AI Features

AI Assistant

Every Project and Program has a dedicated AI Assistantaccessible from the "AI Assistant" tab. This assistant:

• Understands the full context of your project (efforts, wiki, forum history)
• Can answer questions about your research area
• Suggests relevant papers and approaches
• Helps with LaTeX formatting and proof structure
• Can summarize lengthy forum discussions

Personal AI Assistant

Visit /assistant for your Personal AI Assistant — a research companion that knows your interests and helps you stay current.

Interest Configuration

Configure your research interests (e.g., "algebraic topology", "analytic number theory") and the assistant will tailor its recommendations accordingly.

Google Scholar Integration

The assistant monitors Google Scholar for new papers matching your interests and alerts you to relevant publications.

Personalized Feed

Your assistant curates a personalized feed of new results, trending discussions, and relevant activity across Mathub based on your interests and project participation.

Plan Agent

The Plan Agent activates during project creation. When you describe a research problem, the Plan Agent:

1. Parses your natural-language description
2. Identifies the core mathematical questions
3. Searches Mathub for related programs and projects
4. Suggests MSC classification codes
5. Proposes an initial project structure

Init Agent

After the Plan Agent, the Init Agent performs automated project setup through a 7-stage pipeline:

1. Literature Crawl — Searches arXiv, Google Scholar, and Wikipedia
2. Key Result Extraction — Identifies theorems, lemmas, and definitions
3. Effort Generation — Creates initial Efforts
4. Dependency Mapping — Builds logical dependency graph between efforts
5. Wiki Population — Generates Wiki pages
6. Workspace Setup — Configures branches and milestones
7. Forum Initialization — Creates starter discussion threads

Patrol Agent

The Patrol Agent continuously monitors external sources for new developments relevant to your project:

Configure Patrol Frequency

Set how often the Patrol Agent checks for updates — daily, weekly, or custom intervals. Configure this in Project Settings → AI → Patrol.

External Crawlers

The Patrol Agent uses three external crawlers:

• arXiv — Monitors new preprints matching your project's topic
• Wikipedia — Tracks changes to relevant mathematical articles
• Google Scholar — Discovers newly published papers and citations

Auto-Apply Results

When the Patrol Agent finds relevant new material, it can automatically:

• Create new reference efforts in the Workspace
• Update Wiki pages with new information
• Post summaries in the Forum

Wiki Verification

The AI can verify the mathematical claims made in Wiki pages. It checks:

• Correctness of stated theorems against known results
• Validity of referenced sources
• Internal consistency across pages
• Completeness of definitions and assumptions

Verification results appear as annotations on the wiki page — green for verified, yellow for uncertain, red for potential issues.

AI Rewrite

Select any text in a wiki page, effort document, or forum post and click "AI Rewrite" to:

• Improve clarity and readability
• Formalize informal arguments
• Expand proof sketches into detailed proofs
• Simplify overly technical language

Wiki Sync

Projects can enable Wiki Sync to automatically keep wiki pages up to date as new efforts are verified and research progresses. Configure this in Project Settings → AI → Wiki Sync.

• Schedule — Set how often wiki sync runs (e.g., daily, weekly)
• Effort-Wiki Sync — Automatically generates or updates wiki pages when efforts change status (e.g., a verified proof updates the corresponding wiki theorem page)
• Forum-Wiki Sync — Key insights from forum discussions can be surfaced and merged into relevant wiki pages

Rebuild

Re-run the Init Agent pipeline on an existing project or program to refresh literature references, regenerate wiki content, and update effort structures based on new external sources. Available from Project Settings → AI → Rebuild.

LLM Usage Dashboard

Platform administrators can monitor AI token consumption across the platform via the Admin → LLM Usage dashboard. It provides:

• Token usage breakdown by project and time range (24h, 7d, 30d)
• Per-agent usage (Plan Agent, Init Agent, Patrol Agent, AI Assistant)
• Cost tracking and budgeting tools