Science/Retrace

The description states that Science/Retrace is a local workspace built with GPT-5.6 and subagents (Luna medium), offering five distinct modes:

• Chat: For project-aware discussion and inspection, keeping files read-only.
• Learn: Turns up to ten local source files into items for questions, presentations, and quizzes.
• Review: Runs structured reviews of PDF, DOCX, TXT, or LaTeX documents, producing findings, annotations, and revised outputs.
• Deep Research: Creates editable LaTeX source and compiled PDF reports on a topic.
• Auto Research: Allows users to define objectives, budgets, permissions, workflows, and stopping conditions for agentic research loops.

It is built using Codex CLI and integrates tools like Git, GitHub, Playwright, Python, Node.js, SQLite, and others. The author notes that no manual coding was done; everything was built with the help of GPT-5.6 and subagents.

The author positions Science/Retrace as an alternative to traditional AI research tools that rely solely on chat interfaces, which they claim are inconvenient for serious work due to lack of traceability and file management. The goal is to provide a guided approach that is accessible to non-coders but still supports advanced features like LaTeX integration.

They describe it as:

• A local workspace
• Approachable for non-coders
• With tools like LaTeX contained inside
• Designed for learning, reviewing, researching, experimenting, and writing with retraceable evidence

This positioning implies a shift from generic chat-based AI tools toward more structured, reproducible research environments.

The description does not clearly identify a specific customer segment or ideal customer profile (ICP). However, it suggests:

• Users who engage in serious AI research
• Individuals who need to learn from source material, review papers, or develop research documents
• People working with LaTeX, PDFs, and structured workflows

There is no indication of whether the tool targets students, researchers, academics, or professionals.

No information about pricing, monetization strategy, or business model is provided in the description. The project appears to be a hackathon submission with no commercial intent evident.

The system is built using:

• GPT-5.6 and Luna subagents
• Codex CLI
• Tools like Git, GitHub, Playwright, Python, Node.js, SQLite, LaTeX, Markdown, JSON-RPC, YAML, TOML, CSS3, HTML5, JavaScript
• A package manager called Pixi

The author claims:

• No manual coding was involved.
• The system uses loops and automation via Codex and GPT-5.6.
• It supports multiple output formats (LaTeX, PDF, etc.)

There is no evidence of traction, adoption, or user engagement beyond the developer’s own testing. The project was submitted to a hackathon and described as being larger than initially anticipated. It has not yet reached a production-ready state.

The author does not reference any competitors or existing solutions in the AI research space. The project is positioned as an alternative to chat-first tools, but no comparison with other platforms or tools is made.

1. No real-world usage: The tool appears to exist only in prototype form and lacks any evidence of actual use by others.
2. Unverified claims: All functionality is self-reported without external validation.
3. Lack of commercial clarity: No pricing, monetization, or business model described.
4. Dependency on unproven AI agents: Reliance on GPT-5.6 and subagents implies potential instability or unreliability in execution.
5. No team or structure: The team size is listed as zero, indicating no formal development team.

1. What specific research tasks have you used this tool for? Can you walk us through a concrete example?
2. How does the system handle data privacy and security, especially when working with local files?
3. Are there any known limitations or edge cases in how the different modes interact?
4. Have you tested the tool across different operating systems beyond Ubuntu?
5. What are your plans for scaling beyond the current prototype stage?