ResearchPilot for VS Code

Institutional-grade, multi-agent AI research, document RAG retrieval, and empirical fact-checking directly inside Visual Studio Code.

1. Product Overview

ResearchPilot connects your editor directly to an autonomous, multi-agent research pipeline designed to tackle complex technical, scientific, and empirical questions without leaving your code or workspace.

Unlike conventional conversational chatbots that hallucinate or rely on ungrounded pre-training memory, ResearchPilot decomposes your inquiry across specialized agents that plan sub-questions, gather external web evidence via Tavily, retrieve internal workspace PDF knowledge via pgvector semantic search, normalize cross-source findings, audit claims with a dedicated fact-checker, and synthesize rigorous, cited reports.

2. Key Features

Activity Bar & Dedicated Sidebar:
- Always-accessible research hub in your activity bar.
- Submit research queries, monitor live milestone streams, and inspect past research history.
Multi-Agent Orchestration:
- Real-time execution telemetry streamed from 5 cooperative agents:
  1. Planner Agent: Decomposes high-level questions into granular investigation hypotheses.
  2. Web Research Agent: Queries live web sources using Tavily Search API.
  3. Document Research Agent: Performs semantic cosine similarity retrieval over ingested PDF embeddings.
  4. Evidence Normalizer: Reconciles dates, units, claim targets, and duplicate citations.
  5. Fact Checker & Report Writer: Cross-audits claims against source quotes and computes confidence scores.
Fact-Checked Claims Audit & Contradictions:
- Tabulated claim audits categorizing each assertion as SUPPORTED, PARTIALLY_SUPPORTED, CONTRADICTORY, or UNSUPPORTED.
- Dedicated discrepancies section highlighting conflicting claims across academic or industry sources.
Context Menu "Research Selected Text":
- Highlight any function, architectural block, or paragraph in any editor, right-click, and select Research Selected Text with explicit user confirmation.
Export to Workspace Markdown:
- Export completed research briefs, complete with citations, metadata, and verdict tables, directly to your workspace as structured .md files.
Secure Token Storage:
- Uses VS Code's OS-native SecretStorage keychain. JWT tokens and credentials are never written to plain-text settings, logs, or webview DOMs.
Quota & Spending Transparency:
- Direct feedback on tier limits, daily submissions, hourly ceilings, and HTTP 429 Retry-After timeframes.

3. Installation & Getting Started

Installation

Open Visual Studio Code.
Open the Extensions view (Ctrl+Shift+X or Cmd+Shift+X).
Search for ResearchPilot.
Click Install.

Quick Start

Click the ResearchPilot icon in the VS Code Activity Bar (left sidebar).
If you do not have an active session, click Log In in the sidebar.
Enter your account email and password.
Type your research inquiry into the query box (e.g., "Compare zero-knowledge proofs vs. optimistic rollups for L2 scalability").
Select your desired depth (Quick, Standard, or Deep) and click Launch Research.
Follow the live agent stream as each milestone completes, then click View Report to open the rich report viewer.

ResearchPilot authenticates against your configured backend instance using standard JWT credentials:

Press Ctrl+Shift+P (or Cmd+Shift+P) to open the Command Palette.
Run ResearchPilot: Log In to Platform.
Enter your registered email address (e.g., admin@researchpilot.local or your organization email).
Enter your account password.
Upon successful authentication, your access and refresh tokens are securely encrypted in VS Code SecretStorage.
To log out at any time, run ResearchPilot: Log Out.

Tokens are automatically rotated in the background when access tokens expire.

5. Research Workflow

   [ Editor / Selected Text / Sidebar Prompt ]
                       │
                       ▼
            [ 1. Planner Agent ]
         Decomposes inquiry & scope
                       │
         ┌─────────────┴─────────────┐
         ▼                           ▼
[ 2. Web Agent (Tavily) ]  [ 3. Doc Agent (pgvector) ]
 Live external search       Internal document RAG
         │                           │
         └─────────────┬─────────────┘
                       ▼
         [ 4. Evidence Normalizer ]
           Aligns facts & metrics
                       │
                       ▼
          [ 5. Fact-Checker Agent ]
          Validates claim verdicts
                       │
                       ▼
       [ 6. Synthesized Report Viewer ]
         Interactive audit & Markdown export

6. Backend Connection Settings

The extension is pre-configured to connect directly to the official Render cloud API service. You can customize the backend target if hosting your own instance:

Open Settings (Ctrl+, or Cmd+,), search for ResearchPilot, or edit settings.json:

Setting	Type	Default	Description
`researchpilot.backendUrl`	`string`	`https://researchpilot-api-cb0c.onrender.com`	FastAPI backend URL. Can be set to `http://localhost:8000` for local Docker setups.
`researchpilot.defaultDepth`	`enum`	`standard`	Default investigation depth (`quick`, `standard`, `deep`).
`researchpilot.enableWeb`	`boolean`	`true`	Enable live external web search via Tavily.
`researchpilot.enableDocuments`	`boolean`	`true`	Enable internal document semantic search via pgvector.

ResearchPilot is designed around strict developer privacy and zero silent transmission:

No Silent Telemetry: The extension never scans, monitors, or uploads your workspace files or code in the background.
Explicit Confirmation for Selected Text: When triggering Research Selected Text from the editor context menu, the extension displays a confirmation dialog previewing the selected characters before any network call is dispatched.
Credential Isolation: No API keys (such as Google Gemini or Tavily keys) are ever distributed to or stored inside the extension client. All third-party provider calls execute exclusively on the authenticated backend.
Webview Isolation: All webview panels enforce strict Content Security Policies (default-src 'none'), disallowing arbitrary remote script execution or external style injection.

8. Current Limitations & Free Tier Notice

Free-Tier Sleeping: When connecting to a free Render deployment, the backend web service spins down after 15 minutes of inactivity. The initial request after sleeping may incur a cold-start delay of 50–90 seconds while the container boots.
In-Memory Job Lifecycle on Free Tier: Because free cloud tiers cycle containers periodically, deep research tasks requiring extended runtimes should be initiated with quick or standard depth to complete within 2–4 minutes.
Ephemeral Uploaded Files: On free container tiers without persistent disks, physical PDF binaries are cleared upon server restart, though text chunks and vector embeddings remain permanently queryable in PostgreSQL.

9. Available Commands

Access these commands at any time via the Command Palette (Ctrl+Shift+P / Cmd+Shift+P):

ResearchPilot: Start New Research: Opens prompt to start a research job.
ResearchPilot: Research Selected Text: Researches highlighted editor text with confirmation.
ResearchPilot: Save Active Report as Markdown: Saves the currently viewed report as a .md document in the workspace.
ResearchPilot: Log In to Platform: Prompts for credentials and logs into the backend.
ResearchPilot: Log Out: Clears securely stored tokens.
ResearchPilot: View Quotas & Usage: Displays current rate limits, monthly quotas, and tier allocations.
ResearchPilot: Refresh Research History: Syncs recent research jobs in the sidebar.

10. Support & Feedback

Issues & Bug Reports: GitHub Issues
Documentation & Deployment: See docs/RENDER_FREE_DEPLOYMENT_GUIDE.md in the project repository.
Platform Web Interface: ResearchPilot Web

ResearchPilot

Gowri Mahesh