IDStream
Stable conversion between token IDs, packed integer streams, arrays and deterministic receipts.
Toolbox / local-first
Tools for inspectable pipelines.
The OKC toolbox collects small reversible utilities around compact representation, corpus inspection, tokenizer pressure, binary/text roundtrips and benchmark economics.
$ okc inspect corpus.txt
bytes, chars, ids, rare ids, structure
$ okc pack stream.u32
bit-level transform + roundtrip receipt
$ okc compare baseline
ratio · speed · memory · restore
Modules
Small surfaces. Serious questions.
Each tool exists to make a representation or claim easier to inspect, reproduce and price.
Rengrams
Repeated symbolic-structure scanning with deterministic outputs other modules can consume.
Sentinel Lite
Macro stream replacement and restoration: input IDs → macros → restored IDs → verification.
Redumb / DUMB
Reversible baseline transforms for exact reconstruction, bitpacking and reproducibility checks.
Vocabulary pressure
Detect singleton/rare IDs, inspect active vocabulary pressure and produce readable report tables.
TCN notes
Small neural-prediction experiments over cleaner reversible streams, with claims kept deliberately narrow.
Benchmark lab
Measure the economic outcome.
Compression is not just a smaller file. Track bytes transferred, decode throughput, peak memory, metadata, random access, quality risk and the hardware cost a customer might avoid.
Request a measured pilot ↗$ okc report workload
baseline → transform → codec
ratio: recorded
decode: recorded
memory: recorded
restore: required
No receipt, no claim.
Commercial boundary
Free surface, paid work.
Independent human use is free under project terms. Corporate evaluation, private runtimes, SaaS, internal automation, resale, integration or custom research requires a defined agreement where marked. Packaged artifacts may follow when stable; there is no paywall around the public proof trail.