What is MinusPod?

A self-hosted podcast processing server that eliminates advertisements before episodes reach listeners. Episodes get transcribed using faster-whisper or NVIDIA Parakeet models locally, then an AI model identifies ad segments for removal. The system learns from corrections to recognize repeat sponsors efficiently. Users can integrate Anthropic Claude, Ollama, OpenRouter, System One, or any compatible OpenAI endpoint. Modified feeds output as re-cut RSS with versioned audio files, replacement sounds where cuts occur, loudness normalization, and lossless MP3 streaming when available.

Should you self-host MinusPod?

MinusPod scores 50/100 on selfhost.directory (how scoring works) and ranks #7 of 14 in Audio Streaming. It has 508 GitHub stars.

It shipped 11 stable releases in the past year. The latest, v2.100.24, came out on 27 October 2026. New versions land roughly every 1 days.

Setup is moderate: expect some configuration beyond a single command. An official Docker image is available. It is MIT-licensed and written mostly in Python.

Its closest alternative on selfhost.directory is Snapcast (72/100, 8 038 stars). It scores higher overall, so compare the two before you pick. See MinusPod vs Snapcast side by side.

How to configure MinusPod

ANTHROPIC_API_KEY
ARTWORK_BADGE_POSITIONBadge corner: bottom-right (default), bottom-left, top-right, top-left
Badge corner: bottom-right (default), bottom-left, top-right, top-left
ARTWORK_WATERMARK_ENABLED
AUTO_PROCESS_ENABLEDDeploy posture (env-backed: seed the initial state for fresh deploys)
Deploy posture (env-backed: seed the initial state for fresh deploys)
BASE_URLhttp://localhost:8000Server Configuration (BASE_URL required before first boot)
Server Configuration (BASE_URL required before first boot)
CHAPTER_BOUNDARY_MAX_TOKENS
CHAPTER_BOUNDARY_REASONING_BUDGET
CHAPTER_BOUNDARY_REASONING_LEVEL
CHAPTER_BOUNDARY_TEMPERATUREChapter Boundary Detection
Chapter Boundary Detection
CHAPTER_TITLE_MAX_TOKENS
CHAPTER_TITLE_REASONING_BUDGET
CHAPTER_TITLE_REASONING_LEVEL
CHAPTER_TITLE_TEMPERATUREChapter Title Generation
Chapter Title Generation
DETECTION_MAX_TOKENS
DETECTION_REASONING_LEVELDETECTION_REASONING_BUDGET= # Anthropic only, 1024-65536
DETECTION_REASONING_BUDGET= # Anthropic only, 1024-65536
DETECTION_TEMPERATUREAd Detection (Pass 1)
Ad Detection (Pass 1)
FEED_AUTH_ENABLED
LLM_PROVIDERanthropicOption 1: Anthropic (default), direct Claude API access
Option 1: Anthropic (default), direct Claude API access
MAX_AUDIO_DOWNLOAD_MBSize caps (env-backed: these seed the defaults; values saved in Settings win)
Size caps (env-backed: these seed the defaults; values saved in Settings win)
MINUSPOD_ALLOW_PUBLIC_PROCESSINGKeep public podcast clients from starting paid processing. A valid global or feed-scoped subscriber credential can still start processing.
Keep public podcast clients from starting paid processing. A valid global or feed-scoped subscriber credential can still start processing.
MINUSPOD_BIND_ADDRESSBind only to a private interface when an ingress proxy is the sole caller.
Bind only to a private interface when an ingress proxy is the sole caller.
MINUSPOD_MASTER_PASSPHRASEStrongly recommended before first boot: unlocks encrypted storage for provider API keys. Without it, keys are stored as plaintext.
Strongly recommended before first boot: unlocks encrypted storage for provider API keys. Without it, keys are stored as plaintext.
MINUSPOD_MAX_ARTWORK_BYTES
MINUSPOD_MAX_RSS_BYTES
MINUSPOD_PORTInternal listen port. Default 8000. Useful for host networking or running multiple instances on one host. Ignored when GUNICORN_BIND is set.
Internal listen port. Default 8000. Useful for host networking or running multiple instances on one host. Ignored when GUNICORN_BIND is set.
MINUSPOD_REQUIRE_AUTHCompose refuses protected API access until an application password is set. Remote first-password setup requires this one-time header value; remove it after setup. Loopback setup does not require a token.
