Skip to content

Oathra v0.1.2 — standalone transcript checks

Choose a tag to compare

View all tags
@FORIFOR FORIFOR released this 13 Sep 18:48
v0.1.2
1f056cb

Superseded release: The current public asset is v0.1.10, which adds the consent-based intake demo and the latest setup path. Use the current release command before starting a new installation.

Oathra v0.1.2 — check the transcripts you already have

Add completion checks to an existing voice agent without moving your carrier or model. This release adds a local transcript CLI and a standalone TypeScript SDK, and fixes an English time ambiguity that could incorrectly verify 7:30 as 07:30.

Install

Node.js 22+:

npm install https://github.com/FORIFOR/oathra/releases/download/v0.1.2/oathra-0.1.2.tgz
npx oathra verify ./transcript-check.json

For the browser simulator:

npx --yes --package=https://github.com/FORIFOR/oathra/releases/download/v0.1.2/oathra-0.1.2.tgz oathra demo

The npm registry still serves 0.1.0; use the asset above for this release. English integration guide · 日本語の導入手順.

What changed

  • oathra verify <json | -> checks final transcripts locally, returns JSON evidence and uses exit codes 0 (complete), 2 (not complete) and 1 (invalid input/error). It rejects empty requirements, duplicate IDs, reversed timestamps and invalid reference dates.
  • oathra and oathra/evidence now export EvidenceEngine, defineCall, evaluate and verifyTranscript with bundled TypeScript declarations. Prior package metadata pointed to SDK files absent from the tarball.
  • English bare clock times such as 7:30 remain unresolved. Explicit am/pm or an unambiguous 24-hour time and renewed agreement are required. We deliberately do not infer meridiem from earlier conversation.
  • A LiveKit integration function observes committed conversation items. It treats interrupted speech as requiring review. An optional RTC peer conflict with current LiveKit Agents is resolved.
  • Japanese/English site and Zenn articles now link directly to using your own transcripts, alongside the browser demo.

Validation and limits

  • 138 tests passed; 1 live connection test skipped. New regression coverage uses the observed issue #8 and the existing public GPT-4o mini/simulator recording, not invented call data.
  • The installed tarball reproduces the saved record's fields. The actual prefix before the first verified confirmation returns incomplete (exit 2); empty stdin returns exit 1. Root and subpath SDK results match the CLI.
  • Clean installation alongside @livekit/agents@1.8.1 succeeds without --force or --legacy-peer-deps. The integration example and existing RTC wrapper type-check with installed @livekit/rtc-node@0.13.35.
  • Dependency checks, site TypeScript/build and scenario validation pass. Existing scenario evaluation: 0 false completions. Existing adversarial simulator: 0 / 10,000 false completions; this is not a real-call benchmark.
  • Real browser: the reported English sequence stays incomplete and explains the missing meridiem. Japanese mobile integration section is readable without horizontal overflow.

LiveKit session behavior, interruptions over a live connection and PSTN are not verified by this release. The checker implements supported Japanese/English rules; it does not inspect a venue's booking database or prove general semantic correctness. No new paid phone calls were placed, and the 100-call readiness evaluation remains outstanding.

Asset: oathra-0.1.2.tgz, 173,992 bytes, 28 files including Apache-2.0 license and SDK declarations. SHA256: 8a406f39a3ce2ce95b50463060cdc9bf8b942ee2ddab3ab367640da751bf312e.

Acquisition follow-up

The measured bottleneck and next experiments are recorded separately. Stars are still 0 at the starting snapshot. A 12-hour follow-up is active to inspect external signals, fix reported friction and change ineffective tactics. Shipping this release does not establish external adoption or business demand.

Assets 4
Loading