An ok os.launch result is held while the launched app is not on the
bus and delivered when it registers (within 8 s), rewritten as "<App>
is running now. Its tools: … Call them directly." with the tool table
rebound mid-turn where the model can (the local model prepends the
update to the tool result; Claude keeps its native table until the next
turn and gets the names in the text); past the patience the result says
the app has not connected yet. An assistant entry that never got
visible text is dropped before a tool call and at the end of a turn,
and the panel draws an empty one as nothing — the blank block above
the first card is gone. The model's context now opens with the running
apps and their tool names and lists the others as os.launch candidates;
the WM's launch description says a running app's tools are already
there, and its answer for one that is already running says so and
brings it to the front.
Proven on the fleet model: "open the photos app and find the pictures
about dogs" → launch (running 4 s later) → search, summary, search →
the answer, one turn, the tile beside the pane. ai-services 38, wm 153,
aichat 2; aichat checks for wasm32.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The hub chat session's election no longer stops at this machine: a
co-located serving holder first, then a fleet chat node heard by
discovery and role-allowed for chat (a short patience for beacons),
then the weights on this machine — and an honest answer naming both
misses and where weights may be put when none of those exists. Tool
packs ride every route: the system text carries the tool table as the
node's chat_system, one splitter strips thinking and collects
<tool_call> bodies across deltas, the in-process worker's parser is the
one parser, and tool results go back as tool turns. A node that fails
mid-turn ends that turn with the node named; the person's next line
re-elects, served first on the new route. The tools-only guard on the
proxy is gone. The session exposes its route, and the panel's chip
shows it.
The local lookup is independent of the working directory: the env
override, then the makepad home's weights (Qwen preferred, largest),
then the checkout the binary came from, then the cwd — so an assistant
launched from a binary copy finds the same weights as one run from the
checkout, and the Local provider always builds even with none.
Proven live with no local weights: from the sheets overlay a plain
line reached the fleet's 27B (the node with the model resident won the
pick), the model called sheets.summary and answered in 12 s; the app
stayed at 254 MB. hub_chat + local_llm 11, services 34.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
apps/photos (lib + bin) puts libs/image_tiles' TileGrid in a tile: it
opens a baked library found by name (IMAGE_TILES_HOME, the cwd's
local/image-tiles, the checkout the binary was built in — the smbc
collection first), shows a status line with the bake command when
there is none, and exposes photos.search (every query word in a title
or link), photos.show (the camera glides onto one picture) and
photos.summary — the same answer for the port and for the module's
executor. PhotosModule mints the view inside an isolate with an open
schema that takes a collection name, never a path; the WM links it
behind app-photos and lists it in the menu. image_tiles learned to
decode GIF and WebP and to read a manifest's pictures from disk, so
the bake can take local/smbc's mirror: the last 300 comics are baked
under local/image-tiles/smbc (293; seven of the mirror's files are
saved error pages). fabric joins the launcher with a title and a
polite close.
Driven hidden: the wall in a process tile beside the left pane, panned
and zoomed; search and show cards; the same wall as an in-process
module; fabric launched and closed; standalone photos with the F10
overlay on the left. photos 7, image_tiles 5, wm 152; the photos lib
checks for wasm32.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
- libs/sim: DeckStrip (deck.rs) — every laid corridor registers its walk
deck; movers pin to it exactly instead of climbing 2.5 m slab kerbs;
feet inside a static box lift onto its top; the world composes floor,
terrain, materials and voxel in one sampler
- libs/sim + libs/render: camera_path_limit takes the filmed body and never
treats it as an obstruction (the coaster's ride boom was clamped to 1 m);
the shadow gate's settle test gets a hair of float slack
- libs/asset/chat: the in-game builder never asks before building; the
brief teaches game.driver and the race countdown, under budget
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The desk (layout, tiles, the shell, the AI pane, the module host, the
bus) is one program everywhere; what differs around it is asked through
host.rs: a monotonic clock in platform seconds (std::time traps on
wasm), whether child processes exist at all, a child's environment, the
theme choice kept in the desk's own storage namespace. Every Instant in
the WM is gone; the hub, the warm pool and process launches are skipped
where there are no processes, and the bar's clock falls back to the
platform epoch when it cannot fork date.
