The previous bump to 8G (PR #153) didn't take effect on prod. Verified
post-deploy:
Cache size (GB): 2.0 / 2.0 (99.95%)
Cause: $CCACHE_DIR (/var/cache/ccache) is a named docker volume. Any
config we wrote into it via `ccache --set-config max_size 8G` during
the RUN step gets masked at runtime by the mount of the existing
volume — which still contained the original 2 GB config from when the
cache was first populated.
Fix: set CCACHE_MAXSIZE / CCACHE_COMPRESS / CCACHE_COMPRESSLEVEL as
ENV in the Dockerfile. ccache reads those at runtime and they
override any conf-file value, including whatever stale config is
sitting in the volume.
After this lands and the submodule bumps, `ccache -s` should report
the new 8 GB max immediately on container start without needing to
wipe or re-init the cache volume.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Empirical confirmation from a smoke test on prod after PR #152
(persistent build dir + dedup) landed:
Cache size (GB): 2.0 / 2.0 (99.96%)
Hits: 1 / 86886 ( 0.00%)
Misses: 86885 / 86886 (100.0%)
The cap was set to 2G when ccache was first wired in (PR #149) under
the assumption that base ESP-IDF would fit. With arduino-esp32 +
external Arduino libraries (Adafruit BMP280 + BusIO + Unified_Sensor +
GFX + …) the cacheable object set comfortably exceeds that — we
observed 86k misses while the cache was permanently at 99.96% full,
meaning entries get evicted before subsequent compiles can hit them.
Disk usage on the prod VPS at /var/cache/ccache will rise from ~2 GB
to ~6-8 GB. The host has plenty of headroom and the named volume sits
on the same partition as the rest of /var. Compression at level 6 is
already on, so the disk impact is the on-disk size after compression.
The 13-second warm compile we measured today already proves the bulk
of the perf win came from ninja's incremental cache + persistent build
dir; this is the cherry on top that lets ccache start contributing
across container restarts (when ninja's in-build-dir cache is fresh
but ccache should still hit at the C/C++ object level).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
InstallLibrariesModal — auto-install prompt that fires when an
example needs libraries:
- Title, subtitle (with the "Installing X of Y" progress
interpolation, the all-done success state, and the singular /
plural prompt explaining the requirement count).
- Per-row status badges (pending / installing… / installed /
error) plus the Wokwi-hosted-library tooltip.
- Footer buttons: Close / Skip / "Install All ({{count}})" with
loading variant.
SerialMonitor — multi-board tabbed serial console:
- Empty-state when no board is on the canvas.
- Right-side tab controls: Autoscroll checkbox, Clear button +
tooltip.
- The "(Open IoT Gateway)" inline link rendered next to detected
AP IP addresses.
- Output-area placeholders for the running-but-no-data and
before-start states.
- Send button + input placeholder (different copy for MicroPython
REPL vs raw Serial input).
- The line-ending dropdown options (None / Newline / Carriage
return / Both).
Hand-translated for all 9 locales. Hotkeys (Ctrl+C) and dropdown
values stay untranslated (constants the firmware reads).
Pending in Phase 3:
- Oscilloscope panel, custom-chip dialog, sensor control panel.
- ComponentPropertyDialog (per-component property forms).
- Admin / Profile / Project pages.
- Long-form docs prose (DocsPage 2715 lines, AboutPage long
paragraphs).
- 15 SEO landing pages (intentionally English for keyword targeting).
ComponentPickerModal — the "Add Component" dialog:
- Header (title + close button), search input placeholder + clear,
category tabs (All Components / Boards), loading + empty-state
copy, "Clear filters" button.
LibraryManagerModal — the Arduino library browser:
- Window title, Search / Installed tabs, filter input placeholder.
- Search-tab states: searching-for-query, generic loading, no
results (with optional query interpolation).
- Per-library row: "by {{author}}" caption, Install / Installing /
Uninstall / Uninstalling button labels.
- Installed-tab empty-state with the prompt to use the Search tab.
Brand and product names left untouched ("LIBRARY MANAGER" stays
all-caps in English; the localised variants follow each language's
convention for product UI titles). Hand-translated for all 9
locales.
InstallLibrariesModal still pending — it's the auto-install
prompt that fires when an example needs libraries; smaller scope
but lives in the same area.
The /examples gallery is fully localised:
- Header (heading + subtitle).
- Search input placeholder + aria-label + the clear button.
- Match-count tag with i18next pluralisation (handles _one /
_other and Russian's _few / _many).
- Category and Difficulty filter labels + their button labels
(basics / sensors / displays / communication / games / robotics
/ circuits; beginner / intermediate / advanced).
- Per-card "Copy shareable link" tooltip.
- Empty-state copy with two variants (with-search / without-
search) interpolating the search query.
- Reset-filters button.
- The library-install progress overlay copy from
ExamplesPage.tsx ("Installing libraries (N/M)") with done/total
interpolation.
Internal /editor link uses localize() so a Spanish reader who
clicks an example lands on /es/editor.
Hand-translated for all 8 non-English locales. Per-example titles
+ descriptions are NOT i18n yet — they live in the
src/data/examples* tables and would need a separate pipeline.
DocsPage (2715 lines of prose) deferred too — best handled by
running scripts/translate-i18n.mjs once the keys are extracted.
Three coordinated fixes that together close the "ESP-IDF compile takes
5-7 min every time" gap and prevent the failure mode where a user clicking
compile multiple times spawns six ninja processes that peel each other
apart on a modest VPS.
What was wrong
- /compile/start generated a fresh uuid4 every call, so 6 clicks = 6
independent builds racing each other. Saw load average 30 on the prod
VPS during a real BMP280 attempt today.
- No concurrency limit anywhere; asyncio.create_task() fired without
gating.
- ccache was wired in last week (PR #149) but reported 18,350 cacheable
calls and **0 hits** because the build dir was a fresh
tempfile.TemporaryDirectory(prefix='espidf_') per compile. The random
/tmp/espidf_<random>/ path baked into -I and -fmacro-prefix-map flags
→ different command line every compile → ccache hash miss every time.
