Release 0.0.8: Better Search, Refined Editors, and Pipeline Improvements
- Steven Fusco
- Technology , Updates
- November 13, 2026
Table of Contents
ELEMENT.FM 0.0.8 is out, and there’s quite a bit to cover. Since 0.0.7, we’ve shipped 152 commits across 14 repositories—from the screens you use to manage your shows to the services that import, process, and publish your episodes.
Here’s what’s changed.
Find episodes with better workspace search
Workspace search has a new interface that fits the rest of ELEMENT.FM’s dark, neon-accented design. Results show the episode title and its show, with clearer loading, empty, and error states.
The search itself now supports partial-word matches and typo tolerance, searching episode titles, show names, descriptions, and transcripts. Titles and show names receive more weight, helping put recognizable matches nearer the top.
You can focus search with Ctrl+/ or Cmd+/, navigate results with the arrow keys, press Enter to open an episode, and dismiss the results with Escape.
Refined workspaces, show editors, and episode editors
We’ve refreshed the workspace overview and the show and episode editing screens, bringing their layouts, controls, and status displays into a more consistent design.
The RSS sharing panel also has a clearer feed-copy control and a refreshed presentation for podcast-directory links.
Clearer Value for Value controls
The V4V interface has been reworked around funding amounts and recipient splits:
- Clearer recipient cards and controls for adding, removing, and adjusting splits.
- SATS and USD amount entry, backed by a server-side Bitcoin price endpoint.
- Visible loading and cached-rate states, with SATS entry still available if currency conversion is unavailable.
- Stronger validation of recipient fields and percentages before saving changes.
These changes refine the existing V4V workflow; they aren’t a new payment method or a pricing change.
Reworked summaries and automatic chapters
Summary generation now uses the OpenAI Responses API, with a structured final response for HTML show notes. Larger transcripts use token-aware chunks, ordered section notes, and bounded consolidation rather than treating every transcript the same way. The service can select a configured fallback model when the preferred model isn’t available.
Automatic chapter generation also uses structured responses. Long transcripts are split around WebVTT cues where possible, processed with bounded concurrency, and combined in their original order with chapter indexes reassigned consistently. Timestamp normalization and model-access fallback handling are included as well.
Both workers also share a single queue consumer per process, replacing the previous pattern of setting up separate consumers for each concurrent worker.
These are changes to processing and output handling—not a claim that generated content is always correct or that we’ve established a new speed benchmark.
Publishing and Podcasting 2.0 fixes
Several fixes address the boundary between editing an episode and making it available to listeners:
- RSS includes only published episodes, rather than every episode that isn’t a draft. An episode without publishable audio now causes an explicit export failure instead of silently being skipped.
- Unrated shows can be published, with fixes to preserve unrated content metadata through import and export.
- Chapter JSON uses
application/json+chapters, both in feed declarations and stored chapter files. - RSS output now includes an XML declaration, and feed uploads calculate their length using UTF-8 byte counts.
- Public-page sync failures are surfaced. If RSS publication succeeds but the public show page fails to synchronize, the interface displays a warning rather than hiding that distinction. CMS synchronization has bounded retries, and the CMS accepts bounded larger show payloads.
Safer imports and audio processing
RSS, artwork, audio-enclosure, and video downloads received stricter destination checks and isolated outbound download paths. Feed redirects are checked before each connection, and proxy credentials are redacted from downloader failures.
Audio imports now use the same validated upload path as direct uploads. MP3 inputs must pass header, stream, duration, and decode checks before storage updates. Media tools are restricted to local MP3 inputs, and staged audio is checked before transcription. Remote transcript-result downloads also receive destination validation.
Transcription submissions are now limited to 10 per hour per user, across workspaces and credential types. We also fixed a case where the feed-import consumer could stop after reporting an error.
Workspace, account, and API safeguards
This release includes a substantial set of access-control and request-handling changes:
- Tighter workspace boundaries for show operations, analytics queries, and live job-progress subscriptions.
- Signed, short-lived capabilities for realtime subscriptions instead of accepting raw room identifiers.
- Hashed API-key storage, one-time display of newly created keys, and more careful handling of sensitive logs.
- Explicit confirmation when accepting workspace invitations, same-origin POST logout, and allowlisted browser API origins.
- Atomic workspace-deletion and invitation-revocation changes.
- Stronger job-completion callback checks, malformed-JSON handling, and input validation.
- Ordered, deduplicated subscription webhooks, plus safer show-mailbox management and invitation email handling.
Managed mailboxes now follow a one-address-per-show rule, without a separate per-user mailbox cap.
Redis authentication and workload isolation were tightened as well, alongside dependency updates across the services. These are specific safeguards and fixes, not a claim that any system is completely secure.
RustFS support for self-hosted stacks
The local and hardened staging Compose stacks now use RustFS instead of a MinIO server. Application storage configuration prefers generic S3_* settings while retaining compatibility fallbacks during migration.
The release includes S3-to-S3 copy tooling and a guarded migration runbook. This is not a claim that the hosted production CDN was moved to RustFS, and an existing MinIO data directory should not simply be attached to a RustFS container.
For self-hosters, review the storage configuration and migration instructions before upgrading an existing installation.
Better visibility behind the scenes
We’ve added OpenTelemetry tracing across all 13 application services, with internal HTTP and message-queue context propagation. Traces are stored in a privately accessible SigNoz installation alongside Kubernetes metrics and events.
The new tracing is limited to technical metadata such as timings, service relationships, route templates, and status. It does not export audio, transcripts, prompts, request bodies, authorization headers, SQL statements, or exception payloads into traces. This work doesn’t add browser tracing or blanket application-log collection.
New root traces are sampled at 10% by default. We verified stored traces from a production frontend-to-CMS request and ingestion checks for all 13 service names. That is not an end-to-end certification of every authenticated processing workflow.
Feedback
If you run into an issue or have a suggestion:
- Join our Matrix chat.
- Report it on GitLab.
- Email webmaster@element.fm.
Thanks for using ELEMENT.FM and helping us improve it.