Self-hosted · Single-user
AGPL-3.0 · No telemetry
Your own up-to-date record of what you hold, across every broker you use, in one place you control. Every trade is an append-only row in your own Postgres, tagged with the account it happened in; positions, cost basis and P&L are recomputed from raw lots on every read — never persisted, never rewritten — per account, and merged across all of them.
A full history. Nothing up my sleeve.
| Symbol | Acct | Side | Qty | Price | Fees | Trade date |
|---|---|---|---|---|---|---|
| VOO | IBKR | BUY | 8 | 512.40 | 1.00 | 2025-11-03 |
| NVDA | IBKR | BUY | 10 | 184.20 | 1.00 | 2025-11-21 |
| AAPL | BANK | BUY | 12 | 228.15 | 1.00 | 2025-12-05 |
| TSM | IBKR | BUY | 15 | 168.25 | 1.00 | 2026-01-09 |
| NVDA | IBKR | BUY | 5 | 171.90 | 1.00 | 2026-01-27 |
| AAPL | BANK | SELL | 6 | 241.30 | 2.50 | 2026-02-13 |
| MSFT | BANK | BUY | 6 | 415.30 | 1.00 | 2026-03-02 |
| NVDA | IBKR | SELL | 8 | 205.75 | 2.50 | 2026-03-18 |
| TSM | IBKR | BUY | 10 | 174.60 | 1.00 | 2026-04-07 |
| VOO | PENSION | BUY | 4 | 538.10 | 1.00 | 2026-05-15 |
| Append-only · rows are never updated, never deleted | ||||||
Scroll for fees & trade date →
Each line is one lot: side and a positive quantity encode direction, and a correction is a new row, never an edit. BUY fees inflate cost basis; SELL fees reduce proceeds. Every lot carries an account — matching is scoped per symbol and account, so one broker’s cost basis is never matched against a sale at another, while the VOO lots split across IBKR and PENSION still read as one merged position at right. A SELL exceeding open BUYs is truncated with a warning — shorts are not supported, and the data-health page will name the orphaned sell in print.
| Open | Qty | Avg cost | Unrlzd |
|---|---|---|---|
| AAPL | 6 | 228.23 | 112.00 |
| MSFT | 6 | 415.47 | 159.80 |
| NVDA | 7 | 175.59 | 271.35 |
| TSM | 25 | 170.87 | 463.25 |
| VOO | 12 | 521.13 | 481.40 |
No positions table exists — these figures are derived in your browser from the rows at left (against synthetic last prices), exactly as the product derives them from Postgres on every read.
curl -fsSLO …/docker-compose.yml
curl -fsSL …/.env.template -o .env
docker compose up -dFIFO and moving-average lot matching run in parallel, with per-match fee attribution — BUY fees inflate cost basis, SELL fees reduce proceeds. Monetary math is Decimal end to end; floats appear only at the display layer.
There is no positions table. Recomputing from lots on every read is the design.
Snapshots are collected on your schedule into an append-only (symbol, ts) series. Quotes that are stale upstream are rejected rather than written under a fresh timestamp. Splits live in corporate_actions and are adjusted at read time — history is never rewritten, and an adjustment is undone by deleting one row.
An optional read-only MCP server exposes the same engines the dashboard uses to Claude Code, Claude Desktop or Cursor — over Streamable HTTP with Bearer auth.
47 tools · 7 resources · 7 prompts, including get_portfolio_review_snapshot, get_concentration, pre_trade_check and compare_methods.
A data-health page answers it per symbol: price freshness judged against the collector’s own runs, missing cost basis, orphaned sells, suspected splits. When a figure is doubtful, the paper says so — in print.
500+ tests across two suites, including a parity suite asserting the MCP server’s KPIs match the dashboard byte for byte.
Deliberate scope, stated before you invest an evening — the same section the README leads with.
Runs on your infrastructure. Your keys, your disks, your rules.
AGPL-3.0: run a modified version as a service and those modifications must be published too.
Four core tables, all append-only. The database lives in a volume you control and back up.
Outbound requests: price lookups, your LLM provider if you enable the advisor, Financial Datasets if you enable enrichment. That’s the list.
Dashboard, scheduler, MCP server and CLIs ship in one pulled Docker image. No toolchain on your machine.
Which is why the correctness work is real — and why the roadmap follows one portfolio’s needs.
Docker with Compose. Nothing else.
# a directory and two fetched files
mkdir portfoliodb && cd portfoliodb
curl -fsSLO https://raw.githubusercontent.com/amosgeva/PortfolioDB/main/docker-compose.yml
curl -fsSL https://raw.githubusercontent.com/amosgeva/PortfolioDB/main/.env.template -o .env
# set two matching passwords in .env, then
docker compose up -d
docker compose run --rm dashboard python app/apply_schema.py
# dashboard at http://localhost:8501Want the 47 MCP tools too? They sit behind a separate, off-by-default profile. Set PORTFOLIODB_MCP_TOKEN in .env, then docker compose --profile mcp up -d mcp.
The image is pulled, not built: no Python, no toolchain, no compile step. Upgrade with docker compose pull.
It is a record-keeping tool. It keeps an accurate, up-to-date record of assets you already hold, wherever you hold them, and computes what follows arithmetically from those records.
It is not a broker, a custodian, or an adviser, and it replaces none of them. It holds no money, connects to no broker, places no orders, and has no way to move a single share. It never tells you what to buy, when, or where — and nothing in it is investment advice.
The author is not a licensed investment adviser, investment marketer, or portfolio manager — under Israeli law (Regulation of Investment Advice, Investment Marketing and Investment Portfolio Management Law, 5755-1995) or under the law of any other jurisdiction. Nothing in this project, its documentation, or its output constitutes investment advice, investment marketing, or a substitute for personal advice from a licensed professional that takes your individual data and needs into account.
The optional advisor runs on your model, not ours. You supply the API key or point it at a local model on your own machine; the project supplies no model, no key, and no opinions. What it reads is a one-page document you wrote describing goals and rules you set for yourself, so the question it answers is narrow: does what I hold still match what I said I wanted? Any suggestion in its output is your model’s text, generated for you, measured against your own rules. You are the only decision-maker, and you own the consequences.
The numbers are informational, not tax figures. Cost basis, realized P&L and returns are computed for portfolio tracking. They are not prepared under any tax authority’s rules — lot-matching method, currency conversion dates, and inflation adjustment may all differ — and are not suitable for tax reporting without independent verification.
Language models state wrong things confidently, and so do spreadsheets with a bad formula. Verify anything that would move money.
Use of this software is at your own risk. It is provided “as is”, without warranty of any kind, as set out in the AGPL-3.0 license.