sciqnt

A local-first, agent-native portfolio tracker & cross-asset financial data layer. One canonical, point-in-time-correct schema for your positions, transactions, cash and prices — fed by community connectors for each broker / exchange / data source, rendered in a fast TUI, and consumable by any AI agent (Claude Code, Codex, …) through plain CLI + versioned JSON surfaces.

Deterministic code computes the numbers. Agents reason and explain. You own every byte: your credentials live in your OS keyring, your data on your disk — there is no server.

  net worth (EUR) │ daily

  24,570 ┤                                                       ⢀⡤⠖⠋⠹
         │                                 ⣀⡤⣄⣤⣀⣀⣀⣀⣀  ⡤⠤⠤⠤⣄ ⣀⣀⣀⣀⡤⠞
  22,019 ┤                            ⢀⣠⢤⣰⠒⠃     ⠉ ⠈⠉⠛⠁   ⠈⠉⠁⠉⠈⠁
         │                    ⢀⣰⠒⠚⠉⠉⠓⠒⠋
         │⣀          ⣀⢀⣀⣠⠤⠖⠋⠛⠉⠉
  19,468 ┤⠈⠙⠒⢦⣀⡤⠤⣤⣄⣠⠴⠋⠉
         ╰────────────────────────────────────────────────────────────
          2026-01-02                                        2026-06-12

  total value             24,121.05 EUR      TWR        13.70 %/yr
  total P/L (lifetime)    +3,340.95 EUR      XIRR        9.79 %/yr
  dividends (lifetime)      +395.10 EUR      max DD     −14.5 %

That's the built-in demo portfolio — synthetic and deterministic. sciqnt runs in demo mode until you connect an account; no real finances appear in any screenshot or doc, ever.

Install

git clone https://github.com/sciqnt/sciqnt && cd sciqnt
python3 -m venv .venv && .venv/bin/pip install pydantic prompt-toolkit keyring
./bin/sciqnt install        # adds `sciqnt` to your PATH
sciqnt                      # the interactive home (demo portfolio until you connect)

PyPI packages (uv tool install sciqnt, pip install sciqnt-degiro, …) are coming with the first tagged release. macOS note: use a Python built against modern OpenSSL (e.g. Homebrew python@3.13) — the system Python's LibreSSL is fragile against financial-API TLS.

What you get

• The TUI: portfolio home with net-worth chart (1D…All ranges, 5-minute intraday), positions / exposure / income / news / flows / history tabs, account drill-downs — everything keyboard-driven.
• Honest money math: Decimal end-to-end, FIFO/LIFO/AVG lots, fees-inclusive cost basis, TWR (GIPS-style breaks), XIRR, max drawdown, benchmark comparison — computed from your raw transaction history, point-in-time-correct (sciqnt --asof 2024-12-31).
• Connectors as self-contained bundles (modules/sq-*): manifest + agent-facing SKILL + a living quirks log (FINDINGS) + conformance tests. Degiro (CSV + live), Robinhood, Kalshi, Polymarket, Yahoo, Tiingo, ECB FX, SEC EDGAR, FIRDS, OpenFIGI, RSS news — and a scaffold + harness for building your own.
• Agent-native, both directions: every view is reproducible from the CLI (sciqnt --help maps it; --json gives versioned, Decimal-as-string data — sciqnt.portfolio/v1, sciqnt.history/v1, …). Summon your coding agent from any screen and it receives where you are, what's on your screen, and the command that reproduces it; agents can leave findings on your home screen (sciqnt insight add).
• A point-in-time price archive: append-only, bitemporal, yours.

For agents

Run sciqnt --help. Every screen of the app has a CLI form; add --json for structured data. Skills ship in-repo (sq-portfolio, sq-connectors) and install into Claude Code / Codex automatically when summoned from the app. AGENTS.md is the codebase map — start there if you're an agent.

Build a connector for your broker

The platform ships the contract + conformance harness + a scaffold — the long tail of connectors belongs to the community (and to your coding agent). See CONTRIBUTING.md; the short version:

"build a sciqnt connector for <my broker>"   # tell your coding agent

Independent connector repos install with sciqnt modules add owner/repo (conformance runs locally before first use — trust is earned by the harness, not claimed).

Going deeper

• FOUNDATION.md — the worldview + the 13 Founding Articles.
• PRINCIPLES.md — the 18 operating principles.
• research/ — the grounded reasoning behind every decision.

Principles, the short list: local-first and sovereign — fire us and keep everything. Deterministic core, probabilistic edge — LLMs never touch the money math. Data first — rendered text is for humans, versioned JSON is the contract. Read wide, execute gated. Synthetic fixtures only.

License

MIT — see LICENSE.