Treeline

Treeline Money

Chat with your finances. Ask questions like "What's my net worth?", "How much did I spend on groceries?", or "Am I over budget?" and get instant answers from your own financial data.

Quick Start

# 1. Install the CLI (OpenClaw handles this automatically)

# 2. Enable demo mode (sample data)
tl demo on

# 3. Try it out
tl status

First Time Setup

For agents: If tl commands fail with "command not found", the CLI needs to be installed. OpenClaw handles installation automatically via the skill metadata. Start with demo mode so users can try queries immediately.

Verify the CLI is available with tl --version. Start with demo mode so users can try queries immediately.

Optional: Download the desktop app for visual exploration of your data.

Demo Mode

Demo mode loads sample data so users can try queries without connecting a bank:

tl demo on

To switch to real data later:

tl demo off

Demo data is separate from real data.

CLI Behavior Notes

• tl demo on prints a success message — if it seems to hang, wait a few seconds (first run initializes the database)
• Use tl demo status to verify demo mode is enabled
• Some commands may take a few seconds on first run due to database initialization
• If you see errors about missing tables, try tl demo on again

Connecting Real Data

When the user is ready to move beyond demo mode, direct them to set up a data source with the guides linked below.

Data source options:

• SimpleFIN ($1.50/month, US & Canada)
• Lunch Flow (~$3/month, global)
• CSV Import (free)

Setup guides: Bank Sync · CSV Import

Once set up, use tl sync to pull bank transactions or tl import to load a CSV.

What is Treeline?

Treeline Money is a local-first personal finance app. All your data stays on your device in a local DuckDB database. No cloud accounts, no subscriptions required (sync services are optional), full SQL access to your financial data.

Limitations

Encrypted databases not supported. If the user has enabled database encryption in Treeline, CLI commands will fail. They'll need to either:

• Disable encryption if they want OpenClaw access
• Use the Treeline app directly for encrypted databases

If you see "database is encrypted" errors, explain this limitation.

Response Formatting

Format all responses for mobile/chat:

• Use bullet points, not markdown tables
• Round numbers for readability ($1,234 not $1,234.56)
• Lead with the answer, details second
• Keep responses concise — chat isn't a spreadsheet
• Use line breaks to separate sections

Example good response:

Your net worth is $125k

Assets: $180k
- Retirement: $85k
- Savings: $25k
- Checking: $10k
- Home equity: $60k

Liabilities: $55k
- Mortgage: $52k
- Credit cards: $3k

Example bad response:

| Account | Type | Balance |
|---------|------|---------|
| My 401k Account | asset | 85234.56 |
...

CLI Commands

Read commands (run freely)

These commands are read-only and safe to run autonomously:

tl status              # Quick account summary with balances
tl status --json       # Same, but JSON output

tl query "SQL" --json  # Run any SQL query (database opened in read-only mode)
tl sql "SQL" --json    # Same as tl query (alias)

tl backup list         # List available backups
tl doctor              # Check database health
tl demo status         # Check if demo mode is on/off

Note: tl query and tl sql open the database in read-only mode by default. They cannot modify data unless --allow-writes is passed (see write commands below).

Use tl status for quick balance checks — it's faster than a SQL query.

Write commands (ask the user first)

These commands modify local data. Always ask the user for confirmation before running them, unless the user has explicitly allowed them in PERMISSIONS.md (see Agent Permissions).

tl query "SQL" --allow-writes --json  # Run a SQL query with write access
tl sql "SQL" --allow-writes --json    # Same (alias)

tl sync                # Sync accounts/transactions from bank integrations
tl sync --dry-run      # Preview what would sync (read-only, safe to run)

tl import FILE -a ACCOUNT          # Import transactions from CSV
tl import FILE -a ACCOUNT --dry-run  # Preview import without applying (read-only, safe to run)
tl import FILE -a ACCOUNT --json   # JSON output for scripting

tl backup create       # Create a backup
tl backup restore NAME # Restore a backup

tl compact             # Compact database (reclaim space, optimize)

tl tag "groceries" --ids ID1,ID2  # Apply tags to transactions

tl demo on|off         # Toggle demo mode (sample data)

Tip: --dry-run variants are read-only and safe to run without confirmation. Use them to preview before asking the user to confirm the actual operation.

