This index lists every module currently present under seen_std/src. Detailed pages cover the stable/user-facing families; lower-level platform and runtime modules are listed here so missing API-reference coverage is visible.
The separately versioned seen/seen-tokenizers package lives under packages/seen_tokenizers. It provides bounded strict tokenizer JSON, reproducibly generated Unicode 16.0.0 classification tables, deterministic GPT byte-level BPE, validated decoding, and explicit special-token policies. Its byte-offset and Unicode-codepoint APIs are intentionally distinct.
| Family | Purpose |
|---|---|
async | Future and async runtime helpers |
audio | Audio backend/device wrappers |
collections | Collection types, buffers, maps, sets, pools |
core | Core types, conversion, reflection, binary helpers |
crypto | Hashing utilities |
env | Environment helpers |
ffi | C type and interop helpers |
framework | Component, routing, middleware, store, VDOM, hot reload |
fs | Raw paths, move-only files, directories, links, durability, and traversal |
graphics | GPU, renderer, shader helpers |
hash | Hash helpers |
input | Gamepad/input abstractions |
io | File, buffered, and stdio helpers |
json | JSON values, parser, builder |
math | Math helpers |
memory | Allocation budget, statistics, and region helpers |
net | Polling and TCP wrappers |
platform | OS-specific windows/audio/input/GPU bindings |
process | Process execution helpers |
random | Random number generators |
scripting | Lua integration |
security | TEE/enclave helpers |
simd | SIMD vector/math helpers |
str | Character, escape, and string utilities |
sync | Mutexes, atomics, channels, queues, barriers |
thread | Threads, affinity, worker pools |
time | Time and duration helpers |
uww | Deterministic UWW/fixed-point helpers |
asyncasync/futureasync/runtimeaudioaudio/audioaudio/openalcollectionscollections/bit_setcollections/btree_mapcollections/btree_setcollections/byte_buffercollections/hash/modcollections/hash_mapcollections/hashsetcollections/linked_listcollections/list_utilscollections/mapcollections/poolcollections/std_stringcollections/string_buffercollections/string_hash_mapcollections/veccollections/vecdequecorecore/binarycore/bitfieldcore/compresscore/convertcore/intrinsicscore/json_derivecore/optioncore/ordcore/packetcore/preludecore/reflectcore/resultcore/to_stringcore/unitcryptocrypto/md5crypto/sha256envenv/envffiffi/c_typesffi/cinteropframeworkframework/componentframework/executorframework/hotreloadframework/middlewareframework/routingframework/storeframework/vdomframework/vdom_litefsfs/filefs/modfs/operationsfs/pathfs/typesgraphicsgraphics/gpugraphics/renderergraphics/shaderhashhash/modinputinput/gamepadinput/gamepad_unifiedioio/bufferedio/bytesio/directio/fileio/stdioio/bytes defines the native ByteBuffer, ByteSlice, and MutableByteSlice ownership contract. Byte ranges use checked nonnegative 64-bit Int geometry, including offsets above 4 GiB. Immutable views share an owner generation; mutable views are exclusive. Closing views and releasing the owner are idempotent, and any view whose owner was released is rejected with the stable byte.invalid code. Cancellation-aware acquisition reports byte.cancelled. See seen_std/examples/byte_slices.seen for a compiling example.
ByteBuffer also provides bounded growable storage. Call tryByteBufferWithCapacity or tryZeroedByteBuffer to select a checked logical capacity and power-of-two alignment up to 32 bytes. reserve, push, and append reject growth while any view is borrowed, preserve existing bytes, and cap every allocation at BYTE_MAX_CAPACITY. Native Seen arrays are 32-byte aligned by the runtime; smaller requested alignments are compatibility requirements satisfied by that storage. toArray is the explicit copying escape hatch; slices remain generation-checked zero-copy views.
