mrkeyoor.comAsk

NoFxAiOS/nofxtested 27 Sept 2026commit 638d404

nofx review

NOFX merits a supervised evaluation if you want to inspect AI-generated trading plans or study model-directed execution.

  • Install 67 s
  • Build 119 s
48/48tests passed
  • 270 packages
AI Toolsevaluationupdated 01 Nov 2026
+19stars / 7d

Consider NOFX for supervised experimentation with AI-assisted trading if you can inspect exchange orders and manage the host yourself. Our run on September 27, 2026, at commit 638d404 recorded successful install, build and test command outcomes in our sandbox; it did not establish live-trading readiness. NOFX is a self-hosted trading system with two execution modes: Frontier requires approval for each trade, while Autopilot can place trades while running.

Verdict

NOFX merits a supervised evaluation if you want to inspect AI-generated trading plans or study model-directed execution. Our September 27, 2026 run at commit 638d404 records successful command outcomes, not a live-trading validation. Confirm your adapter's entry, protection, cancellation and recovery behavior before considering live funds. Choose Freqtrade for its documented dry-run and backtesting workflow, or Jesse when you want to define and research your own Python strategies.

We ran it

Lab card: what happened when we ran nofx
Lab card from our sandbox run
Screenshot of nofx (vergex.trade/explore)
vergex.trade/explore
Install✓ · 67s270 packages
Build✓ · 119s
Tests✓ · 119s48 passed · 0 failed of 48 (go test)
Repo647 files~122,028 lines of source · 27.5 MB · 9 CI workflows

Answers from our run

Does nofx build from source?

Dependencies installed in 67 seconds (270 packages), and the build succeeded in 119 seconds. We cloned commit 638d404 into a clean Debian container with 3 CPUs and no project-specific setup.

Do nofx's tests pass?

Yes. The project's own test command finished in 119 seconds and reported 48 passed, 0 failed (go test).

Who should not use nofx?

Anyone seeking evidence of profitability or safe unattended trading: our sandbox run supplies neither.

What are the alternatives to nofx?

Freqtrade, Jesse. NOFX merits a supervised evaluation if you want to inspect AI-generated trading plans or study model-directed execution.

Setup3/5Our install, build and test command outcomes succeeded; browser setup, exchange execution and remote deployment remain unverified.
Docs4/5The current README explains execution modes, installation, remote exposure, separate fee balances and recovery limitations.
Community4/5Last push on 2026-09-05 and 13,283 stars.Editor’s note: The supplied primary extracts do not establish maintainer response times, contributor continuity or issue-resolution performance.
Maturity2/5Entry and protection are documented as non-atomic, and our stored run does not validate trading execution or recovery.

How we score

Who it’s for

Developers evaluating model-directed trading who can test their chosen exchange adapter before funding live execution.
Traders who want to inspect an AI-generated plan and approve execution through Frontier.
Operators prepared to protect local credentials, signing material, configuration and backups.

Who it’s NOT for

Anyone seeking evidence of profitability or safe unattended trading: our sandbox run supplies neither.
Users who cannot afford trading losses or cannot monitor positions at the exchange.
Teams requiring demonstrated browser, deployment, reconciliation and recovery coverage before adoption: those checks are not established by the supplied run.

Setup reality

Our run on September 27, 2026, tested commit 638d4042118995fbf1a38d3822b1139aa3c6b467 in a sandbox configured with 3 CPUs, 8 GB RAM and the golang:1.24-bookworm image. The attached harness records installation at 67 seconds, build at 119 seconds and the test step at 119 seconds, each with exit code 0 and no timeout. Installation recorded 270 dependencies. The install tail shows that Go downloaded toolchain go1.25.11, so the container image name alone does not identify the toolchain used.

The stored parser summaries record 48 passing Go package results and 0 failing package results. These are package summaries, not 48 individually verified tests or exhaustive coverage. The raw record also lists 20 packages without test files; those are not failed or skipped tests. No new execution or complete-log reconciliation was performed for this correction.

The current README documents installation through Docker with Compose and published images. Development without Docker now refers readers to the Go version in go.mod, Node.js 24 and TA-Lib. Those current instructions are separate from our older run. A zero command exit does not demonstrate that the application started successfully, that its browser workflow worked or that an exchange order completed.

Should you try NOFX?

Use NOFX for a controlled evaluation if you want AI-generated trading plans or a continuously running model strategy and can verify execution yourself. Start in a paper-trading or testnet environment where the chosen integration supports it, as the project's disclaimer recommends. Our evidence supports evaluating the software; it does not support entrusting it with unattended capital.

The distinction between its two modes matters before you connect an account. Frontier requires approval for each trade, while Autopilot can place trades while running. The current README says generating a Frontier plan does not submit an order, and starting or stopping Autopilot is a separate action. Choose the mode deliberately rather than treating every AI interaction as a trade preview.

