The table ranked models on tool-calling correctness alone, which recommended
Kimi-K3 at $3/$15 per 1M tokens. The chat panel runs up to 50 agent steps and
resends tool schemas each step, so price compounds and that was the wrong
default.
Add a cost column, move the recommendation to Kimi-K2.7-Code (same clean
behaviour, ~1/4 the price), and note that gpt-oss-120b is ~25x cheaper than
K3 and would be the obvious default once the vLLM shim exists.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Whether a provider works in the web build depends on CORS headers being set
on the response (not just the preflight), and whether a model is usable in
the AI chat panel depends on its streaming tool calls being well-formed.
Neither is documented by providers, so record what has been measured.
Covers 11 providers and 5 models, each row dated. Documents two recurring
failure modes: vLLM misrouting the final tool-call argument chunk to
index + 1 without an id, and reasoning models returning null content under
the connection test's maxOutputTokens: 1.
Includes copy-paste curl recipes so others can add rows.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- Move variables roadmap into the VitePress development docs
- Replace the stale renderer profiler plan with current usage notes
- Refresh eval scripting docs with the supported API surface
- Remove the stale root PLAN document
- Move the variables UI roadmap under docs/development
- Update contributor guidance to point at the development roadmap
- Merge add-vitepress-docs: docs site with 6 locales, SEO, user guide
- Remove VitePress cache files that slipped into git
- Fix all 11 oxlint warnings: replace non-null assertions in
use-collab.ts with local const captures inside closures
- Update changelog with docs and lint fix entries
Profiler modules in packages/core/src/profiler/:
- FrameStats: rolling 120-frame buffer tracking FPS, CPU/GPU times, node counts
- GpuTimer: EXT_disjoint_timer_query_webgl2 wrapper for GPU-side timing
- DrawCallCounter: WebGL context proxy counting draw calls per frame
- PhaseTimer: Chrome DevTools Performance panel custom tracks via User Timing API
- HudRenderer: in-canvas overlay with stats text and frame time bar graph
- CaptureStack: per-node profiling with speedscope JSON export
- RenderProfiler: orchestrates all layers with zero cost when disabled
Integration:
- SkiaRenderer.render() instrumented with phase markers for scene, picture
recording/replay, section titles, component labels, selection, rulers, flush
- Node counting and viewport culling tracking in renderNode()
- WebGL2 context passed through for GPU timer and draw call counter
- Shift+P keyboard shortcut to toggle HUD overlay
- toggleProfiler() added to editor store
New docs/user-guide/ section covering: canvas navigation, selection &
manipulation, drawing shapes, text editing, pen tool, layers & pages,
context menu, exporting, auto-layout, components, and variables.
Each article includes Mac+Win/Linux shortcuts tables, cross-platform
note, cross-links between related articles, and consistent structure.
VitePress sidebar updated with User Guide as first group (12 entries).
