litell
If AI is going to learn from the real world, its output has to be traceable to a source. litell is where that capability is trained — a research engine where every answer carries its provenance.
The problem.
Literature search ranks by keyword and recency, not by whether a paper actually answers your question. You get a list, then do the synthesis yourself — one abstract at a time.
And the quality signals that matter — citations, journal standing, whether the idea is even novel — live in separate tools you check by hand. None of it is traceable by default.
The approach.
litell treats search as a decision, not a lookup — and makes every output carry its source.
- Signal-aware search — ranks by citations and quality, auto-broadens when recall is thin
- A cross-paper brief that answers the question, not one point per paper
- Prior-art check — is this idea actually new?
- Journal quality inline, with a direct link to check JCR impact factor
- Exposed as an MCP server — any agent can search, brief, and check prior art directly
This is the discipline the firewall sells as a product: AI output must be traceable. litell's provenance trail is what the firewall's data-source proof cards are built on.
Where it stands.
litell is live and shipping weekly. The web app runs, and the MCP server is registered locally so agents call it as a tool.