← All workCase 4 of 4
Case 04EXPLORATION PROJECT · iOS

Your saved content, back automatically when it's relevant

An iOS app that turns your Instagram, TikTok and other saved content from a black box you never return to into useful information, surfaced only when it's relevant, using AI transcription and MCP.

RoleBuilt solo
StackClaude Code, Railway, MCP
StatusPaused
Read3 min
Quiver home and library
Three-step story of saving a post, losing it, and never going back, then Quiver finding it
ChatGPT answering from saved Quiver posts through MCP
1 / 3 · Quiver home and library
Step 01 / 05
Use case

The user moment

How many reels have you saved and never opened again? A recipe, a workout, a business tip, all buried in a saved folder you'll never scroll back through. The moment you actually need one, you can't find it.

Step 02 / 05
Problem

The problem

Social apps make saving easy and finding impossible. The value of a save shows up weeks later, at the moment it's relevant, and by then it's lost in the pile.

Instagram reelsTikToksAny app's share sheetQUIVERAI reads it for you01Transcribe the video02Pull out key takeaways03Save to your libraryWHEN YOU NEED ITClaude + ChatGPTsurface the right savethrough MCPWHERE SAVES COME FROM
Step 03 / 05
Validated

How I validated it

15 conversationsWith friends and family about what they save and why they never go back to it.
Reddit threadsPeople describing the same lost-saves problem in their own words.
PodcastsPodcast discussions of the same problem, confirming it isn't just my circle.
Related appsRodeo turns saved posts into plans, Pestle and other recipe apps save dishes from Reels, Soonlist saves events, and Reelyze and Rescroll serve creators. Saved social content has demand in many niches.
Step 04 / 05
Decided

Decisions I made

Every call was about retrieval, not storage.

ConsideredStore the full videos
ChoseAI transcribes and keeps the takeaways
Meta and TikTok privacy rules don't allow storing their videos.
ConsideredOwn AI takeaways and chatbot in the app
ChoseMCP connection to Claude and ChatGPT
Going to ChatGPT or Claude is already the natural flow. Quiver meets people there.
ConsideredStorage as the pitch
ChoseRetrieval at the moment of relevance
Storage isn't the bottleneck. Finding what you saved when it matters is.
ConsideredOne top model for every step
ChoseThe cheapest model that does each job
Claude Haiku 4.5 handles the breakdown. It keeps the target under $0.12 per analyzed save.
The path

How the idea evolved

Same codebase, three names.

1
CairnSave and recall

The original idea: save and recall for anyone saving content online, with it returning at the moment it matters. The problem: the user focus was too broad.

2
Cairn ContentAimed at creators

A narrower market: creators, a group technical enough to use AI constantly and see the MCP potential. The problem: I repurposed an app built for anyone saving content, and it didn't connect.

3
QuiverRenamed Sept 19, 2026

Redesigned around creators' biggest problem: saving content ideas that work and adapting them to their audience. A fun, personalized UI with a mascot. But competition grew after Muse launched, and user behavior changed.

Under the hood

How it's built

Cheapest model that does each job. MCP brings the saves back.

1
SaveiOS share sheet

One tap from Instagram or TikTok through a native Swift share extension. The app never stores or plays the source video.

2
TranscribeApify + AssemblyAI

Media is pulled server-side, then transcribed. Nothing scraped lives in the app.

3
Break downClaude Haiku 4.5

Turns each transcript into format, hook, structure, pacing and CTA. Picked as the cheapest model that does the job; target cost under $0.12 per analyzed save.

4
IndexVoyage AI + pgvector

Embeddings make saves searchable by meaning, not keywords.

5
ReturnMCP server

A custom MCP server with OAuth exposes search_saves, get_recent_saves and get_save_details to Claude, ChatGPT and Cursor. I wrote the tool descriptions like UX copy.

App in React Native (Expo). Server on Node and Railway. Database and auth on Supabase. No third-party keys in the app, row-level security from day one.
Step 05 / 05
Status

Where it stands

How far has the idea moved?3versionsCairn, then Cairn Content, now Quiver
What did I build and learn?iOS
RAG
MCPiOS development, working on my phoneOn my phone, it analyzes saved content and returns it through MCP. I learned RAG, vector search and MCP connections and databases, and how to build all of them. My first React and mobile app.
Is it worth continuing?No-GodecisionAI is being adopted fast, and Muse now automates much of what Quiver does.
What I learned

Riverline taught me to test demand before building the business. Quiver put that to the test: Muse and other AI agents now deliver my key value prop, at a much wider scale and faster pace. So I stepped back, re-evaluated, and decided not to carry this opportunity through. You don't have to continue every product. Adapting to the market, customer needs and competitors is a core part of product management.