January
Core engine Shipped
HTML parser (html5ever), semantic tree builder, Trust Shield with 40+ injection patterns, WASM build. First proof that goal-aware extraction works: HTML in, ranked nodes out.
February
Agent capabilities Shipped
Intent API (find_and_click, fill_form, extract_data), HTTP fetch with cookies and robots.txt, semantic DOM diffing (80-95% token savings), goal compiler with topological sort. HTTP API server and Python/Node bindings.
March
Intelligence layer Shipped
QuickJS sandbox with full DOM bridge (200+ APIs), Arena DOM in Rust, SSR hydration for 10 frameworks (Next.js, Nuxt, SvelteKit, RSC, Qwik), event loop (setTimeout, Promises, MutationObserver), YOLOv8 vision, session management, workflow orchestration. QuickJS optimized to 2µs per eval (215x faster).
March 29
CRFR published Shipped
Causal Resonance Field Retrieval — the core algorithm that makes Slaash possible. 97.8% recall, 99.2% token reduction, 14ms cold start. No neural networks, no GPU. Learns from feedback. Validated on 50 live websites across 10 domains.
April — now
Public alpha Live
Landing page, /try playground, MCP server with 13 unified tools. Anyone can try Slaash on any URL right now. Deployed on Render with persistent CRFR memory (SQLite). Currently working on broad_mode for abstract queries — a new CRFR mode that auto-detects interpretive questions (tone, arguments, philosophy) and returns wider context for LLM analysis, with mandatory feedback to bootstrap learning on these harder cases.
Known limitations
What we're honest about
Abstract queries are hard. CRFR excels at factual extraction — prices, dates, statistics, specific entities — where keyword matching and structural signals are strong. But interpretive questions like "what is the tone of this article?" or "summarize the philosophical argument" lack specific keywords to anchor on. We've shipped broad_mode as a first step: wider retrieval + LLM feedback + causal learning. The system gets better with use, but the first query on a new abstract topic will return more noise than a factual query. We're actively improving this.
May
Developer experience Next
Public API with authentication and rate limiting. Developer documentation and quickstart guides. npm and pip packages for direct integration. Usage dashboard with per-query analytics and CRFR learning curves.
June
DOM engine maturity
Push WPT compliance from 91% to 95%+ on core DOM, 60%+ on CSS selectors. iframe support with per-origin trust levels. Improved SPA coverage for React 19, Next.js 15, and Vue 3 hydration patterns.
July–August
Enterprise readiness
Self-hosted deployment option (single binary, zero dependencies). Team accounts with shared CRFR memory across agents. SLA guarantees on uptime and latency. SOC 2 preparation. Volume pricing for high-throughput customers.
September
Ecosystem integrations
First-class plugins for LangChain, CrewAI, AutoGPT, and other agent frameworks. WebSocket streaming API for real-time extraction. A2A protocol support when spec stabilizes. Webhook notifications for CRFR learning milestones.
October–December
Slaash 1.0
Full SPA coverage (target: 98%+ WPT core DOM). CRFR v2 with cross-domain transfer learning — knowledge from one site improves extraction on similar sites automatically. The perception layer that makes every AI agent better at reading the web.