Read time … min

Run vectorbt Research on TradingHost

Run open-source vectorbt research notebooks and parameter sweeps from GitHub on compute-sized TradingHost runtimes.

vectorbt is a vectorised Python backtester powered by NumPy and Numba. It is built for evaluating thousands of strategy configurations in seconds. TradingHost gives that research workload a sized cloud host.

By the TradingHost engineering team
  • GitHub-native deployment
  • Vectorised NumPy and Numba core
  • Thousands of backtests per second
  • Massive parameter sweeps
  • pandas-native data flow
  • Compute-sized runtimes
  • Encrypted secrets
  • Region selection
  • MCP-ready

Introduction

New to vectorbt?

vectorbt is a vectorised Python backtesting library built on top of NumPy, pandas, and Numba. Its design idea is straightforward: instead of evaluating one strategy path one bar at a time, evaluate thousands of strategy paths simultaneously across entire NumPy arrays, accelerated by Numba’s JIT compilation.

The result is a research engine that turns hours of grid search into seconds. Sweeping a dual-SMA strategy across 10,000 parameter combinations on a multi-asset dataset is a normal Tuesday for vectorbt, not a weekend project.

Two important boundaries before you plan a deployment. First, the open-source vectorbt is for research and backtesting only; it does not include live trading. Second, vectorbt is licensed under Apache 2.0 with the Commons Clause (a "fair-code" variant) and is a different product from vectorbt PRO, the paid commercial library by the same author.

TradingHost runs vectorbt research workloads on cloud runtimes sized for the work. When a strategy is ready for live trading, hand it off to a framework with a live loop, like Freqtrade, Jesse, NautilusTrader, or your own CCXT-based code.

Section 01 · At a glance

At a glance

What vectorbt is
Open-source vectorised Python backtesting library
Maintained by
polakowo/vectorbt on GitHub
License
Apache 2.0 with Commons Clause (fair-code)
Open source vs PRO
vectorbt (this page) is the open-source library; vectorbt PRO is a separate commercial product by the same author
Language
Python (with Numba JIT and an optional Rust engine)
Core capability
Evaluate thousands of backtests in parallel via vectorised NumPy arrays
Live trading
Not included in the open-source build (research and backtesting only)
Data sources
pandas DataFrames, Yahoo Finance via YFData, custom feeds
Deployment source
GitHub
TradingHost role
Compute-sized runtime for research jobs and parameter sweeps
Affiliation
TradingHost is independent and is not affiliated with the vectorbt project
Best for
Strategy screening, parameter sweeps, indicator combinatorics, walk-forward research

Section 02 · Where vectorbt fits

How vectorbt fits into the trading stack

vectorbt sits at the research end of the stack. It is not a complete trading system; it is a screening engine that narrows a large parameter space down to a handful of configurations worth running live.

That leads to a clean two-step pattern on TradingHost:

  • Path 01

    Screening and parameter sweeps

    Use vectorbt to evaluate thousands of strategy configurations against historical data. Score each configuration, rank, and pick the candidates that justify forward testing.

  • Path 02

    Hand-off to a live framework

    Once a configuration is selected, port the rules to a framework that does live trading: Freqtrade, Jesse, NautilusTrader, or your own CCXT-based code. vectorbt does the screening; the live framework does the execution.

Both halves run on TradingHost. The research host is sized for compute; the live host is sized for uptime.

You
Your research notebook owns the strategy ideas and parameter spaces.
vectorbt
Runs the vectorised backtester and scoring across the parameter space.
Live runtime
A separate framework (Freqtrade, Jesse, NautilusTrader, custom CCXT) runs the chosen configuration live.
TradingHost
Provides the compute for vectorbt research and the runtime for whichever live framework you pair it with.

Section 03 · Why on TradingHost

Why deploy vectorbt research on TradingHost

vectorbt research workloads are compute-bound rather than uptime-bound. A serious parameter sweep wants memory for the array dimensions, cores for parallelism, and the tolerance to run for hours without crashing your laptop. Running it locally is workable but slow, fragile, and ties up your machine.

TradingHost gives a vectorbt research workload a sized cloud runtime, with the option to pair it with a live runtime when a configuration is ready to trade.

What you get

  • GitHub-native deployment
  • Compute-sized runtime for sweeps and walk-forward research
  • Multi-core hosts for parallel parameter searches
  • Persistent runtime for long-running notebooks
  • Encrypted secrets for any API-keyed data sources
  • Streaming logs and basic health
  • Save and re-run jobs without leaving them on your laptop
  • Pair with a live runtime when a configuration graduates
  • MCP-compatible AI IDE workflows

What you keep

  • Full control over your vectorbt notebooks and feature definitions
  • Direct ownership of any data licences and API keys
  • Choice of pure vectorbt or pairing with downstream live frameworks

Section 04 · Data and compute

Data sources and compute scale

vectorbt is exchange-agnostic. It does not need broker credentials. What it needs is good data and enough compute to run a sweep.

