Changelog

CHANGELOG

Version 0.7.3 (2026-05-27)

UI rebuild — 3-way theme toggle

  • Replaced the single icon theme button in the header with a sliding 3-way toggle: System / Light / Dark

  • Active selection indicated by a sliding pill with per-mode icon colors:

    • Dark mode: black pill, white (system), amber-400 (sun & moon)

    • Light mode: blue pill, black (system), amber-600 (sun), amber-400 (moon)

  • ThemeMode type ('system' | 'light' | 'dark') added to App.tsx; default is dark

  • System mode derives isDark from window.matchMedia('(prefers-color-scheme: dark)')

  • Rebuilt frontend bundle and copied static assets to src/audia/ui/static

Version 0.7.2 (2026-05-27)

UI rebuild only

  • Rebuilt frontend bundle and copied static assets to audia/src/audia/ui/static

  • No backend or logic changes in this release — this version solely captures the updated compiled UI

Version 0.7.1 (2026-05-10)

Code quality fixes

  • Fixed E402 lint errors in ui/routes/convert.py by moving _logger assignment to after all imports

  • Removed unused local variables (result, src_dirs, mock_browser) in ui/routes/library.py, tests/test_cli.py, and tests/test_text_cleaner.py (F841)

  • Wrapped long lines exceeding 100 chars in agents/research.py, agents/text_cleaner.py, config.py, ui/routes/convert.py, and tests/test_cli.py (E501)

Version 0.7.0 (2026-04-29)

Move studies between projects

Backend

  • New POST /api/library/papers/{paper_id}/move?project=<src> endpoint in library.py

    • Accepts { "target_project": "<name>" } body

    • Validates target project name via validate_project_name(); rejects same-source-and-target moves

    • Copies the PDF to ~/.audia/<target>/uploads/ and each linked audio file to ~/.audia/<target>/audio/, avoiding filename collisions with a {paper_id}_ prefix

    • Inserts new Paper + AudioFile rows in the target project’s SQLite database with the updated absolute file paths

    • Deletes source database records (cascade-deletes audio rows) and removes the original files from disk

    • Returns { status, source_project, target_project, new_paper_id }

Path repair utility (scripts/repair_paths.py)

  • New repair mode: scans a project’s DB for pdf_path / file_path values that no longer exist on disk and re-matches them by filename (with and without hash prefix) against files currently in the project’s uploads/ and audio/ directories

  • New recover-from mode (--recover-from <source>): migrates DB records from a source project whose file paths are broken into a target project by matching filenames against the target’s on-disk files — useful for recovering from manual file moves

  • Usage: .venv/bin/python scripts/repair_paths.py --project <name> [--recover-from <source>] [--dry-run]

Version 0.6.1 (2026-04-22)

UI enhancements

Version 0.5.1 (2026-04-19)

Bug fixes

PDF preview broken for saved papers after project-layer introduction

  • Sidebar-selected papers built their preview URL as /api/library/pdf/{id} without the ?project= query param, while the endpoint requires it to locate the file in the correct project subfolder

  • App.tsx now appends ?project=<name> to the sidebar preview URL (matching the existing behaviour in Main.tsx)

Renaming an audio file in the database editor only updated the display name

  • PATCH /api/library/audio/{id} updated filename in the DB but left the MP3 on disk untouched and file_path stale

  • The endpoint now renames the file on disk, preserving the original extension if omitted, and updates file_path in the same transaction

  • After a successful cell save, the database table re-fetches its rows so file_path reflects the rename immediately, and the sidebar refreshes via the existing onConverted / refreshKey mechanism

Version 0.5.0 (2026-04-18)

Breaking change — existing data stored under ~/.audia/ must be migrated manually. Move ~/.audia/audia.db, ~/.audia/audio/, ~/.audia/uploads/, and ~/.audia/debug/ into ~/.audia/default/ to preserve your current library under the default project.

Project-based storage

All files are now namespaced by project under ~/.audia/<project>/

Previously everything was written flat to ~/.audia/. The new layout keeps projects fully isolated — each has its own SQLite database and audio/upload/debug directories.

