELATION

Relations and evolution

Organize papers, authors, concepts, methods and citations as a structure that can be followed through time.

02 / EVIDENCE

Evidence and provenance

Move from a directional judgment to relations, abstracts and source evidence, with a path back to each source.

03 / STATE

State and boundaries

Keep search history, excluded paths and next-step suggestions outside the model while separating graph-confirmed, not-found and blind-spot states.

04 / INTERFACE

Agent-callable tools

Expose research capabilities through MCP in a form that matches agent tool use and reduces irrelevant context load.

Agent-Native

Let agents continue along an evidence path

A static knowledge graph expresses entities and relations. Idea Network adds ontology, continuous ingestion, research sessions and tool interfaces, so an agent can start from a question, disclose evidence progressively, and attach the next exploration turn to the state of the last one.

Drill into papers, authors, concepts and source evidence
Disclose details progressively instead of flooding model context
Preserve exclusions and pivots as a resumable research process

Science of Science / Trajectory Flywheel

A model can learn how to use an open book. It cannot memorize a book that keeps changing.

New papers, relation changes, not-found boundaries and team judgments evolve faster than another model-training cycle. Idea Network keeps dynamic facts, research state and evidence paths outside the model, so agents can work against a current research map.

When researchers accept, reject, rewrite, continue or pivot, those human-directed trajectories become Science of Science feedback signals that can refine attention, gap ranking and research-frontier judgments.

Agents expand evidence, compare paths and maintain state. Researchers choose directions, inspect evidence and make the final decision to begin an experiment.

Research questionDirection and constraints

Agent explorationSearch and relation expansion

Evidence pathSources, unknowns and blind spots

Human judgmentAccept, reject or pivot

Research stateFeeds the next call

Product Relationship

One knowledge environment,
two real research workflows.

Idea Network: Infrastructure core. Organizes literature structure, source evidence and research state, and exposes them to agents through MCP.
Idea Interflowing: Pre-experiment research-decision SaaS. Creator forms structured research proposals; Supervisor reviews papers or completed work and exposes the provenance behind each judgment.
Research Claw: Local institutional research-agent environment. Connects private data, models, tools and Idea Network while preserving process and organizational knowledge.
Research FDE: Field adaptation and delivery. Builds domain ontology, integrates tools, validates real tasks and defines procurement boundaries inside research institutions.

FAQ

About Idea Network

What is Idea Network?

Idea Network is an agent-native research knowledge environment. It organizes papers, authors, concepts, citations, source evidence and research state, and exposes them to tool-using agents through MCP.

How is it different from search, RAG and a knowledge graph?

Search and RAG answer what to retrieve; a knowledge graph represents entities and relations. Idea Network additionally organizes how work connects and evolves, where the evidence boundary lies, and what research state persists across turns.

Why not train all of this into the model?

A model can learn efficient retrieval and comparison, but new papers, citation relations, unknown boundaries and human judgment trajectories keep changing. An external environment updates faster and keeps provenance inspectable.

How does Idea Network relate to Idea Interflowing?

Idea Network is the knowledge infrastructure. Idea Interflowing is the research-decision SaaS built on it. Creator and Supervisor show how the environment enters direction finding, proposal formation and evidence review.

Can I use it now?

Yes. Idea Network is live at wentor.ai/network.