Roadmap.

The highest-priority items — either underway now or first in line.

• Live cost estimate during generation. The pre-generation cost range updates in real time as the script streams in, instead of relying on a fixed worst-case margin.
• An optional way to clear out old drafts. A setting (off by default) to clean up drafts you started but never finished, past an age you choose — with a confirmation before anything is actually removed. Nothing happens automatically unless you turn it on.
• Smarter duration limits for local models. Today, using a local model caps you at a shorter episode length as a blanket safety margin. This checks the actual model's context window instead, so a capable local model isn't held back unnecessarily.
• More resilient script generation. If the model declines to turn dense or sensitive source material into a two-voice dialogue, Aludel offers to fall back to a plain Summary instead of just failing.
• More reliable project/channel importing. Re-importing a project or channel now correctly matches it back to your existing LLM/TTS providers and voices, even when they're saved under a slightly different name.
• Publish directly to Apple Podcasts and Spotify. Not just export — Aludel generates a valid RSS feed and, for self-hosters, uploads it and the episode audio to your own server. The biggest missing piece between "I made a podcast" and "it's live."
• A command-line version of Aludel. Same six-stage pipeline, same license, no GUI — configurable providers, intro/outro music, JSON output for scripting. Available now via Homebrew or from source; see the CLI docs for what's there and what's still missing (Windows/Linux, shell completions/man pages).

Real ideas, not yet scheduled — priorities shift as the ones above land.

• Music library polish. Retagging, renaming, and reordering your imported intro/outro/background tracks without leaving the library view.
• Import a script you already wrote. Skip AI script generation entirely — bring your own dialogue, summary, or audiobook narration in a defined template and go straight to voice generation.
• Audiobook-style export for long, multi-chapter sources. Group chapters auto-split from a long book back into one unit, and export it to an Audiobookshelf-compatible library folder.
• A lightweight, listen-only build of Aludel. For sharing a finished episode with someone who just wants to press play with a synced transcript — no generation pipeline, no provider setup.
• Multiple languages from a single job. Generate more than one output-language rendition of the same episode in one pass.
• Native Apple Silicon voice engine. Removing the current Rosetta 2 dependency for local text-to-speech, once a fully native build is available upstream.