Backend

  • config.py: new DEFAULT_PROJECT = "default" constant; ProjectDirs dataclass bundles db_path, audio_dir, upload_dir, debug_dir for a given project root; validate_project_name() enforces lowercase-alphanumeric names; Settings.get_project_dirs(project) returns the correct ProjectDirs for any project; the legacy flat properties (db_path, audio_dir, etc.) now delegate to get_project_dirs("default")

  • storage/database.py: per-project engine/session-factory registry (_engines, _factories dicts keyed by project name); init_db(project) and get_session(project) both accept an optional project argument; engines are lazily created and cached

  • New GET /api/projects — list all projects with metadata (document count, audio count, disk size, creation date)

  • New POST /api/projects — create a named project

  • New DELETE /api/projects/{name} — delete a project and all its files (protected: default project cannot be deleted)

  • All existing routes (/api/library/*, /api/convert/*, /api/research/*, /api/settings) accept an optional ?project=<name> query parameter (or project form/body field for POST endpoints); omitting it selects the default project

CLI

  • audia convert gains --project / -p — output audio and database entries go to ~/.audia/<project>/; defaults to ~/.audia/default/

  • --output still works as an explicit override on top of the active project directory

Frontend

  • New DatabaseSelector component in the header — dropdown to list, create, and delete projects; active project is highlighted; destructive delete requires a confirmation step

  • activeProject state threaded from App through Header, Sidebar, Main, MainConvert, MainResearch, and MainDatabase; all API calls append ?project=<name> when a non-default project is active

  • Switching projects refreshes the sidebar library immediately

Tests

  • conftest.py: clear_settings_cache fixture (autouse) now sets AUDIA_DATA_DIR to pytest’s tmp_path and clears the engine/factory registry around every test — no test writes to ~/.audia/ any more

  • isolated_db fixture rewritten to inject into the new _engines/_factories dicts instead of the removed module-level _engine/_SessionLocal attributes

  • test_config.py: path assertions updated to reflect the new data_dir/default/ layout

Version 0.4.4 (2026-04-02)

Style: docs & frontend

  • Docs tables: darker cell borders so text is readable in both light and dark themes

  • Docs sidebar: expanded section background changed from RTD default light-gray to the theme dark background

  • Docs buttons (prev/next): purple background with white text; no outline on focus/click

  • Docs code blocks: border removed; left accent changed from purple to lime

  • Docs links: underline removed on hover and active states

  • Configuration diagram: fixed node ordering in the pipeline flow

  • Configuration diagram dropdown: added backdrop blur so the list is legible over the SVG

Tests: coverage 68% → 96%

  • New test_stt.py: covers _ensure_stt_deps, transcribe_file, _transcribe_array, record_and_transcribe (including KeyboardInterrupt path), and distill_search_query

  • New test_async_jobs.py: directly awaits _run_research_job (success, not-found, cancellation, exception, no-query) and exercises the enqueue_conversion background task via httpx.AsyncClient

  • Extended test_text_cleaner.py: Google LLM provider (import error, missing key, happy path, custom api_base); OpenAI/Anthropic happy paths with api_base; progress_cb path in llm_curate; clean_text alias

  • Extended test_research.py: HTML fallback search (parsing, max_results, empty page, API-error trigger, 429 trigger); paper.published = None edge case; HTTP error on download

  • Extended test_cli.py: “all” paper selection, out-of-range selection, download failure with manual path fallback, --open flag, no-subcommand invocation

Version 0.4.3 (2026-04-01)

Research tab: convert button improvements

  • “Convert to audio” button is now lime (matching the Convert tab), shows a spinner and “Converting…” label while jobs are running, and is disabled during conversion

  • A Cancel button appears beside it (not inside each job panel) while any jobs are running; clicking it cancels all active jobs at once

  • Button row stays visible throughout conversion so progress and cancel are always accessible

Version 0.4.2 (2026-04-01)

Bug fixes & UI improvements

LLM provider ignored during query normalisation

  • Root cause: NormalizeRequest only had a query field; the llm_provider and llm_model values sent by the frontend were silently dropped, so distill_search_query always fell back to the global .env defaults (Anthropic)

  • NormalizeRequest now accepts llm_provider and llm_model; the /api/research/normalize handler applies them to the settings object before building the LLM — consistent with how _run_research_job already handled overrides

  • The handler now inlines the LLM call directly (no longer delegates to distill_search_query) so provider/model overrides are guaranteed to take effect

Research sessions not saved to the database

  • Root cause: EnqueueRequest had no query field so there was nothing to store; _run_research_job saved Paper and AudioFile rows but never wrote a ResearchSession