What this PR does
1. Job deduplication (`backend/app/api/routes/compile.py`)
- New `_job_key(files, board_fqbn)` returns SHA-256 of normalised file
names + contents + board. Order-independent.
- New `JOB_BY_KEY: dict[str, str]` indexes hash → job_id.
- `compile_start` checks JOB_BY_KEY before spawning a new task; if a
job for this exact content is already pending or running, returns
the existing job_id (logs `[compile] dedup hit — reusing job <id>`).
- `_purge_expired_jobs` evicts both COMPILE_JOBS and JOB_BY_KEY,
keeping the index consistent. Edge case where two jobs share a key
(old finished, new running) is handled — only evict the key entry
if it still points at the purged job.
2. Concurrency control (`backend/app/api/routes/compile.py`)
- `_COMPILE_SEMAPHORE = asyncio.Semaphore(2)` global cap on
simultaneous compiles.
- `_target_lock(board_fqbn)` returns a per-target asyncio.Lock so
concurrent compiles to the SAME board (sharing the persistent build
dir) serialise. Different boards still run in parallel up to the
semaphore cap.
- `_compile_job` acquires sema → per-target lock → flips state to
`running` → calls `_run_compile`. Pending state now accurately
reflects "queued waiting for resources".
3. Persistent build dir (`backend/app/services/espidf_compiler.py`)
- New `_prepare_persistent_project_dir(idf_target)` materialises
`/var/lib/velxio-build/<target>/project/` from the template on
first use; on subsequent compiles it wipes only `main/` and
`user_libs/` (the per-compile parts) and leaves `build/` alone so
ninja's incremental cache + ccache .o files survive.
- Toolchain version sentinel (`.idf_version`) wipes the whole target
dir if the ESP-IDF or arduino-esp32 version changes — cached
objects from the old toolchain are no longer ABI-compatible.
- `compile()` is now a thin dispatcher: persistent path or fallback
to the legacy `tempfile.TemporaryDirectory()` flow. The actual
build logic was extracted into `_compile_in_dir()` so both paths
share one implementation, no duplication.
- Escape hatch: `VELXIO_PERSISTENT_BUILD_DIR=0` env var falls back
to the tempfile path without rebuilding the image. Critical for
production safety.
4. ccache normalisation (`Dockerfile.standalone`)
- + `ENV CCACHE_BASEDIR=/var/lib/velxio-build` makes ccache canonicalise
absolute paths under that prefix when computing the cache key.
Robustens hits against any future subdir rearrangement.
5. Docker compose (`docker-compose.yml`)
- + named volume `velxio-build:/var/lib/velxio-build` so the persistent
build dir survives `docker compose up -d --build`.
- + env `VELXIO_PERSISTENT_BUILD_DIR=1` (default ON; users disable
without rebuilding).
Expected impact
- Cold first compile per container per target: unchanged (~5-7 min).
- Same sketch re-compiled: ~2-5 s (everything cached).
- Different sketch, same target: ~5-30 s (only user code + new lib steps
rebuild; ESP-IDF base hits cache).
- Different sketch with new libraries: ~30-90 s (new lib component
compiles; rest hits cache).
- Concurrent clicks on same example: 1 build, others poll the same
job_id. No more six-ninja meltdown.
Tests
- `test/backend/unit/test_compile_dedup.py` covers `_job_key` stability +
variance and `_purge_expired_jobs` consistency (including the
"two jobs share a key" edge case).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The visible chrome of /about now reads from i18n in all 9 locales:
- Hero title + subtitle
- 7 section headings (Story / How It Works / Open Source Philosophy /
Creator / Recent releases / Community & Press / CTA)
- Final CTA card (title, subtitle, "Open Editor" button)
- Footer copy switched to t('footer.about') so it shows the AGPLv3
About-Velxio paragraph instead of the stale MIT/avr8js credit
- Footer + CTA Links wrapped in localize() so /es/about's "Open Editor"
routes to /es/editor
Long-form prose (Story body paragraphs, Open Source Philosophy
paragraphs, Creator bio, Releases blurbs, Personal-story quote, Press
section) deliberately stays in English in this commit. Each is a
multi-paragraph chunk that benefits from a curated translation pass
rather than an inline machine pass — slate it for a follow-up.
Tech-stack tags (Java, Python, React, Docker, etc.) and the creator's
name + role + GitHub/LinkedIn/Medium link captions stay untranslated:
all proper nouns / brand identifiers.
Both auth forms now go through t('auth.login.*') and
t('auth.register.*'):
- Title + subtitle, email / password / username labels with
username placeholder and password-min-length placeholder.
- Submit button toggles between idle and loading states.
- "or" divider, "Continue with Google" button, and the
switch-to-other-form footer link.
- Inline validation errors: reserved username, username regex,
password length, plus the generic catch-all from the API.
Internal /editor and /login / /register links wrapped in
localize() so a Spanish user who registers stays at /es/editor
after success.
AboutPage (519 lines) deferred to its own commit — too dense for
the same change.
The canvas header (the bar above the simulation area) and the
"Remove board?" confirmation dialog now read from i18n.
Translated:
- Status dot tooltip (Running / Stopped).
- Active board selector tooltip + "No board" placeholder + the
hint that prompts the user to add a board.
- Undo / Redo buttons: aria-label, dynamic title with the action
description and the empty-state fallback. Action descriptions
themselves stay untranslated (they come from the editor history
store as English literals — translating them would mean reaching
into a different store; deferred).
- Serial Monitor and Oscilloscope toggles (button title + label).
- Zoom in / out / reset-view buttons.
- Component count tooltip + Add Component button.
- The error-banner Dismiss button.
- "Remove board" item in the right-click menu, with a localised
"(N wires)" parenthetical via i18next pluralisation.
- The full removal confirmation dialog: title with board label
interpolation, body copy with optional connected-wires sentence,
Cancel + Remove buttons.
Pluralisation uses i18next's _one / _other (and _few / _many for
Russian) suffixes so wire counts read naturally per language.
Hand-translated for all 8 non-English locales. Untouched (deferred):
the property dialog, custom-chip dialog, sensor control panel, and
the various inline tooltips on board pins and wire endpoints —
those are denser and benefit from a separate pass.
