Always remember everything you ever heard.
You've listened to thousands of hours of podcasts. Quick — what did the last one actually say?
podmind is the memory you wish your ears had. Point it at your Pocket Casts and YouTube history and it quietly builds a Karpathy-style LLM knowledge wiki of everything you've actually heard — one cross-linked markdown page per episode, plus people, topics, and show pages, with a listening-state badge on every citation so you can tell what you truly listened to from what's just been sitting in your queue.
Listening history goes in; an Obsidian-compatible second brain comes out. And it compounds: every ingest and every query leaves more cross-links behind.
- You already did the listening — keep the knowledge. Hours of audio leave almost no durable trace. podmind turns listening you've already done into something searchable.
- Answers from your sources, not the web.
podmind query "..."synthesizes a cited answer from the episodes you actually heard — not a generic chatbot guess. - Plain markdown you own forever. Obsidian-compatible files on your disk. Greppable, portable, and they outlive every SaaS knowledge tool.
- It builds itself. A daily job pulls new listens, transcribes, summarizes, and cross-links — hands-off while you sleep.
- Cheap and local-first. ~$0.015/episode for summaries; transcription runs free on Apple Silicon.
Run podmind over an episode and you get a page like this — structured takeaways, quotes with timestamps, and real cross-links, topped with a listening-state badge (🎧 heard · ▶ 38% in progress · ⚪ unheard):
---
show: "Signals & Noise"
date: 2026-05-16
guests: ["Daniel Voss"]
---
▶ 38% [[shows/signals-and-noise]] — Daniel Voss on cost-per-token engineering:
your LLM bill is input tokens, and the cheapest model that emits valid JSON wins.
## Key takeaways
- Summarization cost is dominated by input tokens (~90%), so the bill scales
with transcript length, not model intelligence.
- Three levers — truncation, prompt caching, and model routing — cut costs by
two thirds before any prompt tuning.
## Notable quotes
> Your bill is determined by how much you make the model read, not by how smart
> it is. — Daniel Voss (00:04:10)
## Cross-links
- People: [[people/maya-brennan]], [[people/daniel-voss]]
- Topics: [[topics/llm-cost-engineering]], [[topics/summarization-pipelines]]Every [[link]] resolves to a real page. Topic and people pages gather every episode that touched them — so "what have I actually heard about LLM costs?" is a grep away: filter a topic page's citations for 🎧 and you get exactly the claims you've personally listened to, not just subscribed to.
One real vault, built passively over ~10 years of listening:
| 14,789 episodes tracked | 8,789 hours of audio |
| 1,720 shows & channels | 9,376 episodes distilled into the wiki |
| 19,467 people mapped | 30,589 topics — every one cross-linked |
The busiest threads run from Bitcoin (562 citations) to AI agents (260) to geopolitics — the wiki maps whatever you actually feed it.
Three layers, strictly separated:
- Raw layer (immutable) — episode metadata + transcripts at
$PODMIND_DATA_ROOT/raw/episodes/<show>/<date-title>/. Source of truth, never edited. - Wiki layer (LLM-curated) — markdown pages at
$PODMIND_DATA_ROOT/wiki/. Regeneratable from raw + the schema. - Schema layer — docs/AGENTS-vault.md, the agent instructions that define what a well-formed wiki looks like. It co-evolves with the wiki as new patterns emerge.
The Python package (this repo) is the pipeline. Your data and wiki live in a separate vault directory.
flowchart LR
PC[Pocket Casts<br/>listening history] --> RAW
YT[YouTube<br/>watch history] --> RAW
RAW[(raw/ — immutable<br/>meta + transcripts)] --> CASCADE{{transcript cascade<br/>rss → publisher → podcast-index<br/>→ youtube → whisper}}
CASCADE --> RAW
RAW --> SUM[LLM summarizer<br/>any OpenAI-compat or Anthropic API]
SUM --> WIKI[(wiki/ — episodes, people,<br/>topics, shows, synthesis)]
WIKI --> OBS[Obsidian / any<br/>markdown editor]
WIKI --> STATS[stats.md + charts]
git clone https://github.com/DexMind-AI/podmind.git
cd podmind
uv sync
./scripts/demo.shThen open examples/demo-vault/ in Obsidian (or browse wiki/ in any markdown editor). The vault ships a graph-view config that scopes the graph to the wiki/ layer, so you see the cross-linked output, not the raw transcripts.
