Turn media into evidence your agent can cite.

Local videos, recordings, audio, captions, images, filename-ordered carousels, chat exports, articles, and RSS/Atom feeds share the inspect → preview → read discipline. Supported public video URLs include individual YouTube links; platform access varies. Approved public article and feed fetches create local evidence too. Optional screenshot sidecars preserve capture scope and verify image integrity.

Keep timestamped frames, transcripts, ordered images or entries, and manifests an agent can cite. The CLI and bundled skill preserve separate cloud and model-download approvals. These are the reading foundations; saved-account workflows have their own verification gates.

Instagram triage, public Substack/RSS capture, selected article and podcast reads with verified notes, public YouTube selection and notes, and local recording-inbox processing have implementations in the repository. They remain development-only while independent skill, harness, and real-workflow acceptance is completed. Generated files and passing helper tests alone do not prove every agent environment works.

Public-feed verification includes a real podcast and one subscriber-only article correctly recorded as skipped. That does not establish general paywall detection. LinkedIn has local observation capture, verified notes and a project skill; independent harness acceptance remains, and any future live platform work needs a documented approved route. The current focus is verification: reliable browser access, grounded notes, safe resume behavior, installer checks, and release evidence. Follow the current acceptance milestone.

Build on what exists: a cloud-disabled local inbox schedule, a documented approved route before any future LinkedIn acquisition, and browser pairing and permission design in the separate bridge. A Windows inbox schedule plan can be reviewed and manually registered; it does not create a task by itself. Each is a separate workstream with its own acceptance gates.

A development-only private YouTube queue helper uses the official Data API and a user-owned playlist, not Watch Later. Its live account flow still needs verification. Easy cron setup is not shipped. See the next milestone; no delivery date is promised.

X bookmarks, TikTok favorites, Reddit saves, newsletter collections, more audio adapters, visual change reports, capture benchmarks, multi-model workflows, and a hosted edition are longer-term directions. Each saved source needs its own platform and permission review; public article reading today does not imply private collection capture.

Visual jobs could check an approved page for layout changes or prepare a research digest. First come allowlists, budgets, timeouts, bounded retries, evidence retention, pause controls, and meaningful-change notifications. Iris remains STOPPED for integration review, not because malware has been established: static scan warnings are not evidence of malware. The Codex plugin remains DO NOT INSTALL. Security review gates still apply; neither is an installed capability.

The bright line: Voidscape reads. FollowerManager relationship audits, creator analytics, and LinkedIn job-search tools are separate companion directions, not Voidscape features. Permission to read never implies following, unfollowing, messaging, or publishing.

Explore the later milestone and its dependencies. The offline follower audit is waiting for selected export files.

Simplest tool wins. Deterministic code extracts and compares, authorized APIs fetch structured data, and approved browser tools read visible pages. AI joins only when interpretation adds value — captions before transcription, local before paid.

Future automation stays visible by design: history in local files or SQLite, capped compute and API spend, and a clear stop when access expires. Subscription access is not API credit, and installing Voidscape creates no scheduled job and no account permission.

Not by default. Local reads stay on your machine. Cloud transcription and first model downloads each require a current, separate approval.

Selected frames, ordered text or a transcript, and a manifest that maps every citation back to the source.

Video, audio, images and local carousels, articles, RSS/Atom, supported public video URLs, and explicit on-demand screenshots or short clips.