Compose refuses protected API access until an application password is set. Remote first-password setup requires this one-time header value; remove it after setup. Loopback setup does not require a token.
MINUSPOD_SETUP_TOKEN
MINUSPOD_TRUSTED_PROXY_COUNTTrusted-proxy hop count. 0 = no proxy (default). Set to 1 when running behind Cloudflare / cloudflared / nginx / Traefik so the login lockout keys on the real client IP from X-Forwarded-For instead of the proxy loopback. Required when enabling the tunnel profile in docker-compose.yml.
Trusted-proxy hop count. 0 = no proxy (default). Set to 1 when running behind Cloudflare / cloudflared / nginx / Traefik so the login lockout keys on the real client IP from X-Forwarded-For instead of the proxy loopback. Required when enabling the tunnel profile in docker-compose.yml.
OLLAMA_NUM_CTXOllama context window (Ollama provider only) Default Ollama context is small (often 2048) and silently truncates long prompts. Set this to the model's trained context limit (8192 or higher for most modern models).
Ollama context window (Ollama provider only) Default Ollama context is small (often 2048) and silently truncates long prompts. Set this to the model's trained context limit (8192 or higher for most modern models).
OMP_NUM_THREADSOptional: local CPU transcription tuning on hybrid Intel CPUs (12th gen and newer). Cap OpenMP threads to your performance-core count so work stays off the E-cores. See docs/installation.md for P-core pinning. No effect with a remote Whisper API.
Optional: local CPU transcription tuning on hybrid Intel CPUs (12th gen and newer). Cap OpenMP threads to your performance-core count so work stays off the E-cores. See docs/installation.md for P-core pinning. No effect with a remote Whisper API.
OPENAI_API_KEY
OPENAI_BASE_URL
OPENAI_MODEL
OPENROUTER_API_KEY
PODCAST_INDEX_API_KEYPodcast Discovery (optional) PodcastIndex.org API credentials for podcast search Get free API credentials at https://api.podcastindex.org
Podcast Discovery (optional) PodcastIndex.org API credentials for podcast search Get free API credentials at https://api.podcastindex.org
PODCAST_INDEX_API_SECRET
RATE_LIMIT_STORAGE_URIExact cross-worker rate limits require Redis and a reachable Redis URI.
Exact cross-worker rate limits require Redis and a reachable Redis URI.
RETENTION_PERIODRetention period in minutes (legacy, prefer Settings UI)
Retention period in minutes (legacy, prefer Settings UI)
REVIEWER_MAX_TOKENS
REVIEWER_REASONING_BUDGET
REVIEWER_REASONING_LEVEL
REVIEWER_TEMPERATUREReviewer (applies to both reviewer pass 1 and reviewer pass 2)
Reviewer (applies to both reviewer pass 1 and reviewer pass 2)
SESSION_COOKIE_SECUREFollow BASE_URL for the Secure cookie flag. Explicit true or false overrides.
Follow BASE_URL for the Secure cookie flag. Explicit true or false overrides.
SYSTEMONE_API_KEY
SYSTEMONE_BASE_URL
TUNNEL_TOKENCloudflare tunnel (optional, for remote access)
Cloudflare tunnel (optional, for remote access)
TYPESAFE_API_KEY
VERIFICATION_MAX_TOKENS
VERIFICATION_REASONING_BUDGET
VERIFICATION_REASONING_LEVEL
VERIFICATION_TEMPERATUREVerification (Ad Detection Pass 2)
Verification (Ad Detection Pass 2)
WHISPER_API_BASE_URL
WHISPER_API_KEY
WHISPER_API_MODEL
WHISPER_BACKENDSelect local, parakeet, or openai-api; Parakeet has one fixed model. Remote API example (Groq, OpenAI, whisper.cpp):
Select local, parakeet, or openai-api; Parakeet has one fixed model. Remote API example (Groq, OpenAI, whisper.cpp):
WHISPER_COMPUTE_TYPE
WHISPER_DEVICE
WHISPER_LANGUAGE
WHISPER_MODELLocal defaults come from Compose: GPU uses small/cuda; CPU uses tiny/cpu. WHISPER_MODEL and WHISPER_COMPUTE_TYPE apply to faster-whisper only.
Local defaults come from Compose: GPU uses small/cuda; CPU uses tiny/cpu. WHISPER_MODEL and WHISPER_COMPUTE_TYPE apply to faster-whisper only.