Explore your organizational knowledge, investigate complex questions, and review answers alongside the evidence that supports them.
Knowledge workspace3 sample documents
Documents are processed in this browser. If AI analysis is enabled, relevant excerpts are sent for inference. Upload only documents you are authorized to process. PDF text extraction requires an internet connection.
ASK THE INVESTIGATION AGENT
Keyword search ready · semantic search available
To generate a cited answer, selected document excerpts are processed by the AI service. For sensitive information, leave AI off and use browser-only search.
Checking AI backend…
Ready
Investigation results & evidenceAwaiting query
START HERE
Inspect the evidence behind your answer.
Run an investigation to see each retrieval step and its source evidence. Enable AI-Powered Investigation for a generated answer with citations, or continue with browser-only evidence retrieval.
Generate a report after your investigation
Document library
Knowledge vault
Store searchable text and semantic vectors in your private knowledge database. Access requires your workspace key.
Investigate stored knowledge
Research Follow-ThroughEvidence screening · Local preview
From future-work statements to inspectable evidence
Screen a future-work statement against candidate passages from later papers. This deterministic preview does not invoke an AI model; every proposed conclusion requires source verification and human review.
Import research papers
Upload one seed paper and up to 12 citing/follow-up papers. Text is extracted locally in your browser; PDFs remain in your browser by default. Only selected text excerpts are sent for AI processing after explicit consent.
PDFs must contain selectable text. Scanned PDFs without a text layer need OCR, which is not included. A selected file is not proof of a verified citation relationship.
No research PDFs selected yet.
AI-assisted future-work extraction
Optional: analyze selected pages of the seed paper with Llama 3.1. Requires your private workspace key in Document Library. The selected page text will be sent to the AI model only when you click the button below. This consumes AI inference quota.
No model requests made
Local lexical candidate search · no LLM requests
FUTURE-WORK ITEM FROM A SEED PAPER
AI evidence investigation
Investigate an individual future-work item against selected passages from your uploaded citing/follow-up papers using hybrid retrieval and an LLM. A provisional conclusion and source excerpts are provided for human review.
No AI investigation has been requested.
Maximum 28 excerpts per run. Search includes lexical BM25, dense semantic embeddings and RRF. Only submitted excerpts are considered; citing relationships and factual entailment require independent validation.
CANDIDATE EVIDENCE FROM CITING PAPERS
Separate passages with a line containing only --- . Add one passage per citing-paper excerpt and inspect the original source before accepting a status.
Retrieve evidence from your knowledge vault
Search stored document passages as evidence candidates. This read-only lexical search does not call an AI model. Enter the private workspace key in Document Library first.
Research limitation: keyword/action-cue screening is not evidence entailment, and absence of evidence in these passages does not establish non-fulfillment. Seed/citing relationship verification and independent scientific evaluation remain future work. AI findings are provisional and require human source review.
Evaluation Lab Measured locally
Retrieval benchmark
Explore retrieval performance on a labeled example collection. Results reflect the selected documents and evaluation questions.
Technology
Built for traceable knowledge discovery.
EvidenceFlow combines evidence retrieval, document analysis and AI-assisted investigations in one workspace.
Document intelligence
Import supported documents, browse extracted passages, and organize findings with source references.
Hybrid retrieval
BM25 keyword search with optional semantic embedding retrieval and reciprocal rank fusion.
AI investigations
Model-guided search planning and evidence-grounded responses with source references when AI analysis is enabled.
Evaluation
Inspect retrieval metrics and investigation traces to understand how results are obtained.
AI outputs should be independently verified against their cited sources before consequential decisions. Private server-side document storage is available only with configured workspace access and connected knowledge services.