FAQ

Straight answers to the obvious questions about Forven — is it local, is it free, is my money safe, does it auto-trade, will it make me money.

Short, honest answers to the questions most people ask first. Forven is a free, open-source, self-hosted quant research lab: it helps you build, test, kill, and promote trading strategies through a disciplined walk-forward gauntlet before any real capital is involved. If you want the longer version of any answer, the linked pages go deeper.

This page covers live trading, risk, and results, so the plain version up front: Forven is a research tool. Backtest and paper results are not predictive of future performance, and nothing here is financial advice.

The basics

Is Forven really local?

Yes. Forven runs entirely on your machine. The backend (FastAPI), the UI (SvelteKit), the database (SQLite), and the semantic memory store (ChromaDB) all live on your computer under a single workspace directory, FORVEN_HOME (defaults to ~/.forven). You self-host it — clone the repo and run it. That directory holds your forven.db, config.json, local credentials, and research artifacts.

There is no Forven cloud doing your research, and no Forven account. The only network calls Forven makes are the ones you'd expect from a tool that studies markets: pulling market data, talking to your chosen LLM provider (if you use the optional agent layer), and — only if you choose to wire it up yourself — talking to an exchange (testnet by default).

Where does my data live?

In FORVEN_HOME. A typical layout:

# Default workspace location on Windows
echo $env:USERPROFILE\.forven
# Contains: forven.db, config.json, chromadb/, and your datasets

You can point FORVEN_HOME somewhere else with an environment variable before launch. Your strategies, backtests, and settings never leave that folder unless you move it.

Is my money safe? Is my account safe?

Forven cannot touch money it has no keys for. There are several deliberate layers between you and an accidental trade:

Paper is the supported mode. Execution defaults to paper (local simulation), and there is no supported live / mainnet path. Paper is optionally validated against the HyperLiquid testnet (no real funds). Wiring real exchange keys to trade live is unsupported and at your own risk.
Live (mainnet) is default-deny. Real-money execution on HyperLiquid requires you to set FORVEN_ALLOW_MAINNET=1 explicitly. The default is 0.
Your exchange keys stay on your machine. HyperLiquid credentials are read from environment variables, a local file (~/.forven/hyperliquid.json), or encrypted settings — never sent anywhere but the exchange itself. Set FORVEN_ENCRYPTION_KEY to encrypt secrets at rest.
Hard risk controls. A kill-switch, daily-loss halt, drawdown limit, and per-trade risk cap sit in front of every order.

See Going live safely for the full pre-capital checklist.

Cost & licensing

Is Forven free?

Yes — completely. Forven is free and open source under AGPL-3.0-or-later. There are no paid tiers, no subscription, no license key, and no account — every feature is included the moment you clone and run it. That includes the full local lab, the autonomous research daemon, paper trading, the risk controls, and the AI agent layer. See Everything is free.

Do I have to pay Forven for AI usage?

No. Forven never resells model tokens. The agent layer is bring-your-own-key: you connect your own provider account, and you pay that provider directly for what you use. Forven routes requests; it doesn't mark them up. See Models & providers (BYOK).

Do I need an LLM key at all?

No. You can ingest data, backtest, run the gauntlet, and paper-trade with no LLM key whatsoever. An LLM key is only needed for the AI features: the agents, the research daemon's autonomous ideation, and the deep-dive strategy chat.

How it trades

Does Forven auto-trade my money?

No. Paper is the supported mode and the default, and there is no supported live / mainnet path. Several things would have to be true at once before a single real order could be placed, and you would have to do all of them deliberately and at your own risk:

You have set FORVEN_ALLOW_MAINNET=1 (default-deny is 0).
You have switched execution mode to live and supplied your own exchange keys.
A strategy has actually survived the pipeline and been promoted to the live stage — usually with operator approval.

Read Execution modes: paper vs live for why live is unsupported and how the (at-your-own-risk) switch works.

What exchanges does Forven support?