The three modal dialogs that fire during the editor's routine save +
share + auth-required flows are now fully localised across all 9
locales.
LoginPromptModal
- Title, body, and the three buttons (Sign in / Create account /
Cancel). Sign in / Sign up Links use localize() so a Spanish
reader prompted to log in lands at /es/login rather than dropping
back to English.
SaveProjectModal
- Title (toggles between Save / Update), name + description fields
with placeholders, save button (toggles between Save / Update /
Saving…), Cancel button.
- Visibility toggle: Public / Private label + hint copy under the
icon.
- All four error paths now go through t() with a {{status}}
interpolation for the generic HTTP failure message.
ShareModal
- Title, public/private label + hint pair, "Make private" /
"Make public" toggle, Copy button, the warning shown when the
project is private, and the Close button.
Hand-curated translations for all 8 non-English locales. Status
codes (403) and shortcut markers preserved.
FileExplorer (sidebar)
- Workspace header label and the new-workspace / save-project icon
buttons now read from t('editor.fileExplorer.*').
- Per-board section: collapse / expand toggle, status dot tooltip
(Running / Compiled / Idle), per-board "new file" button, and the
composite "<board name> — click to edit" hover title (the board
name itself stays untranslated — it's a product noun like
"Arduino Uno").
- File rows: hover title with optional "(unsaved)" suffix,
unsaved-dot tooltip, and the right-click context menu's Rename /
Delete commands.
- Empty-state placeholder when no boards are on the canvas.
- The window.confirm() shown before deleting a file now reads from
t() too, so non-English users see the prompt in their language.
FileTabs (open-tabs strip above the editor)
- Per-tab close button title and the unsaved-changes dot tooltip.
- Inline confirm dialog when closing a modified file: prompt copy,
"Close anyway" and "Cancel" buttons.
Hand-translated for all 9 locales. Hotkey hints (Ctrl+S, Strg+S)
localised per German convention; other locales keep "Ctrl+S" as the
universally-recognised label.
The top toolbar of the editor is now fully localised — compile / run /
stop / reset, the compile-all / run-all variants when multiple boards
are open, the language-mode select, libraries / import / export /
upload-firmware actions, and the missing-library hint banner.
Strings live under editor.toolbar.* in src/i18n/locales/<locale>/common.json.
Hand-translated for all 8 non-English locales. Brand and product
names (Arduino, MicroPython, Ctrl+B, .hex / .bin / .elf / .ihex,
GitHub Sponsors) preserved as-is.
Editor strings still pending: file explorer, file tabs, simulator
canvas, component picker, library manager modal, save / share /
project modals.
Final block. The Support section ("Support the project" / GitHub
Sponsors / Donate via PayPal) now reads from t('landing.support.*'),
and the footer link labels use t('header.nav.*') so they pick up
the same nav translations the header already ships.
This closes the LandingPage rewrite — every visitor-facing string
on velxio.dev/ goes through i18n now. Hand-translated for all 9
locales. "GitHub Sponsors" stays untranslated (proper product name).
Phase 2 remaining work:
- Editor (toolbar, file explorer, simulator canvas, component
picker, library manager, save/share modals, error toasts).
- About / Examples / Docs / Profile pages.
- Login + Register forms.
The Features section ("Everything you need") and the 6 cards
underneath (Real-Time SPICE Analog, 5 Emulation Engines, Custom Chips,
100+ Components, Live Instruments, Monaco Editor + arduino-cli) now
read from t('landing.features.<key>.{title,desc}') for all 9 locales.
Refactor:
- The `features` array in LandingPage.tsx no longer carries title/desc
literals — only an icon + a translation key. The render maps each
card's key to the matching i18n entry. Cleaner and keeps the JSX
structurally stable across languages.
Translations:
- All 8 non-English locales hand-curated. Technical / brand names
(ngspice-WASM, AVR8, ATmega328P, RP2040, ESP32-C3, CH32V003, QEMU,
Cortex-M0+/A53, ILI9341, NeoPixel, Wokwi Custom Chips API,
WebAssembly, .hex/.uf2/.bin, VS Code, arduino-cli) preserved
unchanged in every locale — those are precise nouns where any
translation would degrade meaning.
Replaces the visible header copy of the supported-hardware section
with t('landing.boards.*') keys:
- label "Supported Hardware"
- titleLine1 / titleLine2 ("Every architecture." / "One tool.")
- subtitle (the "19 boards across 5 CPU architectures..." paragraph)
The five engine cards underneath (avr8js, rp2040js, QEMU lcgamboa,
QEMU Xtensa, QEMU ARM) and per-board specs (e.g. "ATmega328p · 32 KB",
"RP2040 + WiFi") deliberately stay in English — those are accurate
technical specs / product names that don't translate, and mixing
locales inside a spec line would hurt readability more than it
helps.
Hand-curated translations for all 8 non-English locales.
Hero strings now go through `t('landing.hero.*')`:
- titleLine1 / titleAccent (split for the gradient span)
- subtitle (one paragraph; "19 boards / 48+ parts" stays inside the
string so locales can phrase the count naturally)
- ctaPrimary / ctaGithub
- trustLine (the "no signup / runs in browser / free & open-source"
reassurance line — was previously emitting NBSP-wrapped middle
dots; the localised versions use plain spaces, which is fine
visually)
- imageAlt (a11y for the editor screenshot)
Internal /editor link now goes through localize() so a Spanish reader
clicking the primary CTA stays at /es/editor instead of dropping
back to English.
Translations are hand-curated for all 8 non-English locales (es,
pt-br, it, fr, zh-cn, de, ja, ru). Brand names (Velxio, Arduino,
ESP32, Raspberry Pi, GitHub, AGPLv3) preserved as-is. The script
arrow "→" is kept in every locale because it carries directional
meaning that translates naturally across languages.
Block 2 (Boards / supported hardware), Block 3 (features grid),
Block 4 (Support / footer copy) and Editor strings still pending.
Cold first compile per container is unchanged (cache empty). Subsequent
compiles drop from ~5-7 minutes to ~30-60 seconds because every ESP-IDF
base object (FreeRTOS, lwIP, esp_wifi, libsodium, soc, hal, …) hits the
cache. The user's BMP280 example, which hangs on cold compile, completes
near-instantly on the second attempt.
