Public technical diligence · verified 2026-09-14

Inspect the boundary between what works and what is still a hypothesis.

LangBeat is a pre-revenue, single-founder music system. This brief makes the public diligence path useful without pretending a private repository, a projection, or a test count is customer traction.

Request private repository access Open the live product

Implemented

Four surfaces a reviewer can ask to inspect

Six-stem separation

Vocals, drums, bass, guitar, piano, and FX are produced by the separation pipeline.

modal/stem_separator.py · FINAL_STEMS
Transition invariants

The renderer is tested for downbeat alignment, low-band headroom, and true-peak limits.

scripts/transition-render.test.mjs
Bounded interactive separation

Interactive stem separation is capped at four concurrent containers. Other GPU functions maintain separate limits.

modal/gpu_budget.py:87 · modal/stem_separator.py:815
TypeScript MCP server

The repository includes MCP tools for stems, sessions, arrangements, harmony, synthesis, and channel control.

mcp/langbeat-mcp/src/tools/

Current source count

1,624 assertion calls across 101 test files

That number is a source count of check( and assert. matches in scripts/*.test.mjs, not a claim that every test was executed for this pitch-deck release. The prior published count—507 across 36—became stale as the product repository expanded, which is why the count is dated.

What this brief does not claim

Revenue to date is zero. The $19 price and customer counts in the deck are modeling inputs, not observed willingness to pay. The product does not claim complete web/iOS feature parity. No per-track GPU cost or gross margin is quoted without a billing export. A previous 725-track library figure was removed because the public claims file did not contain a reproducible aggregate receipt.

Diligence sequence

A useful first review

  1. Use the live browser product and record which controls produce an audible or persistent result.
  2. Request repository access and reproduce the cited source locations.
  3. Choose a representative subset of transition and persistence tests to execute.
  4. Review the standing dead-UI checklist and known credential-dependent integrations.
  5. Treat pricing, conversion, retention, and margin as hypotheses until cohorts and billing exports exist.