Reference
Changelog
Track major public SDK, validation, and docs/frontend milestones during the current pre-1.0 cycle.
Need the shortest path?
Use Quick Start when you want the smallest working route → scan → graph → run flow first.
Something feels off?
Inspect scan output, graph state, active route/context, and dispatch in a practical order.
Need the full surface?
Browse all runtime methods, hooks, extension points, and lifecycle APIs.
Pre-1.0
This changelog tracks major public-facing milestones for the current Cinder cycle.
0.2.x: hosted tier 3, tier 2 embeddings, and agent-native tooling
New packages
@cinder/relay- a hosted, metered tier-3 proxy: tier 1/tier 2 stay free, tier 3 is usage-based (prepaid wallet, real token-based cost metering), with a choice of who supplies the AI provider key per project. Backed by a real, persistent Postgres store, real account authentication, and a working dashboard UI at/relay. Razorpay payments exist but are not yet verified against a real account.@usecinder/cli-cinder integratecrawls a real running app and audits every page for tagging gaps;cinder install-skillinstalls a Claude Code Skill and anAGENTS.mdsection so a coding agent keeps up with new UI automatically.@usecinder/embeddings- tier 2: an optional, free, client-side embedding model that recovers typos/paraphrases of a page's routes/outputs/actions before any network call, plus semantic caching and alias-learning utilities.
SDK additions
Cinder.setLlmCore()/Cinder.resetLlmCore()- swap the single installed llm-core slotrun()escalates to, so tier 2, tier 3, or a fully custom core can take that slot without any other part of the pipeline changing.- Server-side speech transcription fallback (
speech.engine: "auto") for browsers with no Web Speech API at all, reusing the same fallback endpoint text tier-3 already uses. - Bounded tier-1 extensions: negation (clearing a filter) and numeric comparisons, gated on real field metadata rather than guessed.
exportGraphForEnrichment()- a developer-run, offline export that turns a real graph snapshot into something an LLM or coding agent can read and suggest alias/relationship improvements against, for human review before anything activates.- An agent-fetchable onboarding entry point (
/get-started/prompt.md) that a coding agent can fetch and run end to end.
0.1.x: runtime hardening and docs refresh
SDK maturity work
- graph scoping and duplicate-safe node identity improved
- tag-only validation isolated and proven explicitly
- fallback normalization made scope-aware for duplicates
- public extension callbacks made safer against user-code failure
- local graph reasoning strengthened
- domain-specific composite flow moved out of SDK core into an opt-in preset
- fallback transport gained timeout, classification, and telemetry
- mutation-driven rescans gained backpressure and timing metadata
- raw builder runtime state stopped guessing active route/context
- cross-realm DOM handling was hardened
- query candidate search gained indexes, active narrowing, and caching
- focused red-team validation suite completed
Docs and frontend work
- homepage and onboarding docs were rewritten around the real product wedge
- API reference now reflects the modern public SDK surface
- concepts and guides were refreshed for local-first, graph-aware behavior
- server and reference docs were updated for bounded fallback and current product scope
Website infrastructure
- Vite + React + TypeScript + Tailwind + MDX docs frontend remains the base
- navigation, search, sitemap generation, TOC, and metadata remain in place
- docs validation continues to check internal link integrity
Next public milestones
- deeper examples built around real app patterns
- a production-grade ledger for Relay (integer micro-cents) and real Razorpay verification
- an admin/analytics view beyond the current per-developer Relay dashboard
- optional docs performance and loading improvements
Continue reading