Natural Language

MIRA can capture actions from free text entered in the lower panel.

How it works

General pipeline:

1. text normalization
2. interpretation by the active engine (deterministic or local LLM)
3. schema validation
4. action execution

If interpretation fails, MIRA does not stop the app. It shows a help message with examples instead.

Supported actions

• add_income
• add_expense
• report
• data_analysis
• none

Recommended examples

Income:

• "salary income 1200"
• "income 300 from sales"

Expense:

• "spent 25 on transport"
• "paid 80 for utilities"

Technical savings for goals:

• "saved 100 cordobas from change"
• "moved 300 dollars to savings from salary"
• "saved 50 dollars from my paycheck"

Important note:

• Savings phrases are stored as a technical outflow in an internal savings category so MIRA can update savings goals.
• That outflow is not treated as real consumption expense in reports, analysis, or budgets.
• Categories marked as savings do not appear as budgetable categories.

Reports:

• "show report"
• "show expense report"
• "analyze my last month data"

Quick buttons

When the system answers with action none, quick actions appear for:

• Add Income
• Add Expense
• View Report

These buttons preload a template into the text box.

Command history

Inside the input box you can use:

• Up arrow: previous command.
• Down arrow: next command or current draft.

Assistant vs Chat

• Assistant: interprets and executes financial actions.
• Chat: free conversation with a local model.

Note:

• Chat is available only when a local LLM is ready.
• If there is no active model, MIRA falls back to the deterministic assistant parser.