Normal agent chat
Good answer, weak memory
Agent summarizes papers in chat. Useful in moment, but claims, gaps, and uncertainty are hard to inspect after thread moves on.
Research-Pilot
See what your research agent understands.
Research-Pilot turns agent research into a read-only project understanding page: questions, claims, sources, gaps, recent changes, and next moves.
Research-Pilot
Project understanding stays visible
Agent runs leave source-backed beliefs, changed claims, open gaps, and promptable next moves on a static project page.
Public Proof Artifact
A real project understanding graph from an agent research workspace
This page embeds the Research Browser workspace island captured from the DemoVisualAffordance project. The project topic is only the sample; the product is the durable graph of what the agent understands.
Research Browser Workspace
Understanding, literature, and experiments in one read-only graph
What This Proves
Agent work can leave an inspectable state
The graph is not a mock illustration. It is a static public rendering of Research-Pilot workspace data from a local agent research project.
The demo topic is replaceable
DemoVisualAffordance supplies realistic content. Research-Pilot is the layer that records and renders project understanding across sessions.
Preserved State
What Research-Pilot keeps visible
orientation
Project questions
The page keeps the research frame visible, so a new agent run starts from the same project intent.
understanding
Claims and evidence pressure
Claims stay attached to support, weak links, limitations, and changed status instead of disappearing into chat.
sources
Literature routes
Papers remain organized by why they matter to the project, not as a raw bibliography dump.
next work
Experiment handles
Candidate evaluations and controls sit next to the claims they can strengthen or weaken.
uncertainty
Open gaps
The graph preserves what the agent still does not know, giving the next prompt a target.
Graph Modes
How to read the workspace
Understanding
Questions, claims, evidence, and limitations
This is the project argument layer: what the agent currently believes and where that belief is weak.
Literature
Paper lineage without replacing the project graph
This layer shows source routes and paper clusters that support, pressure, or contextualize the understanding graph.
Experiments
Evaluation settings and proposed checks
This layer keeps experiments tied to project uncertainty instead of turning them into a detached task board.
Positioning
What this page is not
- Not a paper manager: sources matter because they change project understanding.
- Not a writing pipeline: it records what the agent knows before prose begins.
- Not an internal workflow dashboard: the public surface is read-only and inspectable.
- Not a chat transcript: old turns become durable claims, gaps, and next moves.
Latest Agent Run
From agent work to human-readable change
Agent task: read, compared, and reasoned over project material. Research-Pilot preserved the parts that matter for the next session.
Before
A claim existed as a plausible agent inference.
After
The graph shows support, uncertainty, and the next evidence need.
Changes from agent run
- A broad claim was narrowed after the agent compared it against source evidence.
- A vague uncertainty became a specific evidence gap visible in the graph.
- A next prompt was created from the weakest part of the current understanding.
How Research-Pilot Works
Simple chain
- Chat with agent
- agent reads / compares / reasons
- Research-Pilot records important understanding
- read-only page updates
- user asks a better next prompt
Quick Start
Run Research-Pilot locally
curl -fsSL https://raw.githubusercontent.com/QZhang2111/Research-Pilot/main/install.sh | bash