The pane has a second body: where the chat is linked in as a module —
the web build, or a desktop that switched aichat to module hosting — the
pane instantiates mod.widgets.AiChatOverlay by name, exactly as a
Window's F10 slot does, and the WM's own os service and every module
instance reach it as in-process links (pane_links.rs), never as frames.
The registry's web profile makes every linked module a Module and
answers os.launch of anything else with unavailable. sheets' Open and
Save go to the instance's storage jail, asynchronously, on every
platform.
The assistant starts with the desktop (decision 15): the aichat child
(native) or the overlay (in-process) comes up at startup with the pane
closed and no focus taken, drawn just off the desk's edge so the child
is configured and presents before any F10 — which now only slides it
in. The gallery fixture leaves the desk's DSL for a by-name host (its
construction overflowed the wasm stack).
Platform bug found by the web drive and fixed in libs/wasm_bridge: the
ToWasm reader added a block's absolute end to a relative base, so two
JS→wasm messages in one buffer decoded garbage and trapped; it seeks to
the stored end, with a batched-buffer test.
Verified natively by a hidden drive: the child's first frame 1002 ms
after boot, before any F10; F10 → the pane at once with the composer
focused; F10 → gone. On the web (threaded build, 1.5 MB brotli, COOP/
COEP): desk, pane with No model, os.list, an in-process Sheets tile and
its summary, terminal answered unavailable, zero traps. wm 148, sheets
84, app-module 3, wasm-bridge 1; wasm32 checks and the std::time clippy
gate clean for wm, sheets, app-module, aichat, wasm-bridge.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
"make me a giant forest with roads" answered with a plan and "Want me to
go ahead and build it now?" — the player is holding a controller. The
agentic doctrine and the game brief now say it outright: a request is an
instruction, decide the open details, build in this turn, report in a
sentence; a turn that ends in a question is a failed turn.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Crossroads rebaked its lightmap on every stoplight phase (70 bakes a
minute): the traffic kit's lamp repaint bumped the one static revision
the renderer keys its slabs, occlusion, shadow receivers AND the GPU
lightmap kick on. The world now carries a paint revision beside the
geometry revision; the static slabs repack on either, the bake kick
only on geometry. Tests pin both.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
libs/app_module (makepad-app-module) is the contract (aicontrol §3):
AppModule (id, label, register into an isolate, an OpenSchema, create
into InstanceParts, capabilities), InstanceHandles (an InstanceScope
owner token, the storage jail as a cx.storage namespace, the viewport,
a ReplySink for calls that finish later), a ServiceExecutor addressed
apart from the root, and an OpenSchema that refuses raw paths — a file
is a handle the host issued or nothing. Only types live there.
The WM hosts one (module_host.rs): allocate the isolate, retint its
theme from the palette, register, create — one trusted entry into the
isolate, never a second &mut Cx beside the VM; teardown drops the root
first, runs shutdown in the isolate, frees it. MpModuleView is the tile
for an instance's root, drawing and dispatching with the isolate
INSTALLED on Cx (enter_isolate/leave_isolate, new in widgets) and
gating keys on the WM's focus; a TileHost trait covers both tile kinds
so the desk and the focus logic never ask which. The registry overlay
(apps.rs) links modules per app-* feature; the desktop default stays
Process, switched per app in ~/.makepad/wm/apps.splash or by a dev
--module flag. The bus gains its in-process leg: an instance is an
m<id> endpoint the pane addresses like any other — the leg the web
superbuild runs everything on.
sheets is lib + bin: SheetsModule mints MpSheets{} in the isolate and
answers sheets.summary on its executor; the standalone binary keeps its
Window and F10 overlay behind the standalone feature.
