Documentation

Install from source (primary)

Dependencies are declared as ranges so TekMyra can live inside your application; ranges float, so the published figures were measured against the exact versions recorded in constraints-reproduce.txt. Install against the constraints file if you want to reproduce the numbers rather than only install the library:

pip install -c constraints-reproduce.txt -e .

pip install -e '.[dev]'   # the library plus pytest, to run the suite below

If you do not want the test dependencies, pip install -e . installs the library alone.

Core dependencies are click, numpy, pydantic, scikit-learn, scipy, tiktoken and pyyaml. tiktoken is a core dependency, not an extra: token-reduction figures are unobtainable without it.

Install from PyPI

pip install tekmyra-core

Getting a model

Trained model artifacts are not in the repository and never will be: they are the commercial half. The reference artifact bundle is published as a release asset, tekmyra-reference-artifacts.tar.gz, with its SHA-256 printed in the release notes. Download it, check the digest before extracting, and unpack it at the root of your clone:

curl -L -O https://github.com/laconiq-ai/tekmyra/releases/latest/download/tekmyra-reference-artifacts.tar.gz
shasum -a 256 tekmyra-reference-artifacts.tar.gz   # compare against the release notes
tar xzf tekmyra-reference-artifacts.tar.gz         # unpacks .cache/compressor/1.0.0 and .cache/router/1.0.0

The artifact directories are resolved in this order, and a failure prints every location it tried: $TEKMYRA_ARTIFACT_ROOT/.cache/… if that variable is set, then .cache/… under your working directory, then .cache/… beside the installed package. Details on what the bundle contains and what it does and does not unlock are in the reference bundle section.

Run the benchmark

python -m tekmyra.benchmark compressor --fixture-dir benchmarks/fixtures/synthetic
python -m tekmyra.benchmark compressor --fixture-dir benchmarks/fixtures/long_context_v1
python -m pytest tests/test_public_corpus_notice.py   # public bytes + attribution
python -m pytest tests                    # the suite, green without a model
python scripts/verify_open_core.py        # exits 6 on a clean clone: seam proven, numbers await the bundle

The test suite passes without a model: the tests that need a trained artifact skip, and each skip names every location that was searched.

Without a model, the benchmark refuses to give you a number

That is deliberate rather than a bug, and it follows the same rule as the verifier:

Error: Learned router artifact did not load — refusing to produce a benchmark
number. artifact_dir='<your clone>/.cache/router/1.0.0' exists=False
backend='none'.
could not resolve '.cache/router/1.0.0'; searched, in order:
  absent  <your clone>/.cache/router/1.0.0
Set $TEKMYRA_ARTIFACT_ROOT to the directory holding .cache/ if your artifacts
live elsewhere. Re-run with --allow-rules-only to record an explicitly
rules-only run; such a run cannot emit a headline compression ratio.

Take the escape hatch it offers and the run records exactly what it was: the headline is withheld (not zero and not one, withheld), token reduction is 0.00%, and 68 of 68 locked spans are still preserved across 28 reached fixtures. The safety machinery works without a model. The compression does not.

Verify before extracting

shasum -a 256 tekmyra-reference-artifacts.tar.gz
9c185a524b24693213cd2a226021e466da9905fb29ab5582c8cc7480abb69883  tekmyra-reference-artifacts.tar.gz

The digest above is the one printed in the launch release’s notes (the asset is 5,879,184 bytes). The authoritative record is always the notes of the release you downloaded from: the digest is published beside the asset rather than pinned in the repository, so that you check the artifact you actually have against the record for that release, not against a constant a later release would silently falsify.

What reproduces without it

The four public corpora (benchmarks/fixtures/public_{domain,code,diff,json}_v1) ship in the repository. A clean clone, with no bundle, can verify that their fixture bytes match the committed baseline and that every third-party source is covered by NOTICE:

python -m pytest tests/test_public_corpus_notice.py

What needs it

The headline public-corpus figure is 62.1458% effective byte reduction over 138 fixtures (2,708,761 → 1,025,379 bytes), with 6 refusals (4.35% of fixtures, 3.55% of bytes) contributing zero saving; the on-accepted diagnostic is 64.4356%. The headline is a byte-basis measurement and is not comparable to the ratio and token figures in the quickstart table. Reproducing it requires the reference artifacts: the published bundle carries the versioned 1.0.0 directories the code loads (the baseline artifact, tests/baselines/public_corpus_baseline.json, additionally records a digest over the full private artifact directory; the README's evidence section states exactly which of its digests verify against what you have). The ratchet suite proves the rate reproduces from the bundle byte-exactly. With the bundle unpacked, run:

python -m pytest tests/test_public_corpus_ratchet.py

A clean checkout has no artifacts, so the ratchet’s rate tests skip, visibly and with the fetch instructions in the skip reason: never a silent pass, and never a failure for following the README. The corpus-integrity tests still run and still bind the fixture bytes. The headline benchmark ratios in the quickstart table need the bundle for the same reason; without it the benchmark refuses to emit a headline rather than printing a flattering 1.000.