Skip to navigation
Platform

Agentic search

View as Markdown

Conduit’s agentic search enables organizations to search across millions of documents, emails, images, and other materials. Every result is cited back to the specific document, page, and portion.

Supported data types

FormatExtensionsNotes
Documents.pdf, .docx, .docFull text extraction with layout preservation
Spreadsheets.xlsx, .csvCell-level indexing and cross-reference
Presentations.pptxSlide content and speaker notes
Images.png, .jpg, .tiffOCR with confidence scoring
Web content.html, .mhtmlRendered content extraction
Email archives.pst, .mboxThread reconstruction and attachment indexing

Search modes

Natural language queries matched against document meaning, not just keywords:

curl -X POST https://api.conductorai.com/v1/search \
  -H "Authorization: Bearer $CONDUIT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "export control policies for dual-use technologies",
    "mode": "semantic",
    "limit": 50
  }'

Multi-step search that decomposes complex queries into sub-tasks, retrieves evidence from multiple sources, and synthesizes results with full citations:

curl -X POST https://api.conductorai.com/v1/search \
  -H "Authorization: Bearer $CONDUIT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "identify out-of-date policies and suggest revisions",
    "mode": "agentic",
    "collections": ["policy-library", "regulatory-updates"]
  }'

Citation tracking

Every search result includes a citation chain linking back to the source:

{
  "results": [
    {
      "content": "Section 4.2.1 restricts export of...",
      "citation": {
        "document": "ITAR-Policy-2025.pdf",
        "page": 47,
        "paragraph": 3,
        "confidence": 0.97
      }
    }
  ]
}