A local-first personal finance advisor. Connects to your bank accounts, stores everything in SQLite on your machine, and uses an LLM agent as a conversational advisor — grounded in your real data, not generic platitudes.
Note
All data stays on your machine. No cloud storage. No telemetry. No product recommendations. See docs/PRIVACY.md for full details. If you have any questions please feel free to hit me on X! https://x.com/WesEklund
You: "What did I spend on dining this month?"
Advisor: queries your local DB, shows the answer with merchants and totals
You: "Should I pay off my credit card or keep cash reserves?"
Advisor: walks your actual balances and rates through a priority framework,
accounts for your constraints, gives a specific recommendation with the math
macOS or Linux (Windows via WSL) · Python 3.8+ · SQLite 3 · Homebrew · An Agent Skills-compatible runtime
Verify your setup
python3 --version # 3.8+
sqlite3 --version # any
brew --version # anyAll three come pre-installed on macOS. On Linux, install via your package manager.
git clone https://github.com/weklund/fiduciary.git && cd fiduciary
git config core.hooksPath .githooks
brew install plaid/plaid-cli/plaidFirst-time Plaid setup (free account required)
plaid register # opens browser — create a free Plaid developer account
plaid trial # apply for the free Trial plan (auto-approved in ~60 seconds)
plaid login # authenticate the CLI (opens browser for OAuth)
plaid keys fetch # download your API keysThe Trial plan is free for personal use (up to 10 linked institutions).
plaid link --products transactions,liabilities,investments # connect your bank (opens browser)
bash scripts/sync.sh # pull data + build local DBConnect your agent
Claude Code — open Claude Code in this directory. Skills auto-discover.
Hermes Agent:
cp CLAUDE.md ~/.hermes/SOUL.md
hermes skills add ./.claude/skills/*Other tools — any Agent Skills-compatible runtime works (Gemini CLI, Cursor, GitHub Copilot, 42+ others).
Then run the advisor intake:
/onboard
This asks about your goals, situation, and risk tolerance — then writes a personalized profile so all future advice is tailored to you.
| Command | What it does |
|---|---|
/onboard |
Financial advisor intake — sets up your personalized profile |
/sync-data |
Pull fresh transactions from Plaid |
/finance-query |
Ask any question about your finances |
/next-dollar |
"What should I do with $X?" — applies order of operations to YOUR situation |
/health-check |
Quarterly full assessment — pyramid status, ratios, action plan |
/tactics |
Optimize your cards, accounts, and tools |
/spending-audit |
Find waste, unused subscriptions, overspending |
/weekly-report |
Generate a weekly spending summary |
┌─────────────┐ ┌──────────────┐ ┌─────────────────┐
│ Plaid CLI │────▶│ SQLite DB │◀────│ CSV Statements │
│ (live sync)│ │ (unified │ │ (historical │
└─────────────┘ │ ledger) │ │ backfill) │
└──────┬───────┘ └─────────────────┘
│
┌──────▼───────┐
│ LLM Agent │
│ (skills + │
│ CLAUDE.md │
│ frameworks)│
└──────────────┘
The advisory logic is built on the CFP Board's fiduciary planning process — a layered pyramid where you never optimize a higher layer while a lower one is unstable. See docs/ADVISORY.md for the full framework.
Plaid gives ~30 days. For deeper analysis, add bank CSVs:
# Drop CSVs into data/statements/, then:
python3 scripts/ingest.pySupports: Amex, Chase, Capital One, or any CSV with Date,Description,Amount columns.
This project handles your most sensitive data. We believe you deserve full transparency about what happens with it — not just marketing claims, but a published threat model you can read and challenge.
Your financial data never touches a git-tracked file. The only external calls are to Plaid (for sync) and your chosen LLM provider (for analysis).
Recommended: Commercial API key (7-day retention, no training). Zero-trust: Ollama locally (nothing leaves your machine).
| Document | What it covers |
|---|---|
| docs/PRIVACY.md | Provider comparison, session logs, retention settings, credentials |
| docs/THREAT-MODEL.md | Full STRIDE threat analysis — 15 threats, mitigations, accepted risks, runtime comparison |
| docs/HLD-Fiduciary.md | System architecture, trust boundaries, data classification |
Connecting additional banks
Run plaid link again for each institution:
plaid link --products transactions # checking/savings only
plaid link --products transactions,investments # brokerage accountsVerify connections:
plaid item list # list all linked institutions
plaid balance --all # test that balances come backCommon issues:
- Bank not found? Most major US banks are supported. Some smaller credit unions may not be in Plaid's network.
ITEM_LOGIN_REQUIRED? Bank session expired. Re-runplaid linkto re-authenticate.- 10-institution limit: The free Trial allows 10 linked institutions.
Updating your profile
Run /onboard again anytime your situation changes:
- New job or lost income
- Major life event (marriage, kids, inheritance)
- Goals or priorities shift
- Want to add accounts or update metrics
The skill detects existing context and offers to update rather than start fresh.
Project structure
CLAUDE.md ← Advisory frameworks (committed, no PII)
scripts/
sync.sh ← Pull from Plaid + rebuild DB
ingest.py ← Rebuild DB from all sources
.claude/skills/ ← Conversational finance skills (8 skills)
data/ ← (gitignored) Financial data + SQLite DB
reports/ ← (gitignored) Generated reports
docs/ ← Deep documentation
ADVISORY.md ← How the advisory logic works
PRIVACY.md ← Data retention + provider comparison
Track everything. Automate the boring stuff. Spend deliberately on what builds the future. Cut ruthlessly what doesn't.
This tool exists because most financial apps sell you products or harvest your data, generic advice doesn't account for YOUR situation, a financial advisor costs $200-400/hour, and your bank has the data. You should too, queryable, on your terms.
See CONTRIBUTING.md. Good first contributions: CSV parsers for new banks, new advisory skills, improved category detection.
MIT