Data sourceUse on TradingHost
pandas DataFramesIn-process pandas data, including data downloaded from any source you can pip install
YFData (Yahoo Finance)Built-in Yahoo Finance feed for equities and ETFs
Custom data feedsAnything you can express as a pandas DataFrame: CCXT-pulled OHLCV, parquet files, your own data warehouse
Saved arraysPre-computed NumPy arrays committed to your repo for reproducible research

Why scale matters

  • 10,000-combination parameter sweeps in seconds rather than hours
  • Multi-asset, multi-timeframe arrays evaluated in one vectorised pass
  • Numba JIT compilation for the hot path; an optional Rust engine in v1.0+
  • Memory-bound by the array dimensions, not by sequential bar evaluation
Pick a runtime size

Section 05 · Deployment flow

How a vectorbt deployment works

A vectorbt research run on TradingHost follows the same deployment pattern as a live strategy, with one difference: the runtime is sized for compute rather than uptime.

  1. STEP 01

    Connect or create a repo

    Sign in with GitHub and connect a research repo containing your vectorbt notebooks, parameter spaces, and any pre-computed data.

  2. STEP 02

    Pin vectorbt and dependencies

    Pin the vectorbt version you want (vectorbt open source, not vectorbt PRO). Numba and any data adapters install with the rest of your Python deps.

  3. STEP 03

    Add encrypted data credentials

    If your sweeps pull data from APIs (Polygon, Tiingo, Quandl-equivalents, your own warehouse), add the credentials as TradingHost secrets.

    /platform/secrets/
  4. STEP 04

    Choose a region and runtime size

    Pick a region close to your data source and a runtime size matched to your sweep dimensions. Larger sweeps want more cores and memory.

    /platform/regions/
  5. STEP 05

    Run

    TradingHost runs your vectorbt notebook or script in a managed container. Logs and result artifacts stream back to the dashboard.

    /platform/observability/
Deploy →

Section 06 · Open source vs PRO

vectorbt open source vs vectorbt PRO

"vectorbt" can mean two different products. The open-source vectorbt is the focus of this page. vectorbt PRO is a separate commercial library by the same author. Both run on TradingHost; the choice is yours.

The licence on the open-source build is Apache 2.0 with the Commons Clause, a "fair-code" variant that limits selling vectorbt itself but does not restrict using it for your own research.

vectorbt (this page)

Open-source build under Apache 2.0 with Commons Clause. Vectorised backtester, indicators, and parameter sweep tooling. No live trading.

  • Free to install and use for research
  • GitHub: polakowo/vectorbt
  • Best for parameter sweeps, screening, indicator combinatorics
  • Stable, mature, version 1.0+ as of 2026

vectorbt PRO

Commercial product by the same author. Hybrid backtesting library with a richer API surface and additional features beyond the open-source scope.

  • Paid licence (vectorbt.dev)
  • Distinct from vectorbt open source
  • Not interchangeable; the import paths and APIs diverge
  • TradingHost runs PRO if you supply a valid licence

Section 07 · What we don’t do

What TradingHost does not do

  • TradingHost does not place trades from a vectorbt notebook (vectorbt open source does not include live trading).
  • TradingHost does not provide trading strategies, signals, or financial advice.
  • TradingHost does not modify your vectorbt notebooks, feature definitions, or scoring logic.
  • TradingHost does not custody data licences or API keys beyond encrypted runtime injection.
  • TradingHost does not sell or distribute vectorbt PRO; bring your own licence if you choose to use it.

Section 08 · FAQ

FAQ

No. TradingHost is independent. vectorbt is open-source software at polakowo/vectorbt under Apache 2.0 with Commons Clause; vectorbt PRO is a separate commercial product by the same author.

No. The open-source vectorbt is a research and backtesting library only. It does not include live trading. Pair vectorbt with a live framework like Freqtrade, Jesse, NautilusTrader, or your own CCXT-based code when you are ready to go live.

Vectorised backtesting at scale: evaluating thousands of strategy configurations across multi-asset, multi-timeframe data in seconds rather than hours. Numba JIT and an optional Rust engine accelerate the hot path.

Apache 2.0 with the Commons Clause, often called a "fair-code" licence. You can use it freely for your own research; the Commons Clause adds restrictions on selling vectorbt itself.

No. They are separate products with different licences, different APIs, and different feature sets. TradingHost runs either one. Bring your own PRO licence if you choose PRO.

Anything that can be expressed as a pandas DataFrame, plus a built-in Yahoo Finance feed via YFData. Many users pre-compute data in another step, save it as parquet or NumPy arrays, and feed it into vectorbt for the heavy parameter sweep.

Sized compute. Parameter sweeps want cores, memory, and the freedom to run for hours without locking up a workstation. TradingHost gives a vectorbt notebook a runtime sized for the work and stores the results.

Yes. The most common pattern is: run vectorbt on TradingHost to screen and select configurations, then deploy the chosen configuration to another TradingHost runtime running Freqtrade, Jesse, NautilusTrader, or a custom CCXT-based strategy.

Only if your data source requires them. vectorbt itself is exchange-agnostic and does not need broker credentials. Data API keys (Polygon, Tiingo, your own warehouse, and so on) live in TradingHost as encrypted secrets.

Backtrader is event-driven and supports live trading, well suited to bar-by-bar logic and live execution. vectorbt is vectorised, dramatically faster for parameter sweeps, but the open-source build does not include live trading. Many teams use both.

Run your first vectorbt sweep

Connect GitHub, drop in a vectorbt notebook, pick a runtime size matched to your parameter space, and run. When a configuration earns its keep, deploy it live with one of the live frameworks above.