The drive found a platform bug: a shader compiled on a widget minted in
an isolate hit the draw-shader object cache by object index alone, so
an isolate's Button reused the main heap's entry and drew with zero
instance slots. The cache is keyed by (heap, object) in every backend,
and add_instance refuses a zero-slot draw call instead of dividing by
it. Isolate cost, measured (the P2 entry gate): 12 ms and ~8 MB per
isolate, flat to 32 — no pre-warming needed.
Verified by hidden grabs under the WM: /os.launch sheets → an
in-process Sheets tile (no child process), the apps row Desktop ·
Sheets, /sheets.summary answered through the in-process leg, a warm
terminal beside it, F10, close → the instance torn down and its isolate
freed. wm 145, sheets 84, app-module 3, services 33; platform and draw
check for windows-msvc, linux-gnu and wasm32.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
- libs/sim: meshfloor.rs FloorRaster (one storey per column from a map's nav
graph); composed_surface_sample_at(floor, terrain, materials, voxel) is the
one seam nav, corridors and lots read; in-eval voxel ops no longer bump the
history revision
- libs/asset/chat: world.get_plan / world.set_plan (shape-validated, stale
revisions refused), the authoring state prefixed to every turn, map turns
routed to the plan tools; game.md: the foundation line, the capability kinds
verbatim, no order-independence claims
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The per-family model crates (flux, h3, paint, music, speech, stems,
vision, sfx, rife, trellis, beats, common) and the libs/diffusion
research binaries read some 190 environment variables that were
research knobs: tensor dumps, per-stage timing, oracle-parity and
fixture rigs, experiment toggles. The path taken with none of them set
is the one that ships; every such knob is deleted with the code it
gated, and every losing branch of an experiment toggle goes with its
toggle — dead kernels, fields and functions included. What remains are
the real configuration variables (the FLUX_*_MODE family, FLUX_GRAPH,
the VAE pool cap, the FLUX2 text-encoder residency, H3_VAE_BATCH, the
music3 caches and official modes, the stems/beats f16 switches, the
weight and data roots) and the build-script variables.
Rebased on the runtime cleanup: precision stays explicit everywhere
(GemmPrecision, f16_attention_operands, the H3 text precision, DA3's
StrictF32 in code); no act16, no H3_ACT_F16, no FLUX_ATTN_F16 or
FLUX_VAE_CONV_GEMM reads survive.
Reviewed by the delegate reviewer (APPLY, no findings) and gated on the
Windows CUDA box: all seventeen model crates check, motion and vision
tests, the hub and the diffusion bins — the gate caught one CUDA-only
tap marker the Mac never compiles, removed here. On this Mac: the same
checks plus motion 24, paint 158 and vision 23 tests.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
A headless box's service log showed only the startup banner for a whole
session; now it records the model and wire encodings on open, and on close the
elapsed time, frames in/out, fps, and the dropped/undecodable/unencoded counts.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The AI runtime crates read their precision, activation and kernel
choices from environment variables; the code path taken with none set
is the one that ships, so every knob that selected it is now an explicit
argument and every losing path is gone. GemmPrecision { f16_accumulate,
f16_activations } is passed by each caller: the default {true, true} is
the old unset Flux route; H3's DiT and text encoder pass {false, false}
(H3's >1e4 activations saturate f16 — the policy H3 used to set through
FLUX_GEMM_F16ACC=0 on itself), its VAE {true, false}; DA3's StrictF32
selects f32 packed attention in code; Hy-Motion carries an explicit
f16_attention_operands flag through its text refiner, its double and
single blocks and the CUDA backend (true in production, false only in
its full validator). The libs/diffusion bins — a separate workspace —
are migrated to the same shapes.
