# AiPod — podcast data for AI agents > Like Exa, but for podcasts. Episodes are diarized at ingest into attributed > claims, dated predictions, and verifiable facts — each with a timestamp and a > verbatim transcript quote — then compiled into a cross-episode entity graph. > Queries hit a deterministic resolver: sub-100ms, $0/query, no model at query > time, zero hallucinated citations. You (the agent) synthesize; AiPod grounds. ## Coverage AiPod fully indexes the All-In Podcast, plus recent-episode coverage of Dwarkesh Podcast, Lenny's Podcast, The Diary Of A CEO, The Tucker Carlson Show, Huberman Lab, Invest Like the Best, The Ezra Klein Show, Toplinjen. Corpus right now: 83 episodes, 1565 entity pages, 2439 claims, 393 dated predictions, 1533 facts. Latest episode: 2026-09-04. Knowledge cutoff per show (episodes published after these dates are unknown to AiPod until ingested): - huberman-lab: 2024-04-10 → 2026-07-23 (15 episodes) - all-in: 2026-05-01 → 2026-09-04 (30 episodes) - invest-like-the-best: 2026-05-13 → 2026-07-21 (12 episodes) - the-diary-of-a-ceo: 2026-06-04 → 2026-07-30 (20 episodes) - toplinjen: 2026-06-11 → 2026-06-11 (1 episodes) - the-ezra-klein-show: 2026-07-14 → 2026-07-14 (1 episodes) - the-tucker-carlson-show: 2026-07-23 → 2026-07-23 (1 episodes) - dwarkesh-podcast: 2026-08-11 → 2026-08-25 (2 episodes) - lenny-s-podcast: 2026-08-23 → 2026-08-23 (1 episodes) ## How to use the results Synthesize ONLY from returned material. Cite as [episode title, timestamp, holder]. Every claim object carries: holder, claim, stance, conviction (0-1), subjects[], timestamp, quote (verbatim), episode_id, episode_title, podcast, published_at. The context response carries "covered": false when the corpus does not cover the question — and an empty result set means the same. Say so; do not invent. ## Interface 1: MCP (recommended for agent runtimes) Remote server, no install, no API key — Streamable HTTP at https://aipod-two.vercel.app/mcp: claude mcp add --transport http aipod https://aipod-two.vercel.app/mcp { "mcpServers": { "aipod": { "type": "http", "url": "https://aipod-two.vercel.app/mcp" } } } 5 tools, context-first: aipod_context, aipod_search, aipod_entity, aipod_episodes, aipod_episode. Start (and usually finish) with aipod_context — its resolver routes the question, assembles enriched context, and abstains explicitly on corpus gaps. ## Interface 2: HTTP JSON API Base URL: https://aipod-two.vercel.app - GET /api/context?q={question} — routed + enriched grounded context (start here) - GET /api/search?q={query}&limit=8 — keyword search over claims/topics/facts - GET /api/entity/{name} — compiled cross-episode entity page - GET /api/entities — entity index - GET /api/episodes — indexed episode list - GET /api/episode/{id} — full structured episode record - GET /api/status — corpus stats - GET /api — this endpoint map, as JSON POST /api/agent {"question": "..."} is the exception to "you synthesize": an LLM agent drives the same brain tools and returns {answer, tool_calls, sources, abstained} — one complete cited answer, with the tool-call audit trail. Slower and rate-limited; the deterministic endpoints above are the primary surface. (MCP exposes the deterministic tools only — context-first; the agent's granular tools are internal to /api/agent.) ## Example GET /api/context?q=What+does+Chamath+think+about+AI+capex → { "route": "person-stance", "covered": true, "claims": [ { "holder": "Chamath Palihapitiya", "claim": "...", "quote": "", "timestamp": "8:29", "episode_title": "...", "published_at": "2026-05-01" }, ... ], "predictions": [...], "notable_quotes": [...] } ## Human-browsable content - / — service landing page - /takeaways/ — the biggest takeaways from every episode: a numbered list per episode (thesis, grounding, verbatim receipt + timestamp), written at ingest from the record and grounding-gated. Per episode: /takeaways/.txt (post text) and .json; also rendered under each episode page (#takeaways) - /digest/ — The AiPod Weekly: one editorial issue per corpus week, written at ingest from the episode records; every quote cited and grounding-gated. Per issue: /digest/ (+ .json and a markdown twin via Accept: text/markdown) - /demo/ — demo gallery: three small products rendered from the graph (prediction ledger, entity dossier, weekly briefing) - /allin/ — All-In Podcast index (per-episode structured pages) - /dwarkesh-podcast/ — Dwarkesh Podcast index (per-episode structured pages; recent-episode coverage 2026-08-11 → 2026-08-25) - /lenny-s-podcast/ — Lenny's Podcast index (per-episode structured pages; recent-episode coverage 2026-08-23 → 2026-08-23) - /the-diary-of-a-ceo/ — The Diary Of A CEO index (per-episode structured pages; recent-episode coverage 2026-06-04 → 2026-07-30) - /the-tucker-carlson-show/ — The Tucker Carlson Show index (per-episode structured pages; recent-episode coverage 2026-07-23 → 2026-07-23) - /huberman-lab/ — Huberman Lab index (per-episode structured pages; recent-episode coverage 2024-04-10 → 2026-07-23) - /invest-like-the-best/ — Invest Like the Best index (per-episode structured pages; recent-episode coverage 2026-05-13 → 2026-07-21) - /the-ezra-klein-show/ — The Ezra Klein Show index (per-episode structured pages; recent-episode coverage 2026-07-14 → 2026-07-14) - /toplinjen/ — Toplinjen index (per-episode structured pages; recent-episode coverage 2026-06-11 → 2026-06-11) ## Upcoming: deep answer (waitlist) Today you synthesize from the returned context. Deep Answer (in the lab) will accept one question and return one complete, fully cited answer — an AiPod agent does the retrieval, processing, and inference. Register interest: POST /api/waitlist with body {"email": "you@example.com"} (agents welcome — use your operator's email). ## Notice Unofficial — AiPod is not affiliated with, endorsed, or sponsored by the All-In Podcast or any show in the index. Quotes are short excerpts with timestamps, attributed to their speakers, intended to send listeners to the original episodes. AiPod is at a very early stage and building in public — coverage, interfaces, and data shapes will change.