neosloc

sloccount counted lines and priced them with COCOMO. With LLMs writing code, lines are cheap. neosloc measures what still costs something: how easily other systems and agents can use a project, what it is worth, and whether to adopt it or rebuild it.

Documentation Archive GitHub PyPI JSON Schema

pip install neosloc        # or: pipx install neosloc / uv tool install neosloc
neosloc path/to/repo

What it reports

Ten dimensions

Interface, stability, events, identity, portability, agent ergonomics, embeddability, extensibility, observability and change legibility, each a ladder of checkable requirements.

Per surface

Services, CLIs, libraries, desktop apps and frontends are judged on their own terms; dimensions that can't apply are n/a, not zero.

Make or buy

Tokens, agent hours and human days to rebuild it with agents, against minutes to adopt it as it is.

neoCOCOMO

What a codebase is worth when agents can re-type it, and how much of it lives only in its history.

LLM evaluators

A probe tries real integration tasks from the docs; a judge and a review panel check the results. Anthropic or OpenRouter.

Built for scripts

JSON output and errors that follow a published schema, typed exit statuses, structured logs. No dependencies, Python 3.8+.

The archive

The neosloc archive evaluates public repositories every week and keeps every result: ladders, gaps, make-or-buy and history for each one. Ask for a repository to be added.

Example: a Redis-in, MQTT-out geo server

neosloc 0.7.0 on geomqtt (excerpt):

Surfaces: service

Interface surface   ■■■■  4 exemplary  A second surface: another transport (HTTP and a protocol, two protocols), ...
Events              ■■■□  3 solid      Signed webhooks registrable through the API, or two mechanisms (webhooks, streams, a broker).
Agent ergonomics    ■□□□  1 ad hoc     Errors are machine-readable: ... a standard protocol's native error replies ...
Embeddability       ■■■■  4 exemplary  Composition artifacts (compose, Helm, IaC).
Integrability index (0-4, 9 applicable dimensions):  1.78

Make or buy (over 3 years)
  Make with agents:  177k output + 4.4M input tokens (85% cached) on anthropic:claude-opus-5-5 = $7;
                     2.0 agent-hours, 3.6 human days (2.1 steering + 1.5 rediscovering history)
  Buy (adopt as is): 5 min to running (via service) + 2 min to first use (via service),
                     4.0 h/year of upgrades -> $812
  Verdict:           BUY: adopting this is cheaper (make/buy = 3.48); buying wins below $671/year