  • EnqueueRequest and _run_research_job now accept an optional query parameter

  • After each job saves the paper, a ResearchSession row is written with the search query and the new paper_id

  • Frontend updated to include query: normalizedQuery ?? query in the enqueue payload

Database tab colour scheme

  • audio_files card/heading: violet → lime

  • user_settings card/heading: amber → purple

  • Amber removed from the colour palette entirely; lime and purple added

Version 0.4.1 (2026-04-01)

  • Set the display div for multiline cells to max-h-24 overflow-y-auto, so abstract (and query) cells will be capped at 6rem tall and scroll vertically when the content overflows

Version 0.4.0 (2026-04-01)

Bug fixes

Abstract column not editable in Database tab

  • Root cause: PDF-converted papers have abstract="" (empty string), which rendered as a zero-height invisible <span> with only 2 px of padding — effectively unclickable

  • Empty editable cells now display a dimmed italic (empty) placeholder so they are always visible and clickable

  • EditableCell wrapper gained min-h-[1.25rem] so the click target is never zero-height

  • CellValue now receives isEditable and isDark props to render the placeholder with correct theme colouring

TTS voice not persisted to user settings

  • Root cause: tts_voice was missing from both _DEFAULTS and SettingsBody in settings.py, so the voice selection was silently dropped on every save and never loaded on startup

  • tts_voice added to _DEFAULTS (default: en-US-AriaNeural) and SettingsBody in the settings router

  • tts_voice is now forwarded end-to-end: MainConvert and MainResearch each gained a ttsVoice prop; Main.tsx passes the loaded voice to both; the convert form sends it as voice, and the research enqueue JSON sends it as tts_voice; EnqueueRequest and _run_research_job in research.py likewise accept and apply the new field

Debug text files not written in UI mode

  • The debug save block in convert.py now resolves debug_dir via a fresh get_settings() call (not via the mutated cfg2 alias) and explicitly creates the parent directory before the run subdirectory

  • Text values are guarded with or "" to prevent write_text(None) errors

  • Errors are now logged to the server console via logging.warning(..., exc_info=True) in addition to the job log, so failures are visible even without expanding the progress log in the UI

Version 0.3.9 (2026-03-31)

Bug fixes & UX improvements

Browser caching of stale UI after pip upgrade (#1)

  • index.html is now served with Cache-Control: no-cache, no-store, must-revalidate so browsers always revalidate after a package update; hashed JS/CSS bundles generated by Vite remain cacheable as before

Convert tab: button stays visible during conversion (#2)

  • “Convert to audio” button no longer disappears once a job starts — it transitions to a disabled “Converting…” state with a spinner

  • A Cancel button appears beside it for the duration of the job; both revert once conversion completes, errors, or is cancelled

Audio playback broken after renaming a paper (#3)

  • Root cause: GET /api/library/audio was not returning download_url, so sidebar audio entries had undefined as their playback URL

  • download_url (/api/convert/download/{id}) is now included in every audio record returned by the list endpoint; playback works regardless of any title or filename edits

Tab switch clears convert / research progress (#4)

  • All four tab panels (configuration, convert, research, database) are now permanently mounted and toggled with display: none instead of being conditionally rendered; React state — uploaded file, job ID, live progress — is preserved when switching away and back

Long paper titles overflow the convert drop zone (#6)

  • Added break-words whitespace-normal to the filename display inside the drop zone so long titles wrap instead of overflowing in all browsers

  • Same break-words treatment applied to the post-conversion success message

Debug text files not written in UI mode (#7)

  • _save_debug_texts was only called from the synchronous run_pipeline path, which the web UI never uses

  • The enqueue_conversion background task now saves 1_raw.txt, 2_preprocessed.txt, and 3_curated.txt to ~/.audia/debug/<stem>_<timestamp>/ after every successful conversion, matching CLI behaviour

Version 0.3.8 (2026-03-31)

Rebuild UI with the current state of implementation

Version 0.3.7 (2026-03-31)

Fix browser opening before server is ready

  • Replaced the hardcoded time.sleep(1.2) delay in audia serve with a TCP poll loop that opens the browser only once the server is actually accepting connections (checks every 200 ms, 30 s timeout)

  • Eliminates the “unable to reach” error seen on first load when uvicorn had not finished starting within the fixed delay

