ACCESS MODEL
Three Stages for Reading Public Knowledge
Karon Web Access API combines a fast cache, lightweight requests, and browser fallback to produce content suitable for AI agents.
Supported Content
| Content | Recommended Endpoint | Status | Notes |
|---|---|---|---|
| Official documentation | /v1/agent/browse | Recommended | Best suited to technical documentation retrieval for AI agents. |
| Standards / RFCs | /v1/agent/browse | Recommended | Prefer content-focused markdown extraction. |
| Open-access papers | /v1/agent/browse | Varies by URL | Results may vary with site structure and PDF delivery method. |
| Technical references | /v1/fetch or /v1/agent/browse | Recommended | Use fetch for raw HTML and browse for LLM input. |
| Release note change detection | /v1/watch/snapshot, /v1/watch/diff | Recommended | Use these endpoints to track documentation changes over time. |
| Data behind a login | - | Not recommended | Not a primary target of the public web knowledge access API. |
Processing Path
Each request starts with the lightest path and advances to a heavier path only when needed.
Cache
Quickly reuses public pages that have already been processed.
- Cost: cache/curl 1
- Suitable for repeated requests
- Check response field cache_hit
Curl
Reads static pages and simple public documents with a lightweight request.
- Cost: cache/curl 1
- Low latency
- HTML or content extraction
Browser
Uses browser-based fallback for pages that require rendering.
- Cost: browser 10
- Supports wait_selector
- Can maintain session continuity
Browser fallback is powerful but takes more time and credits. For a new integration, start with the
/v1/agent/browse defaults and add wait_selector only for URLs that need it.AI Agent Perspective
URL input ->
select cache/curl/browser ->
clean content ->
return quality signals and cost_credits
Karon's central role is to read official documentation and trustworthy public sources, then organize them into a form AI can use instead of treating search results as authoritative.