Use tl compact if the user mentions slow queries — it optimizes the database.

CSV Import Details

tl import auto-detects column mappings from CSV headers. Most bank CSVs work out of the box:

tl import bank_export.csv --account "Chase Checking"

The --account / -a flag accepts an account name (case-insensitive, substring match) or UUID.

Always preview first with --dry-run to verify columns were detected correctly:

tl import bank_export.csv -a "Checking" --dry-run --json

All import flags (all optional except --account):

Common patterns for agents:

# Step 1: Find the account UUID
tl status --json

# Step 2: Preview import
tl import transactions.csv -a "550e8400-e29b-41d4-a716-446655440000" --dry-run --json

# Step 3: Execute import
tl import transactions.csv -a "550e8400-e29b-41d4-a716-446655440000" --json

Duplicate transactions are automatically detected and skipped on re-import via fingerprinting.

Agent Permissions

Before running any write command, check for PERMISSIONS.md in this skill directory.

If it exists, read it to see which write commands the user has pre-approved. Pre-approved commands can be run without asking for confirmation. All other write commands still require explicit user confirmation before execution.

If PERMISSIONS.md does not exist, always ask before running any write command.

Template for PERMISSIONS.md:

# Treeline Agent Permissions

Commands listed here are pre-approved — the agent can run them without
asking for confirmation each time. Remove a line to require confirmation.

## Allowed write commands
- tl sync
- tl backup create
- tl demo on|off

User Context

Before answering finance questions, check for CONTEXT.md in this skill directory.

If it exists, read it first — it contains user-specific knowledge:

• Account meanings (which accounts are retirement vs brokerage, etc.)
• Tag conventions and cash flow rules
• Plugin configurations
• Personal preferences

Learning new context: When you discover something about the user's setup:

1. For small observations, note them in CONTEXT.md and briefly mention what you saved
2. For significant assumptions or corrections, ask: "Want me to save that to your Treeline context?"

See the User Context Pattern section at the end for the template.

Quick Reference

Net Worth

tl query "
WITH latest AS (
  SELECT DISTINCT ON (account_id) account_id, balance
  FROM sys_balance_snapshots
  ORDER BY account_id, snapshot_time DESC
)
SELECT
  SUM(CASE WHEN a.classification = 'asset' THEN s.balance ELSE 0 END) as assets,
  SUM(CASE WHEN a.classification = 'liability' THEN ABS(s.balance) ELSE 0 END) as liabilities,
  SUM(CASE WHEN a.classification = 'asset' THEN s.balance ELSE -ABS(s.balance) END) as net_worth
FROM accounts a
JOIN latest s ON a.account_id = s.account_id
" --json

Account Balances

tl query "
WITH latest AS (
  SELECT DISTINCT ON (account_id) account_id, balance
  FROM sys_balance_snapshots
  ORDER BY account_id, snapshot_time DESC
)
SELECT a.name, a.classification, a.institution_name, s.balance
FROM accounts a
JOIN latest s ON a.account_id = s.account_id
ORDER BY s.balance DESC
" --json

True Spending (Excluding Internal Moves)

Check CONTEXT.md for internal_transfer_tags. Default pattern:

tl query "
SELECT SUM(ABS(amount)) as total_spent
FROM transactions
WHERE amount < 0
  AND transaction_date >= date_trunc('month', current_date)
  AND NOT (tags && ARRAY['transfer', 'savings', 'investment'])
" --json

Spending by Tag

tl query "
SELECT tags, SUM(ABS(amount)) as spent
FROM transactions
WHERE amount < 0
  AND transaction_date >= '2026-01-01' AND transaction_date < '2026-02-01'
  AND tags IS NOT NULL AND tags != '[]'
GROUP BY tags
ORDER BY spent DESC
" --json

Recent Transactions

tl query "
SELECT t.description, t.amount, t.transaction_date, a.name as account
FROM transactions t
JOIN accounts a ON t.account_id = a.account_id
ORDER BY t.transaction_date DESC
LIMIT 10
" --json

Database Schema

Core Tables

accounts

sys_balance_snapshots — Source of truth for balances

transactions

Tags vs Categories

Tags are the primary concept in Treeline — transactions can have multiple tags.