What our sandbox actually recorded

Our run took place on September 27, 2026, at commit 638d404. Installation took 67 seconds, the build step took 119 seconds and the test step took 119 seconds. All three command outcomes have exit code 0 in the attached harness record.

The stored parser summaries record 48 passing Go package results and 0 failing package results. The raw measurement identifies these as packagesOk and packagesFail, and separately records 20 packages without test files. Repeated package lines in the retained output are not additional results. These summaries do not establish the number of unique individual tests or exhaustive test coverage.

Our harness records command success, exit status, elapsed time, timeout status and parsed output separately. The build tail ends with vet-exit:0; the test tail contains Go package results. Neither establishes a completed trading workload. Browser onboarding, funded model calls, live exchange execution, protective-order placement and restart recovery were not established by this run. The audit field is null, and dependency scan coverage is unknown.

The MrKeyoor testing methodology explains how command outcomes reach our reviews and why they do not measure test quality or security. That page describes the publisher's current procedure; it does not establish which methodology revision applied to this September run. This correction uses the stored record and includes no new execution.

What the current execution documentation promises

The current NOFX README describes Frontier as a workspace for comparing entries, reviewing account and order details, and approving execution. It currently exposes Binance USDT perpetuals, Hyperliquid, OKX and Bitget. The broader runtime includes additional exchanges, so runtime support should not be read as identical Frontier support.

The README describes shared exchange adapters, runtime limits and reconciliation of uncertain submissions. It also states that entry and protection are not atomic. A filled entry can therefore require attention if protective-order placement fails. Review the exchange's actual orders before allowing further entries, and establish how you will close a position if the application becomes unavailable.

These are documented capabilities and limitations, not observations from our sandbox. The current README includes Frontier workflows that our older run did not validate. Likewise, the README at the measured commit describes an Autopilot-focused terminal; current documentation should not be treated as a description of everything present at that commit.

Deployment adds work beyond installation

The current installer uses published Docker images and generates missing environment keys. For remote deployments, the README says production Compose publishes ports on all host interfaces by default. Configure host bindings and an authenticated HTTPS proxy before exposing an instance.

Local storage includes encrypted exchange credentials and signing material. Protect the installation files and backups as well as access to the application. AI fees, paid-data fees and exchange trading funds are separate; the README warns that model requests can incur charges even when no trade is placed.

Before upgrading, stop active Autopilot traders and review running Frontier executions. Preserve configuration, encryption material and saved state. After restarting, compare positions and orders with the exchange and reconcile executions needing attention. The README explicitly warns that stopping a process does not close its positions.

When another framework fits better

Freqtrade documents dry-run trading and backtesting of buy/sell strategies. Choose Freqtrade instead when your immediate task is developing a crypto strategy and evaluating it through those workflows. Its README also includes machine-learning tools, so describing it as exclusively deterministic would be misleading.

Jesse is a Python framework for researching and defining your own strategies for backtesting, optimization and live trading. Choose Jesse instead when owning the strategy implementation and conducting Python research are the main requirements. Its AI-assistant integration does not make it equivalent to NOFX's continuously running model-directed trader.

These are alternative trading frameworks, not documented safety add-ons for NOFX. We did not run either as part of this comparison and cannot rank their execution reliability from these extracts.

How to read the ratings

The ratings assess the evidence available here, not profitability or investment safety. The rubric is 1 for major demonstrated obstacles, 2 for substantial unresolved operational limits, 3 for a workable starting point requiring operator work, 4 for clear documentation across the relevant tasks, and 5 for unusually complete corroborated evidence within the dimension.

Setup ease is 3 because our commands completed, while deployment and trading setup remain separate work. Documentation is 4 because the current README explains execution modes, installation, remote exposure, fees and recovery limits. Community is unassessed because the supplied primary extracts do not establish response times, contributor continuity or issue resolution. Maturity is 2 because entry and protection are documented as non-atomic and our run does not validate trading or recovery.

Alternatives

ProjectWhat it isPick it when
Freqtrade gh↗A Python crypto trading bot with dry-run trading, backtesting and strategy optimization.Choose it instead when developing and evaluating a buy/sell strategy through dry runs and historical simulations is your main task.
JesseA Python framework for defining strategies, backtesting, optimization and live trading.Choose it instead when you want to own the strategy implementation and use Python research workflows.

Sources

  1. NOFX repository and README
  2. Measured NOFX commit 638d404
  3. NOFX disclaimer
  4. NOFX security policy
  5. NOFX production Compose file
  6. Aster duplicate stop-order report
  7. Backtest and live timing mismatch report
  8. OKX equity calculation pull request

More ai tools reviews