Why a transparent cache is safe: ccache hashes the preprocessed source +
flags + compiler. A cache hit only happens when the input is byte-for-byte
identical to a prior compile. Different sketches with different libraries
still get correct cache misses; there is no path where one project's
output contaminates another.
Changes
- Dockerfile.standalone: install ccache, set CCACHE_DIR=/var/cache/ccache,
IDF_CCACHE_ENABLE=1, configure 2 GB cap with compression. Compression
(level 6) cuts cache disk usage by ~40% with negligible CPU overhead.
- docker-compose.yml: named volume `ccache:/var/cache/ccache` so the
cache survives `docker compose up -d --build` (without it, every image
rebuild discards the cache).
- backend/app/services/espidf_compiler.py: pass `-DCCACHE_ENABLE=1` to
cmake when IDF_CCACHE_ENABLE is truthy. ESP-IDF's project.cmake
(`set_property(GLOBAL PROPERTY RULE_LAUNCH_COMPILE ccache)` on line 374)
is what actually wires ccache in; without the cmake -D flag the env
var alone has no effect because we don't go through idf.py.
Escape hatch: set IDF_CCACHE_ENABLE=0 in compose env to disable without
rebuilding the image.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This is Phase 1 of multi-language support: the visible chrome (header,
footer, language switcher) and routing are wired up for all 9 locales
(en, es, pt-br, it, fr, zh-cn, de, ja, ru) — same set the blog at
velxio.dev/blog/ already supports. The Editor and the long-form
landing-page copy are still English-only and will be translated in a
follow-up.
Infrastructure
- frontend/src/i18n/config.ts: locale registry + LOCALE_META (htmlLang,
native name, og locale, dir). Mirrors pro/blog/src/i18n/config.ts so
cookie sync stays consistent.
- frontend/src/i18n/cookie.ts: read/write the velxio_locale cookie at
Path=/; Max-Age=1y; SameSite=Lax (Secure on HTTPS). The blog reads
the same cookie via an inline script in its Layout.astro.
- frontend/src/i18n/path.ts: getLocaleFromPath / stripLocaleFromPath /
localizedPath / switchLocale / blogUrlFor — match the blog's helpers
one-to-one.
- frontend/src/i18n/index.ts: i18next bootstrap. English bundle is
inlined synchronously for first paint; non-default locales are
lazy-loaded via dynamic import on demand. Initial locale is decided
in priority order URL > cookie > navigator > en.
- frontend/src/i18n/LocaleSync.tsx: top-level wrapper inside <Router>.
On every URL change loads the matching locale bundle, calls
i18n.changeLanguage, writes the cookie, and mirrors the locale onto
<html lang> and dir.
- frontend/src/i18n/useLocalizedNavigate.ts: useCurrentLocale,
useLocalizedHref, useLocalizedNavigate hooks for components that
build internal links.
Routing
- App.tsx: route table extracted to a single ROUTES array, then
registered twice — once at the root (default English) and once
nested under each non-default locale (`/<locale>/...`). Explicit
per-locale parent routes (rather than a generic `:lang` param) so
React Router never accidentally swallows a real top-level path
like `/circuit-simulator` as a locale segment.
Header / Footer
- LanguageSwitcher.tsx + .css: dropdown matching the blog's
LanguageSwitcher.astro. Globe icon + locale code on the trigger,
native names + ISO codes in the menu. Click → `switchLocale()`
rewrites the URL under the new locale; LocaleSync handles the
rest (load bundle, change language, write cookie).
- AppHeader.tsx: every nav label and the auth dropdown copy now
goes through `t('header.nav.*')`, `t('header.auth.*')`. All
internal Links wrapped with localize() so navigation stays
inside the active locale. Added a "Blog" link computed via
`blogUrlFor(currentLocale)` so /es/ → /blog/es/, etc.
- LandingPage.tsx: footer About-Velxio paragraph reads from
t('footer.about').
Translations (Phase 1 strings)
- frontend/src/i18n/locales/<locale>/common.json: nav labels, auth
buttons, footer About copy. Hand-translated for all 9 locales,
AGPLv3 / brand names preserved as-is.
Tooling
- frontend/scripts/translate-i18n.mjs: standalone Node script that
takes the en.json bundles and auto-translates them to the 8 other
locales via DeepSeek (primary) + Gemini (fallback). One LLM call
per (locale, namespace) pair. Run after extracting new strings
with `npm run translate:i18n`.
Phase 2 (deferred)
- Editor (toolbar, file explorer, simulator canvas, component picker,
library manager, error toasts) — hundreds of strings.
- Examples / Docs / About / Profile pages.
- The translate-i18n.mjs script is ready to handle these once the
strings have been extracted into JSON keys.
The previous footer credit ("MIT License · Powered by avr8js &
wokwi-elements") was wrong twice over: velxio is AGPLv3 (with a
commercial license available), and the project now ships much more
than the two libraries it singled out (rp2040js, eecircuit-engine,
QEMU, ESP-IDF, arduino-cli, Monaco editor, ...).
Replace it with a one-paragraph About Velxio that sits at the bottom
of the landing page, mirrors what the blog footer shows at
velxio.dev/blog/, and correctly states the AGPLv3 license.
Widen .footer-copy to max-width: 680px so the longer copy has room
to breathe and breaks across two lines on desktop.
The synchronous /api/compile endpoint forced one long-lived HTTP request
to span the entire build. Cloudflare's 100s edge timeout cuts that off
mid-flight for any cold ESP-IDF compile (BMP280 takes 5-7 min on first
run). The user-visible symptom was HTTP 524 well before the backend
even noticed.
Backend (compile.py)
- New `POST /api/compile/start` returns `{job_id}` immediately and
spawns the actual compile as an asyncio.create_task background.
- New `GET /api/compile/status/{job_id}` returns the current job state
(`pending` | `running` | `done` | `error`). Each poll completes in
milliseconds, far under any edge timeout.
