Name Every Episode File So an AI Agent Can Find the Right One
Name each transcript and show notes file by one rule, upload them to Brain, and build an AI agent in Gumloop that cites the file it found in your archive.
Write one naming rule, such as episode number and guest, and name every transcript and show notes file by it before you upload them to a Brain source in Gumloop. The block then has the AI agent name the files and quote the passages it found, or say Nothing in the archive matches.
When you ask the archive where a guest made a point, the file name is the part of the answer you can check.
Brain is where Gumloop keeps searchable knowledge from your files, and it works from text. Your recordings stay where they are, and the transcripts and show notes you've already got become the archive.
What does the finder hand back, and when does it hand back nothing?
When a snippet comes back with its file name, it gives that name exactly as the file is named and quotes the passage under it. When the search returns nothing that answers your question, it writes "Nothing in the archive matches." and stops.
The block has it search the attached source before it answers any question about an episode, and bars an answer from memory. If your documents live in a folder, What Do I Put in an AI Agent's Knowledge Source? covers what goes in it.
What do you paste into the instructions for the back catalog finder?
You're the back catalog finder for [SHOW NAME]. Everything you may use about my episodes is in the Brain source attached to you. Each file in it is named by this rule: [YOUR NAMING RULE]. When I ask about an episode, search that source before you answer. If a snippet isn't enough, read the full document. Answer by naming every file you found, exactly as it is named, and quote the passage from it that answers my question. List at most [NUMBER OF FILES] files, most relevant first. If the search returns nothing that answers my question, reply with only this sentence: "Nothing in the archive matches." Never answer a question about an episode from memory, from the web or from an earlier task. If a snippet comes back with no file name, write "file name missing" beside it. Don't summarize an episode I didn't ask about. Don't post, send or change anything. I do that myself.
How do you build the back catalog finder in Gumloop?
- 01
Write the naming rule before anything is uploaded.
Put the rule on one line with blanks, such as [EPISODE NUMBER] [GUEST NAME] [TRANSCRIPT OR NOTES]. The name is what you check an answer against. That line goes into the block. Save each transcript and set of show notes as plain text or Markdown. Gumloop lists PDFs, Office files, rich text, Markdown, JSON, XML, YAML and plain text for a File uploads source, up to 25 MB each, and rejects unsupported types. It lists no audio or video and doesn't read text inside images. Checked on 2026-11-10 against Gumloop's Brain documentation.
- 02
Rename the files by the rule, then add them as one source.
In Gumloop, open Brain from the left sidebar, click + Source and pick File uploads. Brain is available on the Pro and Enterprise plans, so check which plan you're on first. Name the source clearly, as Brain's page advises, then add your renamed files. The file upload reference, written for the API, says the name a file is uploaded with becomes its name in the source, and the sync page says direct file uploads change only when you add or remove files. A misnamed file means removing it and adding it again. Wait until the source reads Active. Checked on 2026-11-10 against Gumloop's Brain documentation, the Brain sync help page and Gumloop's file upload reference.
- 03
Attach the source and decide who else gets its answers.
Start a new AI agent, and on the tab named Agent find Knowledge Sources, click + Source and pick yours. Use Custom to leave out files that shouldn't be searched. A Team or Organization source answers everyone who uses the AI agent, and a Personal source works only in your own chats, so decide who uses it before you add an episode that hasn't gone out. It gets Search Company Brain and Read document and decides when to search. Checked on 2026-11-11 against Gumloop's Brain documentation, Gumloop's help page on attaching Brain sources and the section on the tab named Agent in Gumloop's AI agent documentation.
- 04
Keep chat from changing what the AI agent searches.
Turn off Edit Own Instructions, right under the instructions box. Change which episodes it may search yourself, in the Knowledge Sources section. Checked on 2026-11-10 against the self-improving instructions section of Gumloop's AI agent documentation.
- 05
Put the block in and switch off what could stand in for the source.
Paste the block into the instructions box and save. In Abilities, switch off Web Search, Web Fetch and Search Past Tasks, turn the browser off in the Browser section if it's on, and add no connector. The built-in computer is always on with internet access, so the block's line against answering from the web or memory is what keeps answers to the source. Checked on 2026-11-11 against the abilities table and the computer section of Gumloop's AI agent documentation, its browser documentation and its computer and secrets page.
- 06
Test it with a passage you know and a topic you never covered.
In a new task, ask which episode covered something you remember. Look for a file name that follows your rule, with a passage under it. Then ask about a topic no episode touched. The reply you're looking for is only the sentence the block gives it. Open the file from the first answer and find the passage.
I am an ambassador for Gumloop. Code CULTURECON26 takes 20% off your first month.
Why would a passage you know come back as nothing?
Put the file names from an answer beside your list of files. A name that isn't on the list, or a quoted passage that isn't in the file it names, means the block's quote rule needs tightening. If an answer quotes a passage and shows "file name missing," remove that file from the source, type its name as the first line of the file and add it again.
When a passage you know is in a file and the answer says nothing matches, check that the source reads Active and that the attachment's Custom selection still includes that file. Brain's page describes a Search this source box on the source's page as a preview of what's indexed. Try the passage there. Checked on 2026-11-10 against Gumloop's help page on attaching Brain sources and Gumloop's Brain documentation.
Add each new episode as a file named by the same rule. For the show notes for those transcripts, read Where Should an AI Agent Get the Timestamps in Your Show Notes?.
Frequently asked questions
What if a file's name has to change after I've uploaded it?
Remove the file and add it again under the new name.
Can it find an episode from the recording itself?
The File uploads list names no audio or video type. Work from the transcript or the show notes.
Does a new episode show up on its own?
Gumloop's sync page says direct file uploads never sync on a schedule. Add the new episode's transcript as a file, named by your rule.