Corpus-first
Account-based corpus management
Build a named corpus from published packs on your account page, load it in the app, and Research mode activates over that specific data.
Docs / Research Mode
Research mode turns loaded data into a
bounded workspace.
Research is not a second copy of Explore with a bigger model. It is a different lane: a bounded workspace that reasons over a defined packs corpus, not the full catalog.
Boundary
Corpus-bound, not catalog-bound
Research answers from the active corpus. If the needed data is missing from the corpus, it says so. That keeps the reasoning honest and the results auditable.
Direction
Workspace, not chat clone
The deeper workspace layer — stronger memory, bounded artifact tools, notes, and analytical continuity — is in active development.
What Research mode is for
Research mode exists for comparison, synthesis, trend analysis, anomaly detection, explanation, and deeper follow-up on data the user already assembled. The job is analytical reasoning over a bounded corpus, not general source discovery and not routine map loading.
How it works
Corpus management lives on the account page. The user selects published packs, names the corpus, and saves it. Opening the app and loading that corpus activates Research mode over that specific data.
Workflow
- 1. Build a corpus. Pick published packs from your account page, name the corpus, and save it.
- 2. Load it in the app. Open the app, select your saved corpus, and Research mode activates over that data.
- 3. Ask analytical questions. Comparison, trends, cross-domain queries, and export all run against the corpus you defined.
- 4. Missing data stays explicit. If the corpus is missing something, Research names the gap rather than guessing.
What “corpus-bound” means
Research mode should not behave like a broad catalog-aware assistant. The central rule is that the manifest tells Research what exists, and the tools tell Research what the data says. That keeps the prompt smaller, the reasoning more honest, and the answers easier to trust.
Why this matters
This split gives the product a cleaner shape. Explore is the dispatcher and loader. Research is the analyst. Over time that makes Research feel like a real workbench instead of “the same chat, but more expensive.”
What is next
The deeper workspace layer is in active development: stronger memory, better artifact-query tools, and the ability to reopen and continue a prior analytical session. The account page is already the management surface for corpora; richer workspace continuity builds on top of that foundation.