- Existing `POST /api/compile/` kept verbatim for backward compatibility
(AVR/RP2040 builds finish in seconds and don't trip 524).
- Build logic extracted into `_run_compile()` so both paths share one
implementation; no duplicated ESP-IDF / arduino-cli branching.
- Async path opens its own short-lived DB session via AsyncSessionLocal
for metric recording — the request-scoped session is dead by the time
the background task finishes.
- COMPILE_JOBS dict purges entries 30 minutes after completion so a
busy server doesn't grow unboundedly.
Frontend (compilation.ts)
- compileCode() now: POST /compile/start → poll /compile/status every 2s
until state ∈ {done, error}, with a 15-minute client-side cap.
- 30s axios timeout per individual call (not per build) so transient
network blips during a long compile auto-retry instead of failing.
- 404 on /status throws (job expired / server restarted); other poll
errors warn and retry. Surfaces structured error responses verbatim
so the editor's compile-error panel keeps working unchanged.
Limitation: COMPILE_JOBS lives in-process; if velxio ever scales to
multiple FastAPI workers this needs to move to Redis or sqlite. Single-
instance is fine today.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two distinct issues hit the component SVG generation step:
1. `velxio-bmp280` was in ELEMENTS but tries to require
bmp280-element.js from the wokwi-elements CJS dist — that file
doesn't exist because BMP280 is a velxio-native component, not a
wokwi one. Its SVG already ships hand-authored at
frontend/public/component-svgs/bmp280.svg, so the script should
never have tried to extract it. Drop the row and leave a comment
explaining why.
2. `wokwi-ssd1306` failed with "ImageData is not defined" because the
element constructor seeds an off-screen canvas with
`new ImageData(width, height)` — a browser API absent in Node.
We never invoke putImageData (renderSVG() draws the static frame
from scratch), so a minimal global stub that doesn't throw is all
that's needed. Polyfill it on globalThis next to the existing
customElements stub.
After this:
- 38 generated, 0 skipped, 0 failed (was: 1 skip, 1 fail).
- ssd1306.svg now ships in frontend/public/component-svgs/.
Browsers were reporting:
Unsafe attempt to load URL http://velxio.dev/examples/
from frame with URL https://velxio.dev/examples.
Repro:
curl -I https://velxio.dev/examples
→ 301 Location: http://velxio.dev/examples/ ← protocol downgraded
Why: docker/nginx.conf has the container listening on `:80` only — TLS
is terminated by the host nginx in front of it (which then proxies to
http://127.0.0.1:3080). When nginx generates a trailing-slash 301 it
uses the listening protocol (http) for the absolute Location header,
not the X-Forwarded-Proto. The browser correctly blocks the redirect.
Fix: `absolute_redirect off;` makes nginx emit relative Location
headers (`Location: /examples/`), so the browser resolves them
against the original request URL and HTTPS is preserved end-to-end.
After fix:
curl -I https://velxio.dev/examples
→ 301 Location: /examples/
Goes in the same branch as the BMP280 ninja-timeout fix because both
are deploy-blocking and small.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The ESP32 BMP280 example compile was timing out at 98% (1473/1483 build
steps), failing with the unhelpful "ESP-IDF build timed out (300s)"
message even though every individual step was healthy.
Cold ESP-IDF builds that pull in external Arduino libraries — Adafruit
BMP280 + Adafruit BusIO + Adafruit Unified Sensor on top of the base
arduino-esp32 component tree — routinely produce ~1480 build objects.
On modest VPS hardware this takes 5-7 minutes the first time. Ninja's
incremental cache makes subsequent compiles seconds, but the first one
needs more headroom.
Constant lifted to NINJA_TIMEOUT_S so the value used in the timeout
matches the value reported in the error message — the previous code
hard-coded "300s" in two places that were free to drift apart.
Repro before: open the example "ESP32 — BMP280 Barometric Pressure"
on velxio.dev/editor on a clean container, click compile → fails after
5 minutes with timeout. After: completes in ~6 minutes on the first
run, ~5 seconds on subsequent runs.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The hand-drawn SVGs in Bmp280Element.ts and Attiny85Element.ts were
functional but obviously amateur next to a real Fritzing-drawn part.
Both components now mount the equivalent Fritzing breadboard SVG as a
public static asset (`<image href>` in the shadow DOM SVG), with pin
coordinates remapped to the new artwork and pin-name labels overlaid
on top so the user can still read each connector at a glance.
frontend/public/component-svgs/bmp280.svg (new)
Verbatim copy of third-party/fritzing-parts/svg/core/breadboard/
bmp180_breadboard.svg. The Adafruit BMP180 breakout is the
mechanically identical Bosch predecessor — same I2C interface,
same 4-pin pinout. Pin labels lifted from the matching .fzp.
frontend/public/component-svgs/attiny85.svg (new)
Verbatim copy of the Fritzing ATtiny85 DIP-8 breadboard art.
Bmp280Element.ts
Width 80×100 px (Fritzing aspect 28.35:35.43 ≈ 0.8:1, exact uniform
scale of 2.822 px/mm). Pin coords for SDA / SCL / GND / VCC matched
to the connector centres in the source SVG. Pin labels overlaid on
top. Existing wired example (esp32-bmp280) re-routes automatically
because the wire system reads coords by pin name from pinInfo.
Attiny85Element.ts
Width 160×132 px (Fritzing aspect 28.801:23.768 ≈ 1.21:1, exact
uniform scale of 5.555 px/mm). The Fritzing layout puts pins on the
TOP and BOTTOM edges (4 each), not LEFT and RIGHT like the older
hand-drawn version. Pin coords land on clean numbers
(x ∈ {20, 60, 100, 140}, y ∈ {6, 126}). Built-in LED on PB1 stays
as an overlaid circle outside the chip body.
