Murmur
Talk through your spending.
Let a person record spending and revisit money context through a conversation.



Captured from the development build
How it works
- 01
Say it
A user describes a purchase or asks a question.
- 02
Interpret
The model path proposes a reply and action; a fallback parser supports simpler requests.
- 03
Apply the action
The app executes a supported action and updates local ledger or memory records.
- 04
Continue the conversation
Recent messages and saved context shape the next exchange.
The product idea
A conversation that can become a record.
Use conversation to capture spending while making the step from a proposed action to a saved record visible.
- Input design
Start with the person’s words
Turn everyday spending descriptions into supported structured actions, alongside the conversation that produced them.
- User trust
Separate suggestion from completion
Validate an action before saving it locally; a generated reply alone does not mean a transaction was stored.
- Resilience
Keep a smaller fallback available
Provide a simpler local route when the agent route is unavailable, with a narrower set of capabilities.
Independent projectIndependent conversational ledger; external language and voice features depend on configuration.
Stack
SwiftUI · SwiftFinanceCore · SwiftDataStructured reply/action JSONHeuristic parsing fallbackConversation + memory modelsElevenLabs voice SDK
Implementation details
A suggested action and a successful save are distinct steps. Service configuration matters.
- Everyday language
- Describe a purchase in your own words.
- Chat ledger
- Show the conversation alongside structured actions.
- Action agent
- Convert intent into supported log or memory actions.
- Local fallback
- Use a narrower local path when the agent route is unavailable.
- Finance records
- Store validated actions as local records.
- Language & voice
- Configured services support interpretation and optional live voice.