Version 0.3.6 (2026-03-31)

Google Gemini LLM support

  • AUDIA_LLM_PROVIDER=google — new provider option backed by langchain-google-genai / ChatGoogleGenerativeAI

  • AUDIA_GOOGLE_API_KEY — required when using the Google provider

  • AUDIA_GOOGLE_API_BASE — optional custom endpoint (Vertex AI or corporate proxy); mapped to client_options.api_endpoint

  • llm_provider Literal extended to "openai" | "anthropic" | "google" in config.py; validator error message updated accordingly

  • _build_llm in text_cleaner.py gains a google branch with clear import-error and missing-key messages

  • langchain-google-genai>=2.0 and google-generativeai>=0.8 added to core dependencies in pyproject.toml

  • .env.example updated with an “Option C – Google Gemini” block documenting gemini-2.0-flash, gemini-2.0-flash-lite, and gemini-1.5-pro as example models

Version 0.3.5 (2026-03-31)

Rebuild frontend

  • Rebuild UI with the current state of implementation

    • Add new db tab

    • Update the HTML title

Version 0.3.4 (2026-03-31)

Custom API base URLs for OpenAI and Anthropic

  • AUDIA_OPENAI_API_BASE — optional setting to redirect all OpenAI calls (LLM + TTS) to a custom endpoint (Azure OpenAI, corporate proxy, or any OpenAI-compatible URL)

  • AUDIA_ANTHROPIC_API_BASE — same for Anthropic LLM calls

  • Both settings wired through config.py, text_cleaner.py (_build_llm), and tts.py; stt.py inherits the base URL automatically via _build_llm

  • .env.example updated with commented examples for both options

Version 0.3.3 (2026-03-31)

Icon system, TTS voice selector & database editor

Icon system (constants/index.ts)

  • Introduced IconDef discriminated union type — { kind: 'icon'; name; adaptive? } for Iconify icons or { kind: 'img'; src; alt } for image assets — providing a single typed representation for all logos across the app

  • All asset imports (arxiv.svg, systran.svg, hexgrad.webp) moved from individual component files into constants/index.ts

  • PROVIDER_ICONS updated to Record<LLMProvider, IconDef>; OpenAI uses simple-icons:openai with adaptive: true (renders white in dark mode, black in light mode)

  • New exports: STT_ICON, ARXIV_ICON, TTS_BACKEND_ICONS — centralising all service/backend logos in one place

  • renderIconDef(icon, isDark, className?) helper in MainConfiguration.tsx handles both img and icon variants, applying theme-aware colouring for adaptive icons

TTS voice selector (Configuration tab)

  • TTS card now shows a 2-column grid: Engine selector + Voice selector side-by-side

  • Voice options are populated from TTS_VOICES[backend]; switching the engine automatically resets the voice to the first option for that backend

  • ttsVoice state added to Main.tsx, persisted to and loaded from /api/settings via a new tts_voice key

Database explorer — editable cells

  • All cells in editable columns (title, authors, abstract, arxiv_id, pdf_url for papers; filename, duration_seconds, tts_backend, tts_voice, paper_id for audio files; query for research sessions; value for user settings) are now inline-editable

  • Click a cell → <input> or <textarea> (for abstract/query) appear inline; Enter commits, Escape cancels, ⌘/Ctrl+Enter commits multiline fields; blur also commits

  • tts_backend and tts_voice cells use a portal-based custom dropdown that escapes the overflow-x-auto container; tts_voice options reflect the current row’s backend

  • Brief lime flash on successful save; spinner while saving

  • authors is edited as comma-separated text and sent as a JSON array; paper_id/duration_seconds are coerced to numbers; clearing nullable fields sends null

  • New backend PATCH endpoints for all four tables: PATCH /api/library/papers/{id}, PATCH /api/library/audio/{id}, PATCH /api/library/research_sessions/{id}, PATCH /api/library/user_settings/{key}

Database explorer — table UX

  • Table is now horizontally scrollable — uses table-auto min-w-max so columns never line-wrap; overflow-x-auto wrapper scrolls instead

  • Full column sets returned: list_papers now includes abstract, pdf_path, pdf_url; list_audio now includes file_path, duration_seconds (previously omitted)