Wires in the existing attiny85-* examples re-route automatically by
pin name; external components positioned to the right of the chip
may need a manual nudge for clean routing — but they work.
frontend/src/components/simulator/BoardOnCanvas.tsx
attiny85: { w: 160, h: 100 } → { w: 160, h: 132 } to match the new
aspect ratio. Same width as before so the chip occupies the same
horizontal slot in existing example layouts.
scripts/component-overrides.json
BMP280 thumbnail updated to mirror the Fritzing colour scheme
(dark blue PCB, BMP180 silkscreen, four gold connector circles)
so picker and canvas feel consistent.
frontend/public/components-metadata.json
Regenerated.
docs/THIRD_PARTY.md
New "Fritzing parts library" section. Both new assets are listed
with their upstream paths plus the CC-BY-SA licence and link to
the parts repo. Future Fritzing copies must be added there too.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
CI's e2e test_hcsr04_simulation.mjs caught the regression introduced by
a3f21a2 (the issue #101 fix). The worker crashes on boot with
Firmware decode error: No module named 'app'
esp32_worker.py runs as a subprocess via subprocess.Popen([sys.executable,
WORKER_PATH, ...]). When Python launches a script directly, sys.path[0]
is the SCRIPT's directory (backend/app/services/), not the backend root.
So `from app.services.esp32_flash_image import pad_to_flash_size` fails
because there is no `app/` under `backend/app/services/`.
esp32_lib_bridge.py wasn't affected because it runs in-process inside
uvicorn, where backend/ is implicitly on sys.path.
Fix: same try/except + importlib fallback the worker already uses for
esp32_i2c_slaves at the top of the file. First try the package import
(works when imported by the bridge's tests or anything else with the
backend root on sys.path), fall back to direct file loading otherwise.
Verified the fallback works in isolation by simulating the subprocess
context (sys.path containing only backend/app/services/) — the package
import fails as expected and the file-load fallback returns a properly
padded 4 MB buffer.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Closes the loop on the four ESP32 fixes that landed earlier in this
series. Each previously-broken or noisy example now has a regression
test that compiles the sketch through the production ESP-IDF compiler
and runs it in QEMU via esp32_worker.py — the same code path the
WebSocket /ws/{client_id} endpoint drives in production. Testing the
worker directly skips the WS transport but exercises the same compile
→ flash → boot → serial cascade.
Coverage:
- TestEsp32SerialCleanliness — DHT22, Servo+Pot, Joystick, Dual ADC
Asserts the user's Serial.print substring shows up AND no
`I (xxx) gpio:|wifi:|phy:` info-level ESP-IDF logs leak through.
This validates the sdkconfig CONFIG_LOG_DEFAULT_LEVEL_WARN change
from commit b373c97.
- TestEsp32CompileSuccess — BLE Advertise, LEDC RGB
BLE Advertise validates the sdkconfig switch to Bluedroid (was
NimBLE-only, which broke arduino-esp32's BLEDevice.h).
LEDC RGB validates velxio_compat.h's ledcAttach() shim from
commit f6f6f43; the sketch uses the arduino-esp32 3.x one-shot API
on a 2.0.17 toolchain.
- TestEsp32WiFiSketches — WiFi Connect, WiFi WebServer
Regression coverage to make sure the sdkconfig changes didn't break
WiFi association. Connect must reach an "IP Address:" line; Server
must report "Server started".
- frontend/src/__tests__/component-metadata-bmp280.test.ts
Sanity check that the BMP280 entry from commit 1f3f2e0 survives
metadata regeneration.
All ESP-IDF/QEMU tests use unittest.skipUnless on
_toolchain_available() so they no-op cleanly on dev boxes without
libqemu-xtensa, and only do real work in the Docker CI image.
Sketches are inlined verbatim from the public Velxio examples.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
A beta tester reported "BMP280 - module graphic missing" — picking the
example loaded a working sketch but the canvas component fell back to
the MPU6050 placeholder.
Bmp280Element.ts already exists and registers velxio-bmp280 with the
right pinInfo, but it was never injected into components-metadata.json,
so the component picker and CircuitPreview didn't know about it. Add
an entry in scripts/component-overrides.json under _customComponents
(per CLAUDE.md §6b — direct edits to the generated JSON would be
clobbered by the next metadata regen) and regen.
The thumbnail mirrors the GY-BMP280 breakout look from the Web
Component itself: green PCB, black die label, four gold pin pads.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Beta testers saw raw `[0;32m` and `[0m` text mixed into the serial
monitor for several ESP32 examples. Those are ANSI SGR escapes the
ESP-IDF logger emits to color INFO/WARN lines on a real terminal. Our
<pre> renders them literally because there was no ANSI handling in
the path.
Strip the SGR sequences (`\x1b\[[0-9;]*m`) before the IP-linkifier so
the user only sees plain text. Combined with the sdkconfig change that
drops the default log level to WARN, the ESP32 output is now as clean
as the AVR output.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
A user reported the LEDC PWM RGB example failing to compile. The sketch
calls ledcAttach(pin, freq, resolution) — the one-shot API added in
arduino-esp32 3.x. Our toolchain image pins arduino-esp32 to 2.0.17
(matched to ESP-IDF 4.4.7 + the lcgamboa QEMU ROM), where the API is
the older two-step ledcSetup + ledcAttachPin pair. Sketches written
against 3.x docs hit "ledcAttach was not declared in this scope".
Bumping arduino-esp32 to 3.x means moving to ESP-IDF 5.x, which may
break our QEMU fork. Cheaper fix: ship a compat shim header in the
ESP-IDF project template that defines ledcAttach + ledcAttachChannel
in terms of the 2.x API, gated on `!defined(ledcAttach)` so it
disappears the day we bump.
espidf_compiler.py now injects #include "velxio_compat.h" right after
Arduino.h whether the user explicitly included Arduino.h or we
prepended it ourselves.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two reports rolled into one sdkconfig change:
1. Beta testers reported "weird serial output" across DHT22, Servo+Pot,
Joystick, WiFi×2, and DualADC examples. The user's Serial.print lines
came interleaved with internal ESP-IDF chatter:
I (53306) gpio: GPIO[4]| InputEn: 1| OutputEn: 0| ...
I (10626) phy_init: phy_version 4791,2c4672b,...
sdkconfig.defaults didn't pin a default log level so it inherited
CONFIG_LOG_DEFAULT_LEVEL_INFO. Set WARN (level 2) so only warnings
and errors leak into the user's serial output. Sketches can still
esp_log_level_set() per tag at runtime if they want verbose.
