# Onto — AI is reading your website wrong. > Onto serves clean Markdown to AI agents instead of HTML noise. Same content, right format. One line of code. No changes to your site. **Source:** /roadmap **Extracted:** 2026-05-21T20:21:00.655Z --- System evolution ## Where we're _going._ From a free scanning tool to the compatibility layer for the agent web. Six phases. One engine. Read today, Serve alongside, Act as the destination — each layer built on the same cleaning core, the same accuracy scoring, the same edge runtime. Here's how we get there. Roadmap · 6 phases · Act is where the vision lives * Phase 1 Feb 2026 Live ## AIO Score Calculator The entry point. A free, public tool that scores any website's AI-readiness in seconds. Enter a domain, get an instant report with React Tax analysis, semantic richness scoring, and content negotiation check. * Instant domain scanning * React Tax / CSS bloat analysis * Semantic richness scoring * WAF detection & security paradox diagnosis * Before/after payload visualisation * Email waitlist capture * Phase 2 May 2026 Live ## Read API + MCP Server The Layer 1 product. Any AI developer can call api.buildonto.dev with any URL and receive clean Markdown plus an AIO accuracy score — sub-100 ms on cache hits, single-digit-second cold fetches. No SDK needed on the source site. A Firecrawl alternative with confidence scoring and hallucination risk flagging no competitor has. MCP server wrapper integrates directly with Claude Code, Cursor, and any MCP-compatible client. * REST API with usage-based pricing tiers * MCP server in Anthropic's directory * 10× smaller payloads, typically * AIO accuracy score returned per request * Hallucination risk flagging * Sub-100ms cache hits at the edge * Works for any URL on the public internet * Phase 3 Mar 2026 Alpha ## @ontosdk/next SDK Open-source Next.js plugin that compiles your routes into clean Markdown at build time. npm install @ontosdk/next — one line in your next.config, zero code changes to your existing pages. * Build-time route extraction * Cheerio-based HTML → Markdown pipeline * Heading, table, code block preservation * JSON-LD and schema extraction * Pluggable extractors per route namespace * onto.config.ts customisation * Phase 4 Mar 2026 Alpha ## Edge Middleware The bot-aware traffic cop at the CDN edge. Inspects Accept headers and User-Agent strings, routes AI agents to .onto payloads, serves full HTML to browsers. Sub-15 ms overhead. * Accept header content negotiation * User-Agent bot detection (GPTBot, ClaudeBot, etc.) * Rewrite to pre-compiled .onto payloads * Zero overhead for human traffic * Graceful fallback to HTML on errors * Phase 5 May 2026 Coming soon ## Control Plane (Pro SaaS) Multi-tenant B2B dashboard for analytics, context injection, and policy management. See which AI agents crawl your site, what they access, and control what they see. The control plane that powers both Layer 1 API users tracking their usage and Layer 2 site owners managing agent traffic — same dashboard, two views. * Agent traffic analytics by User-Agent & IP * Per-route context injection for AI agents * Multi-domain management * API key management & usage analytics for Read API users * Trend visualisation & anomaly detection * Brand tone controls for AI narratives * ROI metrics for AI-driven traffic * Phase 6 · Layer 03 2027 The destination Where the vision lives. ## Act — agents through the same engine. When agents can read, serve, and act through one engine, the agent web stops being a research surface and becomes a transactional one. Today's web wasn't built for agents clicking buttons through semantic intent. Act changes that. Site owners opt in to make their product agent-interactable. Agents navigate any UI without breaking when the design changes. Built on the same cleaning engine and accuracy scoring that powers Read and Serve — which is why no one else is building this. They'd have to ship Layers 1 and 2 first, and we already did. > Same engine. More powerful interface. The agent web becomes _programmable_, not just readable. What ships * Semantic-intent API (`onto.act`) * Resilient to UI changes — no CSS selectors * Site-owner opt-in for action surfaces * Browser session orchestration at the edge * Multi-step intent chaining with context * Action receipts & audit log per session * MCP tool surface for any agent runtime * Native integration with Read & Serve outputs Why no one else can ship this * 01 #### Requires the cleaning engine Action depends on a faithful semantic representation of the page. That's Layer 1. Whoever ships Act without that ships brittle automation that breaks the moment a class name changes. * 02 #### Requires accuracy scoring Agents need to know what they can trust to act on. Without the scoring model we built for Read, every action is a coin flip. * 03 #### Requires protocol-level adoption Site owners only opt in to an action surface they already trust to serve. That's Layer 2. We'll have it shipped and installed before Act needs it. act.ts · 2027 \# Coming 2027 — same engine, action layer await onto.act({ intent: "buy pro plan", context: "checkout flow", receipt: true }) Capability matrix ### What's free. What's not. The core scanner, SDK, and edge middleware are open-source forever. The Read API is freemium — 1,000 free requests per month. The Control Plane is the only paid tier. Capability Open source Pro AIO Score Calculator @ontosdk/next Extractor Edge Middleware Read API · 1K free requests / month Read API · paid tiers MCP Server · 1K free requests / month MCP Server · paid tiers Agent traffic analytics Per-route context injection Multi-domain management Brand tone controls API key management Priority support For AI developers #### Building an agent? Start with the API. 1,000 free requests per month. No credit card. A Firecrawl alternative with accuracy scoring. [Read the API docs](/how-it-works) For site owners #### Get in early. Join the alpha program for @ontosdk/next SDK access and help shape how AI reads the web. [Get early access](/#waitlist) [Review tech specs](/how-it-works)