  • Column hide/show: each column header has an eye-off icon; clicking hides the column; hidden columns appear as chips above the table that restore on click; hidden set resets when switching tables

  • Table picker replaced native <select> with a fully custom styled dropdown

  • Clicking a papers.id cell (or audio_files.paper_id) opens the PDF in the PreviewPanel — links rendered as rose-coloured buttons instead of editable cells

Version 0.3.2 (2026-03-30)

Convert tab improvements

Instant PDF preview on upload

  • Dropping a PDF or selecting one via the file browser immediately opens the PreviewPanel on the right using a local blob: URL — no conversion needs to start first

  • Clearing the file (“Convert another file”) also closes the preview panel

  • During conversion the panel URL is transparently upgraded to the server-served PDF once the backend has processed it

Collapsible full-progress log

  • The verbose log output (step-by-step stage details, chunk counts, etc.) is now hidden by default behind a “Show full progress” toggle

  • Clicking the toggle expands/collapses the log with a chevron indicator (“Show full progress” ▶ / “Hide full progress” ▼)

  • The log container background contrast raised (bg-white/8 / bg-black/8) so the scrolling area is visually distinct

Version 0.3.1 (2026-03-30)

Frontend refactor & database explorer

Main.tsx split into tab components

  • MainConfiguration.tsx: pipeline diagram and model/backend selectors extracted into a standalone component

  • MainConvert.tsx: PDF upload, ConversionProgress, AudioPlayer, and CONVERT_STAGES extracted; all three are exported for reuse

  • MainResearch.tsx: ArXiv search, voice recording, LLM query normalisation, paper selection, and job progress extracted; exports RESEARCH_STAGES

  • Main.tsx reduced to a 159-line orchestrator: holds shared pipeline config state, loads/saves settings via /api/settings, renders a tab bar, and delegates to each sub-component

Database tab (MainDatabase.tsx)

  • Schema overview: four cards (one per table — papers, audio_files, research_sessions, user_settings) showing all columns with types and PK/FK badges; relationship annotations (FK arrow and JSON logical link) displayed below

  • Mermaid ERD: collapsible erDiagram code block with a one-click copy button

  • Data explorer: dropdown to select any table; fetches all rows from the API and renders them in a full-width table with no truncation; JSON array columns rendered inline

API additions (library router)

  • GET /api/library/research_sessions — returns all research sessions ordered by date

  • GET /api/library/user_settings — returns all key-value settings rows

scripts/explore_db.py

  • Standalone stdlib-only script (no dependencies) that prints a full terminal dump of ~/.audia/audia.db

  • Shows table list, per-table schema (column names, types, nullability, PK), foreign keys, row count, and all cell values with JSON pretty-printing

  • Accepts --db /path/to/other.db to point at a custom database file

Version 0.3.0 (2026-03-29)

Configuration, voice search & LLM query normalisation

Configuration tab

  • Pipeline diagram: animated SVG diagram in the Configuration tab visualises the full pipeline (STT → LLM → ArXiv → PDF → TTS) with correct arrow routing (Text bypasses LLM and enters ArXiv directly)

  • Persistent settings: UserSetting SQLite model stores per-key config; GET /api/settings and PUT /api/settings endpoints load and upsert settings on demand

  • Save button: “Save configuration” button with “Saved ✓” confirmation; config state lifted to Main and loaded from the database on mount

  • Settings applied to pipelines: selected LLM provider/model and TTS backend are forwarded as optional overrides to /api/convert/enqueue and /api/research/enqueue

Voice search (Research tab)

  • Microphone button: browser MediaRecorder captures audio; blob is POSTed to /api/research/transcribe which runs faster-whisper (transcribe_file) and returns the transcript into the query field

LLM query normalisation (Research tab)

  • Normalize query button: calls POST /api/research/normalize which runs distill_search_query from stt.py (same prompt used by the CLI listen command — no duplicate prompt logic)

  • Two-step search flow: raw query → optional LLM normalisation → editable confirmation pane → “Search arXiv” button; normalisation errors surfaced as a dismissable red banner instead of silent fallback

  • Single Search arXiv button: removed duplicate search button; one full-width button below the input row handles both direct and post-normalisation searches; “No results found” only shown after an actual search attempt

API additions

  • POST /api/research/normalize — wraps distill_search_query

  • POST /api/research/transcribe — wraps transcribe_file

  • POST /api/convert/upload — synchronous PDF upload + convert (used by tests)

