Before execution, outline your title, idea, background, direction, goals, constraints, and must-read papers. No implicit black-box prompting.
Upload local PDFs and provide remote paper URLs simultaneously. We enrich metadata (DOI, venue, authors) and automatically chunk/embed for retrieval.
The system drafts an experiment and execution plan first. It must pass an explicit human approval gate before moving to sandbox run.
Attach a local repo path or Git URL alongside setup/run commands. The sandbox executes real-world workloads, tracks timeouts, and collects artifacts.
Monitor the real-time progress of 15+ sub-stages via WebSocket updates, pause, resume, or rollback when an approval gate is reached.
Export true Markdown, LaTeX, BibTeX files, compiled PDF manuscripts, and a full zipped delivery bundle containing all research products.
Set your OpenAI/Codex endpoints, keys, and preferred models right from the web GUI.
Input your research hypothesis, compute budget, API constraints, and artifact expectations.
Import paper links, PDFs and perform live discovery. The engine parses them contextually.
Review the system-generated strategy. Reject or approve to unblock the execution pipeline.
Watch stages transition from pending to running, inspect internal metadata, or rollback on failure.
Gather the evaluated experiment code and download the final LaTeX manuscript with peer-review rubrics.
Your research context belongs to you. Auto Research Pro Max is built with local isolation natively in mind.
DB state, plans, running outputs, and artifacts are all stored in the local SQLite database.
If no API key is provided, the UI functions perfectly using deterministic fallback text.
Uploaded papers are saved strictly to the local disk. No opaque third-party cloud hosting.
All data egress is strictly confined to user-configured APIs and endpoint URIs you explicitly defined.
Everything you need to orchestrate autonomous research, complete with human-in-the-loop gates so the system does the work but you make the calls.