Categories come from the budget plugin (plugin_budget), which maps tags to budget categories. Not all users have this plugin.

Plugin System

Plugins have their own DuckDB schemas: plugin_<name>.*

Discovering Installed Plugins

tl query "
SELECT schema_name
FROM information_schema.schemata
WHERE schema_name LIKE 'plugin_%'
" --json

Common Plugin Schemas

plugin_budget.categories — Budget categories

plugin_goals.goals — Savings goals

plugin_subscriptions — Detected recurring charges

plugin_cashflow — Cash flow projections

plugin_emergency_fund — Emergency fund tracking

Check CONTEXT.md for which plugins the user has and cares about.

Common Patterns

Getting Current Balances

Always use latest snapshot:

WITH latest AS (
  SELECT DISTINCT ON (account_id) account_id, balance
  FROM sys_balance_snapshots
  ORDER BY account_id, snapshot_time DESC
)
SELECT a.name, s.balance
FROM accounts a
JOIN latest s ON a.account_id = s.account_id

Working with Tags

Tags are arrays:

-- Contains a specific tag
WHERE tags @> ARRAY['groceries']

-- Contains any of these tags
WHERE tags && ARRAY['food', 'dining']

-- Note: UNNEST doesn't work in all contexts in DuckDB
-- Instead, GROUP BY tags directly

Date Filters

-- This month
WHERE transaction_date >= date_trunc('month', current_date)

-- Specific month
WHERE transaction_date >= '2026-01-01'
  AND transaction_date < '2026-02-01'

Budget vs Actual

SELECT
  c.name,
  c.expected,
  COALESCE(SUM(ABS(t.amount)), 0) as actual,
  c.expected - COALESCE(SUM(ABS(t.amount)), 0) as remaining
FROM plugin_budget.categories c
LEFT JOIN transactions t ON t.tags && c.tags
  AND t.amount < 0
  AND t.transaction_date >= (c.month || '-01')::DATE
  AND t.transaction_date < (c.month || '-01')::DATE + INTERVAL '1 month'
WHERE c.month = strftime(current_date, '%Y-%m')
  AND c.type = 'expense'
GROUP BY c.category_id, c.name, c.expected

Question Mapping

Tips

1. Always use --json for parseable output
2. Amounts are signed — negative = expense
3. Use classification for asset/liability
4. Balances live in snapshots, not the accounts table
5. Check CONTEXT.md for user-specific account meanings and tag conventions

User Context Pattern

When this skill is installed, create CONTEXT.md alongside it to store user-specific knowledge. This keeps the skill generic/shareable while personalizing behavior.

Template for CONTEXT.md:

# Treeline User Context
*Auto-updated by your assistant as it learns your setup*

## Account Notes
<!-- What specific accounts mean, e.g.: -->
<!-- - "Company 401k" = retirement account -->
<!-- - "Home Equity" = home value estimate (manual) -->

## Tag Conventions
<!-- How the user uses tags -->

## Cash Flow Rules
<!-- Tags to exclude from "true spending" calculations -->
internal_transfer_tags: [transfer, savings, investment]

## Income Sources
<!-- Known income sources for better reporting -->

## Active Plugins
<!-- Which plugins are installed and relevant -->

## Preferences
<!-- Reporting style, rounding, spouse-friendly mode, etc. -->

## Learned Facts
<!-- Anything else discovered about the user's financial setup -->

Maintenance:

• Briefly mention updates for small observations
• Ask before recording significant assumptions
• Periodically validate against live data (accounts may close, tags may change)

Privacy Note

All data is local (~/.treeline/treeline.duckdb). Never share transaction descriptions or account details outside the conversation unless explicitly asked.