YouTube URLs are one supported direct-source path. Harmonia resolves source metadata and media only after the operator attests that the material is authorized for processing.
How Harmonia uses it
The source registry validates the URL and records source rights. During extract_sources, the worker resolves public metadata through YouTube/oEmbed, obtains authorized media, and uses Amazon Transcribe to produce digest-bound timed spoken segments, with Nova for separate visual evidence. Those segments and authorized video evidence join the same normalized source bundle used for documents, web pages, pasted text, uploads, and pinned brand-library snapshots.
Google OAuth configuration is available where an authorized source requires it. Public URLs do not grant reuse rights by themselves; the job retains the operator’s rights declaration and source provenance.
Security and authority
Source extraction cannot authorize publication. Models receive only the bounded normalized source package and may not claim facts outside its segment/frame evidence. Credentials and download details remain server-side.
Failure behavior
Invalid URLs, unavailable or private media, unsupported formats, rights failures, metadata mismatch, and download/transcription errors remain visible. Harmonia does not substitute fake transcripts.
Evidence status
URL validation, rights records, local fixtures, and stage contracts have automated coverage. A real end-to-end source claim requires the captured authorized URL, resolved metadata, transcript, and source digest.
Relevant files
agent/harmonia_agent/youtube.pyagent/harmonia_agent/content.pyagent/harmonia_agent/stages.pysrc/lib/sourceRights.tsdocs/evidence-runbook.mdx
Last modified on September 14, 2026