2. BLE Advertise example failed to compile. sdkconfig had NimBLE
enabled, but arduino-esp32 2.0.17's BLEDevice.h targets Bluedroid.
Switch the stack: enable BT_BLUEDROID + BTDM_CTRL_MODE_BR_EDR_BLE +
BT_BLE so the standard arduino-esp32 BLE library compiles. Sketches
that explicitly include NimBLEDevice.h will not compile under this
config — that's a smaller minority than the BLEDevice.h users.
The runtime side of BLE still depends on stubs in the qemu-lcgamboa
fork; this only fixes the compile path.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Issue #101 reproducer: an ESP32 sketch that pulls in Adafruit_SSD1306 +
Adafruit_GFX produced a "No response from server. Is the backend
running on port 8001?" error in the browser. The compile actually
succeeded backend-side, but the JSON response carrying the firmware
was ~5.5 MB of base64 — the ESP-IDF compiler builds a full 4 MB merged
flash image (mostly 0xFF padding), encodes it whole, and ships it. In
prod that response goes through nginx + Cloudflare, which buffer-fail
or RST the connection on payloads that big — axios then lands in the
"no response" branch with no HTTP status to surface.
Fix: trim the trailing 0xFF padding before serializing, re-pad to a
valid QEMU flash size (2/4/8/16 MB) just before mtd attach. Lossless:
bytes after `last_used` in the merge are 0xFF by construction, so
trim → pad reproduces the original image byte-for-byte.
Numbers from the reproducer (Adafruit_SSD1306 + Adafruit_GFX,
esp32:esp32:esp32 board):
before: ~5.5 MB JSON response
after: 539 KB JSON response (10× smaller)
backend/app/services/espidf_compiler.py
_merge_flash_image now tracks `last_used` across the three placed
sections (bootloader / partitions / app) and writes only
flash[:last_used] to merged_flash.bin.
backend/app/services/esp32_flash_image.py (new)
Shared `pad_to_flash_size(bytes) -> bytes` helper. Rounds up to the
next valid QEMU flash size with a 4 MB minimum, matches the
frontend's existing padToFlashSize logic in Esp32MicroPythonLoader.
Raises ValueError on >16 MB inputs (would indicate a broken upstream
merge, not anything user-recoverable).
backend/app/services/esp32_lib_bridge.py
backend/app/services/esp32_worker.py
Both QEMU consumer paths (in-process and subprocess) call
pad_to_flash_size right after base64.b64decode, before writing the
tmp .bin that QEMU attaches with `-drive if=mtd,format=raw`.
Verified: smoke test confirms trim → pad → original is byte-exact.
Edge cases covered: small payloads pad up to the 4 MB minimum;
firmwares >16 MB are rejected loudly.
Closes#101
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Replaces the window.confirm("Delete X?") modal with a footer that flips
into a "Delete X?" prompt + Cancel / Delete pair when the user arms the
delete. Less jarring on mobile (no native dialog), keeps the user's
flow inside the property panel.
The two hand-rolled curved-arrow SVGs I drew in bd5fd18 looked off — the
arrowheads were misaligned and the curve clipped at the bottom of the
viewBox. Swapped both for the canonical lucide-react icons (Undo2 /
Redo2), which match the visual weight + alignment of the rest of the
toolbar.
lucide-react was already in the OSS deps (added when other parts of the
app started using it). No new dependency.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
UI-facing half of the undo/redo feature. Combined with the previous two
commits, Ctrl+Z (or the toolbar button) now reverses every canvas
mutation: add/remove component, move, rotate, set property, add/remove
wire.
EditorPage.tsx:
- New window-keydown effect for Ctrl+Z / Ctrl+Y / Ctrl+Shift+Z (and the
Cmd equivalents). Uses the same input/textarea/contenteditable guard
as the existing Ctrl+S handler — Monaco's per-file undo and the AI
chat composer keep their own behaviour.
SimulatorCanvas.tsx:
- Two icon buttons (undo + redo) added to the canvas header, between
the board selector and the Serial Monitor toggle. Tooltip surfaces
the next command's description ("Undo: Add LED (Ctrl+Z)") so the
user knows exactly what's about to revert. Buttons disable when the
stack is empty in that direction.
- New `canvas-icon-btn` CSS class for square 32×32 icon-only buttons
(matches the visual weight of the existing Serial button without
the label).
- Subscribes to history / historyIndex via store selectors so the
buttons re-render reactively as commands are pushed/undone.
No new tests — the store-level coverage from 99ed22b already exercises
undo/redo round trips. UI affordances are wired pass-through to those
store APIs.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Wires every user-initiated canvas mutation in SimulatorCanvas to the
recorded variants added in 99ed22b, so Ctrl+Z (next commit) can roll
each one back as a single step.
Routed through record*:
- handleSelectComponent (picker → add) → recordAddComponent
- Delete keyboard handler (selectedComponentId branch) → recordRemoveComponent
- handleRotateComponent → recordRotate (still mutates live via
updateComponent so the rotation visually applies; record stores the
prev/next angles for undo)
- Drag → recordMove on mouseup. Captures component.x/y at mousedown in
a new dragStartPosRef and only records on actual drag-end (skips the
click branch that just opens the property dialog).
- Pin-click "finish wire" path → calls finishWireCreation (which
atomically appends the wire) then pushes a CanvasCommand for that
wire with applyNow:false (state is already at post-add).
- Selected-wire delete (keyboard + SelectionActionBar + PinPickerDialog)
→ recordRemoveWire
- Selection action bar component delete → recordRemoveComponent
- Pin picker dialog component delete → recordRemoveComponent
- ComponentPropertyDialog onPropertyChange → updateComponent applies
live, then recordSetProperty captures prev/next so Ctrl+Z reverts the
value without re-running the raw mutation.
Cleaned up unused destructures of addComponent / removeComponent /
removeWire from the original useSimulatorStore() call — every call site
now uses the record* equivalents.
No new keyboard shortcuts or toolbar buttons in this commit; that's
landing next.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds the foundation for canvas undo/redo. UI wiring, keyboard shortcuts,
toolbar buttons and agent-tools refactor land in follow-up commits.
useSimulatorStore.ts:
- New CanvasCommand type — { description, execute, undo }.