Live order routing goes to HyperLiquid (testnet by default, mainnet only when you explicitly allow it). For market data — the candles you backtest against — Forven supports several sources including Binance/CCXT, Binance Vision, Polygon, Yahoo, and CSV import. Data sources and live execution are separate concerns.

What's the difference between paper and live trades?

Paper trades fill against your local OHLCV candle prices and are recorded in SQLite only. No order ever reaches the exchange. They run in isolated per-session sandboxes.
Live trades are real orders placed on HyperLiquid, filled at exchange prices, carrying real order IDs.

Because paper fills use local mid-prices, paper PnL is not a one-to-one proxy for live PnL — there's no slippage against a real order book, no exchange queue. Treat paper as a discipline check, not a profit forecast. More in Paper trading.

If I do go live, what stops a runaway loss?

Several hard controls, enforced before and during execution:

Control	Default	What it does
`max_drawdown_pct`	`10`	Trips the kill-switch and flattens all positions when portfolio drawdown breaches it
`max_daily_loss_pct`	`5`	Halts new trades for the rest of the UTC day once same-day realized loss breaches it
`max_risk_per_trade_pct`	`2`	Rejects any single order risking more than this share of the account
Circuit breakers	per-breaker	Pause exchange calls when HyperLiquid itself is unhealthy (true outages, not rate-limits)

The kill-switch is autonomous: it can flatten positions without waiting for you. Values above are illustrative defaults — you configure them in Settings. See Risk controls and Going live safely.

Platform & support

How do I install it? Is there a download?

There is no downloadable app or installer. Forven is self-hosted from source: you clone the repository and run a launcher script (start_all.ps1 on Windows, start_all.sh on macOS/Linux), which builds and starts the stack on your own machine. You need Python 3.11+, Node.js, and git. See Install for the exact steps.

What platforms does it run on?

Anywhere the stack runs — Windows, macOS, and Linux. The launcher scripts cover Windows (start_all.ps1) and macOS/Linux (start_all.sh). See Install and Running from source.

Something's broken — where do I start?

The Troubleshooting & recovery page covers the common ones: the scheduler stuck on a stale lock, the daemon offline, a blocked approval queue, and MCP connection problems. For day-to-day health, the /ops dashboard and its soak report tell you what's green, amber, or red.

Results & expectations

Will Forven make me money?

No one can promise that, and Forven won't pretend to. Forven is a research instrument, not a money machine. Its job is to make your process honest: to test strategies out-of-sample, to surface overfitting, and to kill ideas that only looked good in-sample.

The whole design assumes that in-sample metrics lie and that the only durable retail edge is disciplined process. Any number you see in the app — a backtest Sharpe, a paper PnL curve — is an artifact of historical or simulated data and is illustrative only. It does not predict the future. See Philosophy for why the tool is built to be skeptical of its own results.

Why does it keep killing my strategies?

Because that's the point. A strategy must survive the gauntlet — walk-forward analysis, parameter jitter, and cost-stress testing, scored for robustness — before it earns a paper slot. Implausibly good metrics are treated as suspicious, not impressive: a profit factor at or above the lookahead ceiling, or a Sharpe past the implausible ceiling, triggers an automatic reject because they usually signal a future-bar leak. Forven would rather disprove your darling cheaply than fund it expensively.

A note on the names

The marketing site uses simplified words for the pipeline. The app uses the real engineering names, and so do these docs:

public "screen" is the quick_screen overfitting gate
public "candidate" is the paper stage
public "gauntlet" and "live" mean what they say

Wherever a public synonym helps, we note it once and then use the real name.

The honest caveat

Forven is a research tool. Backtest and paper results are simulations of the past, are not predictive of future performance, and nothing in this documentation is financial advice. Paper (optionally on the HyperLiquid testnet) is the supported mode; live/mainnet is unsupported and at your own risk. Go slow, stay in paper while you learn, and decide for yourself.

Philosophy — why the tool is built to doubt its own numbers
Everything is free — Forven has no tiers or paywall; every feature is free
Execution modes: paper vs live — paper simulation vs the unsupported live path
Going live safely — the safety machinery, and why live mainnet is unsupported

On this page