Benches and validators no longer set variables on themselves: llama's
skip-logits is a session option (the CUDA bench turns it on), OCR takes
explicit use_f16_gemm and tiled_roformer options, the lane speculative
probe reads its CLI. The live gates the first cut had deleted are back
as explicit-config tests: MMQ M=129, the strided-f32 MMV path, the
RMS+MUL CPU oracle. The loader's THREADS and CHUNK_MB stay real settings.
The Metal quantized-matmul experiment (metal_qmm and its vendored MLX
kernels) was reachable only through a knob and goes with it.
Reviewed in three rounds by the delegate reviewer (the last round
accepted everything but one Hy-Motion call site, fixed in round four
and reviewed here), and gated on the Windows CUDA box: lib checks of
common/paint/loader/cuda/llm/motion/vision, motion 24 and vision 23
tests, the hub check, the diffusion bins, llm 253 passed / 1 ignored.
On this Mac: the same checks plus the motion and vision tests.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Every Window now carries an AiChatSlot beside the tweaker: zero cost
while off, inert when the window manager hosts the process (the WM's
pane is the chat then), and on the first bare F10 it instantiates the
chat's module root BY NAME — mod.widgets.AiChatOverlay — which exists
when the app links makepad-aichat and calls its script_mod; an app that
does not gets one log line. The overlay slides in from the right over
the body with the WM pane's motion, owns the pointer inside its rect and
the keyboard through its composer, keeps ticking while hidden, and draws
in the window's retained overlay list so it composites over the deferred
body. Requests from outside the tree ride Cx globals, never statics.
An app exposes itself with one call: AiServicePort::open(cx, manifest)
is the hosted transport under the WM and, standalone, an in-process link
parked on Cx (PendingServiceLinks) that whichever chat root is up adopts
into its registry — the overlay today, the superbuild's pane later. The
registry wakes the UI when it sends, so an in-process call is answered
without waiting for a pointer event. NoModel leaves the engine feature
so a build without a model runtime still answers honestly and keeps the
tool console; aichat's engine is a default-on feature.
The bridge: Cx::ai_callback beside tweak_callback, /ai?on=1|0, /ai?say=…
and /ai/transcript (the overlay publishes the transcript as JSON after
each state change), installed by makepad_aichat::script_mod.
sheets links aichat and exposes sheets.summary (name, used range, header
row, selection) through the port: standalone the overlay answers it,
under the WM the bus does.
The local model no longer loads on a registration: its session starts
on the first user line, so an app joining the bus costs nothing.
Verified by hidden grabs: sheets standalone /ai?on=1 → the overlay with
the summary card and no model; F10 closes it; under the WM the sheets
tile shows no overlay and F10 toggles the WM pane. widgets 135 + 4
(two widget_tree find_within tests fail before and after this change),
aichat 33, sheets 83 + 2, services 29; widgets and aichat check for
wasm32.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
A partial route to the nearest reachable cell, a hold at its end, and
re-planning with backoff (15 to 240 ticks) replace the old head-
straight-at-it fallback; cell derivation prefilters entities per chunk.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The live decoder never produced a frame on the body node: MF_E_TRANSFORM_STREAM_CHANGE
was hand-derived as 0xC00D6D60, which is MF_E_TRANSFORM_TYPE_NOT_SET; the real stream
change (0xC00D6D61) was treated as a hard error, the output type was only negotiated
lazily, and without MF_LOW_LATENCY the decoder holds a reorder window a 2-3 frame live
pipeline never fills. Set MF_LOW_LATENCY on the transform, commit NV12 before the first
ProcessOutput, re-negotiate on TYPE_NOT_SET/STREAM_CHANGE/BUFFERTOOSMALL, drain on
NOTACCEPTING, and stamp packets with monotonic 100 ns timestamps. MAKEPAD_H264_DEBUG=<file>
traces packets, HRESULTs, negotiations and frames for headless services.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
