A research agent’s weakest point is often not finding information, but losing track of what it read, why a claim is supported, and which gaps remain. This project treats research as a resumable, auditable process: evidence persists across sessions, and the final report must pass checks before it ships.
What Sets It Apart
- A staged workflow rather than a single prompt. The pipeline moves from query decomposition and source gathering through synthesis, critics, gap-filling, and citation checks. This makes omissions and disagreements easier to surface than in a one-shot answer.
- A vault that compounds. Sources and notes are stored in searchable Markdown and SQLite, so later runs can reuse prior research instead of starting from scratch.
- Constrained revision. After synthesis, the workflow favors small edits over regenerating the report. Quote integrity, retraction checks, and citation-to-claim review add further gates, though they cannot guarantee that every conclusion is correct.
Great Fit If…
You want a coding agent to conduct substantial, source-backed research that can be resumed and revisited, and you are willing to trade speed and model usage for a documented process. Look elsewhere if you need a quick answer, a turnkey general-purpose research app, or factual guarantees. Full runs can take hours; the Claude Code workflow is the benchmarked one, while the newer Codex port has weaker enforcement for some editing constraints and has not been benchmarked.