- HISTORY_MAX = 50 (oldest entries dropped on overflow).
- New state: history[] + historyIndex (-1 = empty).
- New APIs: pushCommand(cmd, {applyNow?}), undo, redo, canUndo, canRedo,
clearHistory.
- New "recorded" actions that wrap raw mutators with a CanvasCommand:
recordAddComponent, recordRemoveComponent, recordMove, recordRotate,
recordSetProperty, recordAddWire, recordRemoveWire, recordUpdateWire.
- recordRemoveComponent captures both the component AND any wires that
cascade with it, so undo restores both atomically.
- recordMove also re-runs updateWirePositions on undo/redo so wire
endpoints follow the component back/forward.
- setComponents and setWires (project-load / clear paths) now call
clearHistory inline — leaving stale commands pointing at IDs that no
longer exist would crash on undo.
Why custom Command pattern over zundo / travels:
- The store has 30+ ephemeral fields (simulator instances, serialOutput
growing byte-by-byte, hexEpoch counter, wireInProgress that ticks 60×/s
on drag). Snapshot/diff middleware would either burn memory tracking
them or need a fragile partialize allow-list.
- Per-op descriptions ("Add LED", "Move resistor") for tooltips come for
free with this approach; zundo/travels would need to infer them.
Tests: 15/15 in src/__tests__/undo-redo.test.ts — covers cap-at-50,
redo-truncation, cascade undo of remove-component, move/rotate/property
round trips, bulk-setter clearing.
Raw mutators (addComponent / removeComponent / updateComponent / addWire /
removeWire / updateWire) are unchanged. UI handlers can keep using them
during live drags for preview frames without spamming history; the
record* actions are what drag-end, click-finish and agent tools should
call going forward.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Replace the "DMC" initials placeholder with the GitHub avatar
(https://avatars.githubusercontent.com/u/47928504?v=4). The CSS for
.about-creator-avatar already had the rounded frame; just swapped to
object-fit: cover so the <img> fills the circle correctly, plus a
subtle ring + drop shadow.
- Add a "Recent releases" section between the Creator block and the
personal-story quote, with two cards:
- Velxio 2.5 (Latest) → /v2-5 (ngspice-WASM analog co-simulation)
- Velxio 2.0 → /v2
Each card has a tagline + 2-3 line blurb. The 2.5 card gets a blue
border + "Latest" tag so it reads as the current launch. About now
surfaces both release pages, which previously were only linked from
the Circuit/Electronics/SPICE simulator pages.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Implemented PinPickerDialog for selecting pins on touch devices.
- Added SelectionActionBar for managing selected items with touch actions.
- Created WireModeBanner to provide feedback during wire creation.
- Introduced useTouchDevice utility for detecting coarse pointer input.
- Refactored WireLayer to utilize useIsCoarsePointer for touch detection.
Previous commit unconditionally enabled preserveSymlinks when
VITE_PRO_BUILD was set. That works for `vite dev` (where the overlay
is wired in via a Windows junction and the resolver needs to keep the
junction path so relative imports back into the OSS sibling dirs
resolve), but it BREAKS `vite build` in Docker — there are no symlinks
to preserve, and Rollup with preserveSymlinks=true fails to resolve
relative imports from the copied overlay tree:
Could not resolve "../../../services/componentRegistry"
from "src/pro/agent/tools/canvas.ts"
Gate the flag on `command === 'serve'` so it only kicks in during dev.
Production builds always run with preserveSymlinks=false (the default).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Discovery regex was matching only single-quoted ID literals, so the
auto-generated examples-100-days.ts file (which uses double-quoted
strings, by convention of its Python emitter) was completely missed.
Same for picow-wifi: the script wasn't reading examples-picow-wifi.ts
at all.
Two-line fix in scripts/capture-example-thumbs.mjs: the regex now
accepts both `'…'` and `"…"`, and the data-file list includes
examples-picow-wifi.ts.
Captured slugs:
- 49 × `100d-*` (100 Days of IoT — MicroPython on ESP32 / Pico)
- 4 × `picow-*` (Pico W WiFi — async LED, relay web server,
servo web, websocket LED)
Coverage: 219/226 examples (97%). The 7 still falling back to
CircuitPreview are component-ID literals (`epaper-1in54-bw`,
`epaper-2in13-bw`, etc.) that the regex over-matches — they 404
on /examples/<slug> because they aren't real example IDs.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Changes that ship to OSS — all benign for self-hosters, but most are
extension points the velxio-prod overlay (and any private fork) needs to
plug an in-editor AI chat into the page.
Editor:
- 3-way view-mode toggle (code / both / circuit) in the unified toolbar.
Lets users hide a pane to give a right-docked sidebar (e.g. the AI
chat overlay) more breathing room. Persisted in useEditorStore.
- Default file explorer narrower (210 → 165 px); min 110.
- Removed the redundant `tb-board-pill` (icon + "Editing: X" tooltip);
the BoardSelector dropdown elsewhere already shows the active board.
- Inlined Import/Export/Upload-firmware buttons; the 3-dot overflow
menu gave up too much discoverability. Removed dead overflow state.
Simulator:
- Fix: global Delete/Backspace handler in SimulatorCanvas no longer
fires when the event target is an INPUT/TEXTAREA/SELECT/contentEditable
— affected any in-page text field, not just the chat overlay.
Overlay extensibility:
- New `data-velxio-slot="agent-chat"` at the bottom of EditorPage so
pro overlays can portal a chat panel into the editor without
forking the page.
- vite.config.ts: preserveSymlinks=true when VITE_PRO_BUILD is set.
Lets local-dev junctions (overlay tree → frontend/src/pro) resolve
bare imports back to the OSS node_modules without resolving symlinks.
Deps:
- Added react-markdown + remark-gfm (rendered chat output) and
@google/genai + zod (overlay agent loop). Tree-shaken from the OSS
bundle when no pro code imports them.
gitignore:
- Ignore backend/app/pro/ and frontend/src/pro/ junctions used by
developers running a private overlay against the OSS dev server.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
davidmonterocrespo24