SparkLang · product loop
One Spark-native path from a reviewable
.spark file to
SPARK_BC, local dump, owned
tiny coder steps, helpers on the wire, and docs on Pages —
CLI first, GUI where it helps.
Inspired by clear CLI→inspect→publish loops elsewhere — rebuilt around our binaries, dumps, and measured model claims. No third-party branding on this page.
Author a .spark program,
then
./spark-bootstrap --compile
(or open an existing pack). Output is Spark’s own
SPARK_BC — orchestration
ops like TRAIN /
STEP, not a weight dump.
Compile docs · Builder · Opcodes / ISA
# Produce SPARK_BC from a Spark program
./spark-bootstrap --compile examples/spark_train_step.spark \
-o out/spark-train-step.sparkbc
Primary SoT:
tools/spark-bc-dump/dump.py
— hex, header, mnemonics. One-shot project loop:
./helpers/spark-analyze
→ local
out/analyze/<name>/
(dump, ops list, REPORT stub; optional
--serve /
--ask). Optional
spark-bc-gui for a
readable pane. Deterministic inspect first; LLM research
is a separate, cited lane.
Decompile docs · Compete / scoreboard · Tools & helpers · Methods vs OpenBin · LLM research notes
# Local analysis folder — no upload
./helpers/spark-analyze docs/examples/spark-train-step.sparkbc
./helpers/spark-analyze examples/spark_train_step.spark --ask
After inspect, ask in text or speech — ears (STT) → dump context + owned TinyCoder → speaking (TTS). Local only; no OpenBin login. TinyCoder is tiny: dump facts stay SoT; open-ended RE answers carry a status note.
Voice ask docs · Voice / STT / TTS · Model aspects
# Text ask (no mic)
./spark-ask docs/examples/spark-train-step.sparkbc \
--text "What opcodes are in this dump?"
# Voice loop — dry stub for CI; --mic for live
./spark-speak-ask docs/examples/spark-train-step.sparkbc --dry
Run model train →
model status →
expect in one file.
Owned spark-coder is a
tiny capability — useful for Spark
factory loops, not a claim to beat frontier models.
Train voice / STT / TTS in 3 steps —
Voice easy
(./spark-voice easy):
tiny CI default or
--scale large on a
consumer GPU (or CPU). Owned heads, trained locally.
Train loop · Spark coder · Voice easy · Model training
# Dry-run — fixtures, no keys
./spark --dry-run examples/train_eval.spark
# Voice easy (tiny / CI)
./spark-voice easy --dry
# Large opt-in
./spark-voice easy --scale large --device auto
Local HTTP surfaces for helpers and shadows — wire a
trainer at
SPARK_TRAIN_URL, probe
serve APIs, keep dry-run as the default gate.
Ship factory docs and examples on sparklang.dev (Cloudflare Pages). Share packs and captures from the gallery below — Spark-appropriate examples, not a malware community feed.
Curated Spark programs and captures you can dry-run or open in the dump tools.
Minimal CI gate with fixture artifacts.
TRAIN / STEP / TRAIN_STATUS in bytecode — dump with spark-bc-dump.
Hex + header from examples/spark_train_step.sparkbc.
Tiny owned coding model notes — capability, not hype.
Edit and dry-run sample programs in the browser.
Compile / decompile / helpers paths as Spark SVGs.