Synchronous byte I/O is specified by the Reader, Writer, Seeker, BufRead, and Flush traits. The native Seen ByteReader and ByteWriter implement the same concrete contracts without foreign policy: each operation accepts an OperationContext, returns exact partial progress, rejects stale views and checked-limit overflow, and reports cancellation or expiration before side effects. ByteWriter is created with tryByteWriterWithLimit; it never grows beyond the caller's bound. See seen_std/examples/byte_streams.seen.
Bounded adapters include LimitedReader, CountingReader, TeeReader, HashingWriter, BufferedReader, and BufferedWriter. copyLimited uses a caller-bounded scratch buffer and returns CopyResult with separate read and write progress, an explicit limit marker, and completion state. Tee, hash, and copy accounting covers only bytes actually accepted; cancellation is checked before progress and every owned adapter closes idempotently. See seen_std/examples/byte_adapters.seen.
UTF-8 codecs provide strict validation, lossy U+FFFD replacement, bounded encoding, and cancellation before side effects. Endian codecs read and write checked 16-, 32-, and 64-bit integer storage in both byte orders. Float16, BFloat16, and Float64 helpers deliberately preserve raw IEEE bit patterns so NaN payloads and signed zero round-trip exactly; scalar interpretation belongs to the scalar type layer. Every codec validates its complete range before mutation. See seen_std/examples/byte_codecs.seen.
Asynchronous byte I/O is specified by AsyncReader, AsyncWriter, AsyncSeeker, AsyncBufRead, and AsyncFlush. The native Seen AsyncByteReader and AsyncByteWriter preserve checked borrowed spans, nonnegative 64-bit progress, exact partial completion, cancellation before side effects, and idempotent cleanup. Memory-backed operations are direct-ready coroutines; readiness scheduling and operating-system queues remain the responsibility of backend implementations. See seen_std/examples/async_byte_streams.seen.
reactorreactor/modreactor/reactorThe reactor provides move-only native handles, generation-safe registrations, finite event batches, prompt wakeups, monotonic timers, fair diagnostics, and a fixed-worker bounded blocking pool. Linux epoll, macOS kqueue readiness, and Windows IOCP completion share one event contract while retaining explicit capability and unsupported-platform diagnostics. See docs/api-reference/reactor.md.
jsonjson/builderjson/canonicaljson/modjson/parserjson/strictjson/valueformatsformats/safetensorsinferenceinference/scalarsinference/spanmathmath/mathmemorymemory/aligned_buffermemory/allocationmemory/mapped_regionmemory/pool_regionmemory/stack_regionacceleratoraccelerator/cuda/error (explicit Linux x86-64 CUDA dependency)accelerator/cuda/mod (experimental hardware maturity)CudaStream.borrowLaunchToken() returns a non-owning CudaStreamLaunchToken for one immediately nested model-kernel adapter call. The token is generation- and device-checked, supports graph capture, and does not authorize retention, synchronization, destruction, or device changes.
netnet/pollnet/tcpplatformplatform/darwin/cocoaplatform/darwin/coreaudioplatform/darwin/gamecontrollerplatform/darwin/metalplatform/darwin/sdl3platform/darwin/windowplatform/linux/alsaplatform/linux/evdevplatform/linux/libinputplatform/linux/pipewireplatform/linux/sdl3platform/linux/steamplatform/linux/steam_wrapperplatform/linux/vulkanplatform/linux/waylandplatform/linux/windowplatform/linux/x11platform/web/webgpuplatform/windows/win32platform/windows/xinputprocessprocess/processrandomrandom/rngscriptingscripting/luasecuritysecurity/enclavesimdsimd/simd_mathsimd/simd_typesstrstr/charstr/escapestr/stringsyncsync/atomicsync/atomic_queuesync/atomic_stacksync/barriersync/channelsync/mpsc_queuesync/mutexsync/orderingsync/rwlocksync/spsc_queuesync/thread_localthreadthread/affinitythread/modthread/pooltimetime/backofftime/formattime/modtime/timetime/timeruwwuww/fixeduww/mod