The demo runs the real pipeline — tick_prep.py scans the vault for pending episodes, tick_finalize.py writes episode pages, people/topic stubs, the show page, the index, and a log entry — over 3 bundled synthetic episodes. The LLM summarization step is replaced by pre-baked summary JSON, so it needs zero API keys.
uv sync # core pipeline
uv sync --extra whisper # + local mlx-whisper transcription (Apple Silicon)
uv sync --extra stats # + matplotlib/pandas for the analytics charts
uv sync --extra all # everything, incl. yt-dlpThe vault is a plain directory with raw/ and wiki/ subtrees; podmind finds it via PODMIND_DATA_ROOT. Export it in your shell profile, or use direnv (dotenv .env in .envrc) — podmind does not auto-load .env. See .env.example.
export PODMIND_DATA_ROOT=/Users/you/my-podmind-vault
mkdir -p "$PODMIND_DATA_ROOT"/{raw/{feeds,episodes,audio},wiki/{episodes,people,topics,shows,synthesis}}Credentials live in ~/.config/podmind/secrets.json (override the path with PODMIND_SECRETS):
mkdir -p ~/.config/podmind
cat > ~/.config/podmind/secrets.json <<'EOF'
{
"llm_api_key": "sk-..."
}
EOF
chmod 600 ~/.config/podmind/secrets.json| Key | Required | Used for |
|---|---|---|
llm_api_key |
yes (or legacy deepseek_api_key) |
the summarizer (bin/summarize.py / bin/ingest_run.py) |
pocketcasts_token |
for Pocket Casts sync | written for you by uv run python -m podmind.pocketcasts login |
podcast_index_key / podcast_index_secret |
optional | the podcast-index tier of the transcript cascade |
embed_api_key |
optional | embeddings for semantic search (bin/embed_all.py, bin/query.py) |
Default is DeepSeek — put one key in secrets.json and you're done. podmind also speaks OpenAI, OpenRouter, Ollama, and Anthropic; see docs/OPERATIONS.md for the provider matrix and embeddings config.
# One-time: log in to Pocket Casts (stores a token in secrets.json)
uv run python -m podmind.pocketcasts login
# Pull listening state + run the transcript cascade on a small batch first
uv run podmind sync --no-whisper --limit 10
# Summarize pending episodes into the wiki
uv run podmind ingest 12The transcript cascade tries five tiers in order, cheapest first, and stops at the first hit:
- rss — Podcasting 2.0
<podcast:transcript>tag in the show's feed - publisher — known-publisher web scrapers (often paywalled)
- podcast-index — transcript URLs registered at api.podcastindex.org
- youtube — yt-dlp subtitles (direct URL for YT-sourced episodes;
ytsearch1for podcasts) - whisper — local
mlx-whisperas last resort (Apple Silicon,--extra whisper)
PODMIND_DATA_ROOT=/Users/you/my-podmind-vault ./scripts/install_launchd.shLoads a daily 04:00 pipeline + an AC-gated whisper loop. Details in docs/OPERATIONS.md.
Pocket Casts has no public API. podmind talks to the same private endpoints the Pocket Casts web player uses — like other open-source Pocket Casts clients. It can break without notice, and you should consider whether that's acceptable use of your account.
~$0.015/episode on the default DeepSeek path (measured); a 100-episode run ≈ $1.50. Whisper transcription is free locally on Apple Silicon. Full breakdown in docs/OPERATIONS.md.
One CLI, podmind <command> (run podmind <command> --help for flags):
| Command | What it does |
|---|---|
podmind ingest [N] |
summarize pending episodes into the wiki |
podmind query <q> |
semantic query, filed back as a synthesis page |
podmind lint |
wiki health pass (near-dup topics, broken links, stale badges) |
podmind sync |
pull Pocket Casts state + run the transcript cascade |
podmind transcript |
transcript cascade only |
podmind digest |
a short digest of recent listening |
podmind stats |
regenerate wiki/stats.md + charts |
podmind embed |
build/refresh the semantic-search cache |
podmind compress-transcripts |
migrate VTT transcripts to/from xz |
podmind refresh-badges |
re-derive listened-state badges |
podmind demo |
the zero-key demo |
Operator detail — launchd automation, the provider matrix, cost deep-dive — is in docs/OPERATIONS.md.
docs/AGENTS-vault.md is the third layer of the system: the agent instructions a coding agent (Claude Code or similar) follows to maintain the wiki — page formats, badge rules, the ingest/query/lint operation contracts, and log-entry types. In Karpathy's original pattern the schema is not frozen: when an operation reveals a missing or ambiguous instruction, the agent updates the schema in the same session and logs the change. The schema is the highest-leverage file in the vault — a one-sentence fix there propagates to every future session.
MIT