  • POST /api/research/convert — synchronous ArXiv download + convert (used by tests)

Code hygiene

  • Removed _QUERY_SYSTEM prompt and normalize_query() from text_cleaner.py; normalisation now reuses distill_search_query directly — two LLMs, two prompts, no duplication

Version 0.2.0 (2026-03-29)

Web UI overhaul

  • Async conversion jobs: PDF upload and ArXiv research conversions now run in the background; a job_id is returned immediately and the frontend polls for live progress

  • Streaming terminal log: each pipeline stage (PDF extraction, heuristic cleaning, LLM curation chunk-by-chunk, TTS chunk-by-chunk) streams log lines into a scrollable terminal pane

  • Cancel button: any running conversion can be cancelled mid-pipeline via a cancel button that calls DELETE /api/{convert,research}/jobs/{id}

  • Inline PDF preview: clicking a paper in the sidebar now opens the PDF in a side panel instead of downloading it; Content-Disposition: inline is set on all PDF-serving endpoints

  • Live PDF preview during conversion: the preview panel opens automatically as soon as the PDF is available (right after upload or ArXiv download), before the pipeline finishes

  • Research async pipeline: POST /api/research/enqueue replaces the old blocking /convert; each ArXiv ID gets its own job with 6 stages (searching → downloading → extracting → pre-cleaning → LLM curation → TTS synthesis)

  • Progress callbacks: llm_curate and synthesize/_edge_tts accept a progress_cb parameter used by the web job runner to emit per-chunk log lines

  • Shared job store: audia.ui.jobs.JOBS dict is imported by both convert.py and research.py routers so cancel and PDF-serve endpoints work across both flows

  • UI fixes: “Convert another file” reset button only appears after conversion completes; PreviewPanel refactored to accept title/pdfUrl directly instead of a Paper object

Version 0.1.2 (2026-03-29)

ArXiv search robustness & CLI improvements

PDF download

  • SDK-free download: PDFs are now fetched directly from https://arxiv.org/pdf/<id> via urllib, bypassing the export API entirely and eliminating 429 rate-limit failures on download

  • Manual fallback prompt: if a download still fails, the user is shown the arxiv.org/abs/ link and prompted to provide a local PDF path to continue conversion

CLI output

  • Results table: added a Link column (https://arxiv.org/abs/<id>) with no_wrap=True — URLs are never truncated so they remain clickable

  • ASCII banner: running audia with no arguments now displays a block-art banner before the help text

  • Paper selection always prompted: audia listen and audia research both show the results table and ask the user to pick papers; --convert flag still skips the prompt for power users

Version 0.1.1 (2026-03-29)

listen command pipeline overhaul

  • LLM query distillation: raw transcribed speech is now passed through the configured LLM to extract a concise ArXiv search query (e.g. “I would like to research agentic AI” → “agentic AI research”)

  • distill_search_query() moved to audia.agents.stt (proper agent layer, not CLI)

  • Confirmation loop before searching: after distillation the user sees the extracted query and can confirm (y), re-record (r), or quit (q) — prevents accidental searches from mis-transcriptions

  • typer.confirm() replaced with explicit typer.prompt() to support the three-way choice

Version 0.1.0 (2026-03-29)

First fully working release: PDF to audio via mandatory LLM curation and edge-tts synthesis.

  • Linear LangGraph pipeline: PDF extraction → heuristic clean → LLM curation → TTS synthesis

  • LLM curation is mandatory; supports OpenAI and Anthropic backends via AUDIA_LLM_PROVIDER

  • Chunk-level LLM curation with tail-context stitching for smooth spoken transitions

  • edge-tts default TTS backend: async with 90 s per-chunk timeout and asyncio context fix for FastAPI

  • Rich per-step progress output; audia --version flag

  • Consistent run ID (<pdf_stem>_<YYYYMMDD_HHMMSS>) shared by audio filename and debug folder

  • Debug text snapshots saved to ~/.audia/debug/<run_id>/ (raw, preprocessed, curated)

  • SQLite storage for papers and audio files via SQLAlchemy

  • FastAPI web UI (SPA) and Typer CLI with convert, research, listen, serve, info commands

  • .env.example with full documentation of all AUDIA_* settings

Version 0.0.1 (2026-03-28)

First release of the basic package structure