Knowledge Infrastructure

Knowledge Engine for AI Agents

Agents aren't searching for information. They're trying to complete work. Subtext transforms enterprise data into trusted, task-ready knowledge that AI agents can use directly—without endless retrieval loops, hallucinations, or token waste.

Faster task completion

Lower token costs

Higher task success rates

Get Started Read the API

The Pitch

Your agents shouldn't have to figure it out themselves.

Today's RAG systems return chunks of arguable relevance and force the calling agent to retrieve, re-read, retrieve again, and synthesize from scratch — roughly 85% of an agent's effort is spent on retrieval.

Subtext flips this. We compile your raw sources into a navigable wiki, and our agent navigates it for you. One call. One cited, confidence-scored answer. Built for the contact-center agent that needs a policy oracle — not a search index.

Typical RAG

8retrievals / answer

Subtext

1call / answer

Citations

verbatimper response

How it works

From a folder of documents to a queryable knowledge layer.

Step 01

Add sources

Drop PDFs, DOCX, URLs, plain text, or a ZIP of documents. We extract and store the raw material.

Step 02

Compile

Claude organizes your sources into a structured wiki with cross-linked articles, backlinks, and an auto-generated index.

Step 03

Query

Humans chat via the UI. Agents call /v1/answer for a single-call, cited, confidence-scored answer.

Surfaces

One knowledge layer. Multiple access patterns.

Compiled wiki

Claude curates your sources into structured articles, complete with [[wikilinks]] and a deterministic index. Article quality stays high as your corpus grows.

Agentic chat

Sonnet 4.6 navigates your wiki via search_articles, list_articles, and read_article tools. Tool calls are visible to your users — every answer is auditable.

Agent API

POST /v1/answer returns a synthesized answer with verbatim citations, a confidence band, and an explicit escalation signal. Built for AI agents that need policy oracles.

Sample wikis

Every new account gets curated sample wikis — try the product without uploading anything. Read, search, and chat instantly.

Built for agents

One call. Cited. Confidence-scored.

The Agent API is a stateless oracle. POST a question, get back a structured answer your downstream agent can quote, cite, or escalate from — no second LLM call required.

Request

curl -X POST http://www.getsubtext.tech//api/v1/answer \
  -H "Authorization: Bearer lmn_live_..." \
  -H "Content-Type: application/json" \
  -d '{
    "question": "What is our refund window after 30 days?",
    "context": { "customer_tier": "gold" },
    "trace_id": "cc-session-abc"
  }'

Response

{
  "answer": "Standard refunds are not granted after the 30-day
            window. Gold-tier customers are eligible for a
            one-time courtesy refund within 60 days; see
            policies/refunds.md §3.",
  "confidence": 0.82,
  "confidence_band": "high",
  "should_escalate": false,
  "escalation_reason": null,
  "citations": [
    {
      "path": "policies/refunds.md",
      "title": "Refund Policy",
      "last_updated": "2026-05-12T10:00:00Z",
      "quote": "After 30 calendar days from purchase, refunds
              are denied except in cases of...",
      "relevance": 0.91
    }
  ],
  "latency_ms": 1840,
  "model": "claude-haiku-4-5"
}