v0.8.0

SynthOS v0.8.0

A major overhaul of the page architecture, a full theming system, 28 external service connectors, agent integrations,
and a new brainstorm mode — all focused on making app creation more powerful while keeping the experience
conversational.

https://www.npmjs.com/package/synthos

Page Renderer v2

The entire page rendering pipeline has been redesigned. Instead of full HTML rewrites, the LLM now returns surgical
JSON change operations (update, replace, delete, insert, style) applied via node-ID diffing. Pages load shared shell
code (theme CSS, helpers, runtime script) injected by the server, keeping individual page files lean. A repair pass
auto-fixes any failed operations, and inline script deduplication prevents bloat over multiple edits.

Pages with the old v1 format are blocked from serving and can be upgraded in-place via the Pages Gallery — the
original files are backed up before migration.

Themes

Two built-in themes ship with this release:

• Nebula Dusk — dark synthwave (deep blue/purple, purple accent glows)
• Nebula Dawn — light pastel counterpart

Themes provide CSS variables (--accent-primary, --bg-secondary, etc.) and a window.themeInfo runtime object so every
page adapts automatically. Theme files are versioned and auto-upgrade on startup when the package ships newer CSS.

Connectors — 28 External Service Integrations

Apps can now call external APIs without handling auth. A full proxy system lets pages use synthos.connectors.call() —
auth credentials are injected server-side and never exposed to the browser.

Supported services: Airtable, Alpha Vantage, Brave Search, Cloudinary, DeepL, ElevenLabs, Giphy, GitHub, Hugging Face,
Imgur, Instagram (OAuth2), Jira, Mapbox, NASA, NewsAPI, Notion, Open Exchange Rates, OpenWeatherMap, Pexels, Resend,
RSS2JSON, SendGrid, Spoonacular, Stability AI, Twilio, Unsplash, Wolfram Alpha, YouTube Data.

Configured connectors are automatically described to the LLM with endpoint hints and usage patterns, so it knows how
to wire them into generated pages.

Agent Support (A2A + OpenClaw)

Two agent protocols are supported:

• A2A (Agent-to-Agent) — standard HTTP-based protocol with streaming via SSE
• OpenClaw — WebSocket-based persistent connections with session management and optional SSH tunneling

Enabled agents auto-connect on startup and are exposed to the LLM so it can wire synthos.agents.* calls into pages.
The page API provides send(), sendStream(), chat.* (history, abort, clear), and capability discovery.

Brainstorm Mode

A built-in ideation assistant accessible from every page via a lightbulb icon next to the chat input. It opens a
full-screen modal for multi-turn brainstorming with the AI, generating clickable suggestion chips at each turn. When
ready, "Build It" copies a clean, build-ready prompt into the chat input.

Table Storage API

Apps get built-in local JSON-file storage, scoped per page. The REST API (/api/data/:page/:table) supports CRUD
operations with pagination. Pages use synthos.data.* helpers that auto-scope by page name — no cross-page data
leakage.

Page Download & Import

Pages can now be exported as .zip files and imported into other SynthOS instances. Import handles name conflicts with
auto-incrementing suffixes and includes path-traversal protection.

Save Line

A visible "Saved" checkpoint marker is injected into the chat timeline when you save a page. It's protected from LLM
modification via the data-locked attribute, giving you a clear visual history of when saves occurred.

Additional Improvements

• Dual-model settings — configure separate models (and providers) for page building vs. in-app AI calls and brainstorm
• data-locked attribute — protect any element from LLM modification
• showInAll flag — hide pages from the "All" filter in the Pages Gallery
• Page modes — locked pages hide the chat form and switch Save/Reset to Copy/Reload
• Save & error modals — proper modal UI with keyboard support replacing browser prompts
• Page copy — duplicate any page (user, required, or legacy) with new metadata
• Settings v2 — restructured, versioned settings with auto-migration from v1
• Full-viewer mode — games and animations can use all available viewport space
• CI pipeline — GitHub Actions workflow with test coverage reporting via Coveralls