Agent-native Research Knowledge Infrastructure
Let agents see how the literature connects.
Zhiway is not another research-agent application. We build agent-native research knowledge infrastructure and knowledge commercialization capabilities for model providers and research institutions. We focus on the stage before experiments begin. The technology community is working to close dry- and wet-lab loops; but if AI lets one laboratory pursue a hundred directions at once, the scarce decision will no longer be how to run experiments, but which hundred routes deserve to begin.
Start experiments.
Stop betting blind.
Idea Network and Idea Interflowing are live, closing the loop between knowledge environment and research decisions
online research tools designed for low reasoning, retrieval, and context cost
Research FDE on site across NLP, materials physics, and materials engineering laboratories
a North China innovation institution and a science-and-engineering university provide pilot and cooperation entry points
Thesis / What is missing
Large models can read papers, but cannot naturally preserve citation structure, field evolution, or research state. What is missing is a trusted, traceable, stateful knowledge environment that updates faster than model training cycles.
Organize authors, concepts, methods, citations, and temporal evolution so an agent can see where a study came from and where it may go next.
Drill from directional judgments into relationships, abstracts, and source evidence, while separating graph-supported, graph-unseen, and graph-blind results.
Maintain search history, rejected paths, human corrections, and next-step suggestions outside the model, without repeatedly consuming context to re-prove old conclusions.
Product Matrix
Idea Network provides the knowledge environment; Idea Interflowing produces pre-experiment decisions; Research Claw runs inside institutions; Research FDE adapts the system into workflows that can be accepted and procured.
Agent-native research knowledge environment
Defines papers, authors, methods, concepts, citation relationships, and research state, then exposes them through MCP. It helps agents see where a study came from, how the field branched, and which boundaries remain unanswered.
Pre-experiment research decision workspace
Turns Idea Network's relationships into inspectable research judgments. Creator identifies structural gaps worth pursuing; Supervisor evaluates novelty, rigor, and reproducibility against a target venue and an explicit evidence path.
Local institutional research-agent environment
Connects an institution's own data, tools, permissions, and workflows inside a local agent environment. Research Claw is the product base; FDE enters the field to identify what each discipline actually needs to preserve and adapt.
From field demand to a procurement boundary
FDE is not a fourth software product. It is Zhiway's delivery team inside the lab: reconstructing real workflows, defining data and knowledge boundaries, configuring the Loop / Harness, and proving acceptance criteria on real tasks.
Journey
A research service first proved demand. SaaS and local-agent products then exposed the same structural bottleneck: research agents need a trusted, stateful knowledge environment. That earned insight became Idea Network and now enters laboratories through Research FDE.
Idea Network × Idea Interflowing / Live
Idea Network is a research knowledge environment agents can call directly for literature relationships, source evidence, and stateful research sessions. Idea Interflowing is the pre-experiment research-decision SaaS built on it, organizing research directions, evidence paths, not-found results, and blind spots into an inspectable delivery chain.
Locate the "voids" in the citation network — structural gaps no one has reached — and let agents work out which research question each void corresponds to.
Identify a field's research frontier, trace how it has advanced over time, and answer "where is the next breakthrough most likely to appear."
Find inspiration along real academic relationships and generate candidate hypotheses where fields intersect — each traceable back to the original literature.
Start from a research direction and follow connected papers, concepts and citations to find structural gaps worth pursuing. Separate established evidence, refutable hypotheses and open validation work, then deliver a structured research proposal that can be reviewed and executed.
Review a paper or completed body of work against target-venue dimensions and expose the evidence path behind every judgment. Explicitly separate graph-confirmed evidence, graph-not-found results and graph blind spots so each score can be challenged and refined.
Science of Science / Trajectory Flywheel
Science of Science has traditionally reconstructed discovery from papers, citations and patents after publication. In the agent era, every human acceptance, rejection, rewrite and redirection lets us observe how a direction forms while research is still happening.
Ontologies, citation relations and temporal evolution show where a direction came from, how it branched, and where current evidence ends.
What an agent queried is only the beginning. Why a researcher accepted, rejected, pursued or abandoned a route is the scarce data behind research judgment.
A model can internalize the skill of open-book research. New papers, evolving relations, not-found boundaries and team judgment trajectories cannot be permanently absorbed by a single training run.
Idea Network organizes published knowledge; Creator and Supervisor enter topic selection, proposal formation and evidence review. Together they begin to reveal not only what science has published, but how scientific innovation forms.
With research intent and institutional data protected, MCP calls and human corrections create continuously updated trajectory signals that can improve attention models, gap ranking and boundary judgments.
Field Validation / Research FDE
Products prove capability online; FDE proves value in the field. We do not begin with “buy an AI platform.” We begin with one task that actually happens every day in a lab, then define data boundaries, acceptance metrics, buyer, and budget.
Map agents, literature sources, specialist tools, roles, and critical decision points.
List open and non-open journals, private data, permissions, and local security requirements.
Build domain ontology and graphs; connect literature sources, specialist tools, and judgment workflows.
Run real tasks and define acceptance metrics, buyer, delivery boundary, and budget.
Idea Network and Idea Interflowing are live; Research Claw is running in pilot labs; Research FDE is already on site in three disciplines.
Team
First-hand research experience, multi-agent engineering, and access to institutional and international markets — a team holding all three at once is itself a scarce asset.
Founder · CEO
CSO · Chief Strategy Officer
Co-Founder · Head of Engineering
Co-Founder · Head of Operations
Head of University–Industry–Research Partnerships
Head of Investment & Financing Partnerships
With Zhiway's assistance, papers completed and authored by the researchers themselves have been accepted at IJCAI, EMNLP, and AAAI — 3 in total, with a 4th under review. Blind peer review at top conferences is the field's most demanding third-party test.
Use a knowledge environment to support pre-experiment decisions, then bring it into institutions through Research Claw and Research FDE so every judgment has evidence and every redirection preserves state.
Start experiments without betting blind.
Insights / Methods
How Zhiway defines research knowledge environments, positions Idea Network among existing technical layers, and brings AI4S into institutions through Research FDE.
Why relations, evidence, state and interfaces must exist outside the model.
From finding content to research lineage, evidence boundaries and stateful exploration.
Turning general capability into real tasks an institution can accept and procure.
Contact
Whether you're a researcher, an institution, or simply someone who cares about AI for Science — we'd love to hear from you.