What's up everyone. Jim here kicking off The Pull Request, a series for Jamie product updates. Two things shipped this week that I've been wanting to build for a long time.
TL;DR
What's Smart Mode? Jamie now parses vague natural language queries into structured filters by guest, topic, feed, and time frame before running semantic search. "What was Saylor saying about AI on TFTC last year" just works.
What's the L402 stuff? Any AI agent with a Lightning wallet can now do deep research across thousands of hours of expert conversation. Prospecting, competitive intel, tracking how a topic evolves over time, pulling the exact quote that proves the point. No human in the loop. The Bitcoin/AI centric internet aka the Machine Payable web is coming to life.
From Quote Lookup to Actual Understanding
Jamie started as a direct semantic lookup engine. Type a query, match against transcript quotes in vector space, get results. I call that Speed Mode. Still there. Still great when you know exactly what you're after.
But people don't always search like that. They search like this:
"What was Saylor saying about AI on WBD last year" (guest + topic + feed + time frame)
"That All-In episode about tariffs from early 2025" (feed + topic + time frame)
"Joe Rogan's interview with that Facebook whistleblower" (host + vague guest reference)
Vague. Full of implied context. Speed Mode would slam this raw text into a vector search and hope for the best. Usually the results were mid.
Smart Mode fixes this. Toggle it on and Jamie routes your natural language through an LLM before the search happens. It extracts the guest, the topic, the time frame, the feed. Maps all of that to meaningful filters and rephrases the query so the semantic lookup actually hits.
"Rogan talking to that CIA guy about China last time he went on" becomes a filtered search scoped to the JRE feed, filtered to his last appearance, focused on CIA officer Mike Baker segments about artificial intelligence.
Not a raw vector match against those exact words. Speed Mode is a scalpel. Smart Mode is a conversation. Both return the same thing: timestamped clips, deeplinks, playable audio. Ground truth you can press play on.
I Built for Machines. Turns Out They Wanted Smart Mode Too.
I originally built direct semantic search anticipating machines as the primary consumers. Agents don't ramble. They can construct precise queries. Speed Mode was designed for them.
But even agents appreciate not having to learn your corpus. An agent researching a topic doesn't know which feeds cover it, which guests are relevant, or how the data is structured. Smart Mode outsources that entire learning curve to Jamie's routing logic. Describe what you want. Jamie figures out where to look.
That made something click. If agents are going to use this for real, they need to discover, index, and explore the full corpus. And pay for it without a human setting up an account. Just give your agent an NWC wallet or preimage and go.
So we went L402.
The Power of Lightning L402: Four Tools. One Wallet. No Paperwork.
Jamie is now on https://402index.io. Need to know everything a prospect has said publicly before a meeting? Need to track how Bitcoin sentiment evolved over 18 months? Need to ground a research brief in real quotes instead of hallucinated ones? That's what agents pay Jamie for. Fractions of a cent per call. One Lightning credential.
Discover Start with a fuzzy goal like "find podcasts the OpenClaw guy was on" and get back matching episodes across 4M+ podcasts. Results flag which episodes already have transcripts ready for deeper search and which ones need indexing.
Search Transcripts The core. Semantic search across millions of indexed transcript segments. Ask a question, get timestamped quotes with deeplinks to exact audio. Supports the smartMode flag for vague queries.
Search Chapters The best way to survey the landscape before going deep. Instead of searching raw quotes, this scans 70K+ LLM-curated chapter labels across the entire corpus. Find out who's talking about a topic, how the conversation has shifted over time, and where to focus before committing to a full transcript dive.
Transcribe & Index Found an episode through discovery but no transcript yet? Submit it for full transcription, chaptering, and permanent indexing. Once done, that content is searchable forever via the other endpoints.
Instead of building their own podcast research pipeline from scratch, agents get a ready-made one for fractions of a cent per call. No corpus to maintain, no embeddings to manage, no transcription to run. Show up with a Lightning wallet and a question, get structured answers back, follow the thread wherever it leads.
As Long as the Bitcoin Ecosystem is Bloomin' I'm Not Doomin'
I wrote recently about how Bitcoin needs applications that agents actually have to pay for. This is that.
Smart Mode makes Jamie useful to humans who think in natural language. L402 makes Jamie useful to agents that pay per action. 402index is the directory where agents find us. The whole loop is live and the network of agent-payable services is getting bigger every week.
If you're building agents that need grounding in real long form conversation from experts, point them at our spec.
If you're a human who wants to search smarter, toggle Smart Mode at pullthatupjamie.ai/app.
More Pull Requests coming soon.
Keep building with me. Never doom. We're gonna win.
Frequently Asked Questions
What is PullThatUpJamie Smart Mode? Smart Mode is an LLM-powered query router that parses vague natural language searches into structured filters. It extracts guests, topics, feeds, and time frames from your query and applies them as filters before running semantic search. The result is dramatically better accuracy for casual, conversational queries.
How does PullThatUpJamie discover podcasts? The /discover-podcasts endpoint searches across 4M+ podcasts to find people, topics, and shows using natural language. Ask something like "podcasts the OpenClaw guy was on" or "bitcoin mining discussions in 2025" and get back matching episodes with metadata and flags indicating which ones are ready for deeper transcript search.
What can I search with PullThatUpJamie's transcript search? The /search-quotes endpoint indexes hundreds of high signal podcast feeds covering 10k+ episodes and millions of transcript segments. Shows include Joe Rogan Experience, Lex Fridman, Huberman Lab, TFTC, All-In, What Bitcoin Did, Acquired, and many more. Search returns timestamped quotes with deeplinks to exact audio. New episodes are indexed continuously, and agents can submit unindexed episodes for transcription on demand via /on-demand/submitOnDemandRun.
What is PullThatUpJamie chapter search? The /search-chapters endpoint searches 70K+ LLM-curated chapter labels across the corpus. It returns chapter headlines, summaries, timestamps, and episode context. Best used as a survey tool to find out who's discussing a topic and how the conversation has evolved before drilling into specific transcript quotes.
How do AI agents use Jamie with Lightning? Jamie's four API endpoints are L402-enabled. An agent hits an endpoint, receives a 402 response with a Lightning invoice, pays it, and retries with the preimage as authorization. No API keys or accounts needed. One credential works across all endpoints. All four are listed and verified on 402index.io.
How much does Jamie's API cost? Discovery and search endpoints run fractions of a cent per call. Full episode transcription and indexing via /on-demand/submitOnDemandRun costs $0.45 per episode. All billing is per-call from a single Lightning credential with no subscriptions or minimums.
How does an AI agent pay for Jamie with Bitcoin Lightning? Jamie's endpoints use L402, a protocol where the server responds with a Lightning invoice instead of requiring login credentials. The agent pays the invoice and uses the proof of payment as its access token. The easiest way to get started is with lnget, Lightning Labs' CLI that handles L402 payments automatically. Just point it at a Jamie endpoint and it takes care of the rest. For agents that need a full Lightning wallet, Alby Hub with Nostr Wallet Connect gives your agent a sandboxed wallet with budget controls. Alby also offers an MCP server that plugs directly into Claude, Cursor, and other agent frameworks. For the full agent commerce toolkit including node setup, remote signing, and scoped credentials, see Lightning Labs' lightning-agent-tools.
