grabowski
1a1b62d475
Deploy web Studio / deploy (push) Successful in 3m3s
✦ Agent in the Studio's top bar opens a floating chat window. The model (any OpenAI-compatible endpoint with tool calling; default B4L's inference server, https://inference.p0.contact/v1) gets the 15 MCP tools plus two page tools (select_object, set_view) and the Studio's state (selection, level, rotation) in its system prompt. Answers and reasoning stream in; tool calls show as rows with arguments and results; object ids link to the object; Stop ends a turn; Undo takes the agent's edits back. The window moves by its title bar and resizes at its top-left corner. - studio/agent.js: shared by both front ends (waits for studio.js's isokit:studio event); MCP over /mcp or isokit.mcp in the web Studio - isokit studio: /api/agent/{config,models,chat} relay the calls on their own threads (ureq, chunked and flushed per read), so the key stays on the machine and the endpoint needs no CORS; settings from ISOKIT_LLM_URL / _KEY / _MODEL or agent.json in the user's config folder - web Studio: calls the endpoint from the page with a key kept in the browser; build_web.py --agent-url names a same-site relay as default; the deploy builds with /llm/v1, which deploy/Caddyfile forwards to the inference server (its nginx refuses the CORS preflight) - tools/mock_llm.py + tools/agent_checks.py: the agent window checks in both check_studio.py and check_web.py
5.2 KiB
5.2 KiB
Roadmap
Each phase ends with tests, a Studio/viewer check in Edge, commit and push.
Phase 1: engine, Studio, MCP (done)
- Core: projection with 4 rotations, primitives (box, prism, cylinder, frustum, face, poly, flat, openings, lines, disc, circle, stem, figures, puffs, fan, raw, groups, sequences), sprites with 90° object frames, depth sort ported from the JS views, SVG writer, tags, tooltips
- Theme: built-in palette, materials and rules converted from the Mooban/CNX CSS; TOML overrides
- Rhai runtime: drawing API, module imports from memory, errors with file:line, safety limits
- Prefab library: site, area, road, parking, wall, fence, lamp, marker, tree, tree row, hedge, building (storeys, windows, door, flat/gable roof, solar, AC, cut-open levels), carport, shed, car (driving), truck, motorbike, person, crowd, tank, container, generator, solar field, chimney
- Scene document + prefab catalog with property schemas
- Project folder, ops (add/update/remove/duplicate/reorder/set_cfg/replace), Workspace (save, undo, reload on outside changes), check, static HTML page, SVG export
- Viewer JS: port of the views' mount() (pan, pinch, wheel, declutter, tooltips, full screen) for prerendered, server and future WASM sources
- Studio GUI: palette, object list, placing, drag-to-move with snap, keyboard, forms from schemas, levels, undo/redo, live follow of agent edits, MCP connection panel
- MCP: 15 tools over stdio and HTTP (
/mcp), validation messages for agents - Demo project, 52 Rust tests, Playwright Studio + MCP test
Web Studio (done)
isokit-wasm: the engine in the browser (wasm-bindgen, 2.8 MB / 0.75 MB gzipped);Storetrait (folder or memory) under theWorkspace; Studio JSON API shared withisokit studiostudio.jsruns on a pluggable backend; the web shell answers it from WASM- Projects in IndexedDB; new, demo copy, delete; export / import
.zip(stored or deflated) and folders (drop or pick); two tabs on one project stay in step - Folder projects (File System Access, Chrome / Edge): edits written to disk, outside changes
followed live and undoable, Build page writes
dist/index.html - Code editor for scripts, prefabs, theme and manifest, with script errors by file and line
- MCP tools inside the page (
isokit.mcp), WebMCP registration when the browser offers it tools/build_web.py(static output),tools/check_web.py(44 checks, Edge)- Deploy a hosted copy; render in a Web Worker for very large scenes
Agent window (done)
- ✦ Agent in the Studio: chat with a model on an OpenAI-compatible endpoint (default: B4L's inference server) that edits the scene with the MCP tools + page tools (select, view); streamed answers and reasoning, tool calls with results, clickable object ids, Stop, Undo
isokit studiorelays the calls (/api/agent/*, key stays on the machine, no CORS); settings fromISOKIT_LLM_*oragent.jsonin the user's config folder- Web Studio calls the endpoint from the page; isokit.buildfor.life relays
/llm/to the inference server (Caddy) so the page needs no CORS from it tools/mock_llm.py+tools/agent_checks.pyin both Studio checks- Keep the conversation per project across reloads; attach a rendered view (PNG, Phase 4) for vision models
Phase 2: richer editing
- Interior prefabs (desk, table with chairs, shelf, machine, bed, sofa, partition walls) and a
building
roomsprop, so the GF/1F levels show something - Path objects in the Studio: click points for areas, roads (polyline with width), fences, walls, pipe runs; vertex handles
- Multi-select (box select, Shift-click), group move, align/distribute
- Copy/paste between projects; prefab thumbnails in the palette (rendered icons)
- Measure tool and grid overlay; numeric resize handles on the selection
- Studio editor for
theme.toml(colour pickers) and for project parameters
Phase 3: embedding in the other projects
- Host pages on the WASM build (
isokit-wasmexists, see Web Studio): the same scene recomputes in the browser when a host page's calculator changes the parameters (cnx_vinyl fleet calculator) isokit build --fragment:<style>+<script>blocks for the Python build scripts of the host pages (replaceinline.py)- Host API parity with CNXIso (
mount,update(cfg),setLevel,focus,info,onPick) - Port the existing scenes: mooban estate, CNX Scuba centre (section cut), CNX Vinyl plant; check each against its old view with the QA checklist from the skill
Phase 4: output and QA
- Section cut layer (underground scenes) from iso-scuba: cut plane, clip-path per cavity
- PNG export (static style resolution + resvg) so agents get an image from
render_view isokit checkrules from the QA checklist: props on paved areas, runs ending in the air, covers drawn over their contents, tag overlaps- Performance budget test (render ms per view, 60 fps pan) on a 3,500-node scene
- Packaged desktop app (Tauri); shared projects for the hosted web Studio (behind login)