VegaLoad
Star Install

Open source · Sponsored by Harness

Load testing
for agentic
developers.

Ask your agent for a load test. It writes the file next to your code and runs it with VegaLoad.

Install VegaLoad View on GitHub
Illustration. Numbers are examples.

01 · The Agentic Workflow

How it changes load testing.

From fragmented external tooling to a test file living right beside your service code.

1 / 3

The Shift

Move to AI-native load testing.

Today the work is spread across a separate tool, scripts in another repo, and a ticket queue. VegaLoad brings it into one file next to your code. Your agent writes it and runs it.

Today

  • A A separate load tool
  • B Scripts in another repo
  • C A ticket in someone's queue

With VegaLoad

  • checkout.js
  • checkout.vl.js

One file next to the code. Your agent writes it, reviews it, and runs it.

Co-location

Write load tests alongside your code.

A test is a plain JavaScript, TypeScript, or Python file. It lives beside the service it tests. Reviewed in the same pull request. Checked in CI before the merge, and in CD after the deploy.

Same repo Same pull request One report per run
  • checkout-service
  • src
  • checkout
  • checkout.js
  • checkout.vl.js
  • cart
  • cart.js

Fast Onboarding

Set up once. Ask in plain words.

Install VegaLoad. Run one command in your project. Your agent then knows how to generate realistic scenarios, validate thresholds, and run tests locally.

Cursor & Claude Code Bundled MCP Server Runs on your machine
  • brew install vegaload/tap/vegaload
  • vegaload init
  • Skills for your agent added to this project
  • MCP server configured for Cursor and Claude

Zero cloud dependencies. Completely private and local.

02 · Engine & Platform

Built for production. Safe by default.

Protocols, baseline gates, strict host allowlists, and a clear path from local runs to enterprise scale.

1 / 4

14 protocols

HTTP/1.1, HTTP/2, gRPC, WebSocket, MQTT, Kafka, PostgreSQL, MySQL, Redis, RabbitMQ, FTP, SFTP, and raw TCP and UDP. All in one binary.

  • HTTP/1.1
  • HTTP/2
  • gRPC
  • WebSocket
  • MQTT
  • Kafka
  • PostgreSQL
  • MySQL
  • Redis
  • RabbitMQ
  • FTP
  • SFTP
  • TCP
  • UDP

Pass or fail gates

Set limits for p95 latency, error rate, and checks. A broken limit exits 3. It stops the pull request in CI, or the rollout in CD.

Baseline comparison

Save a good run. Later runs automatically detect latency or error rate regressions.

Results where your pipeline looks

JUnit XML for your CI or CD system. A rich job summary on the GitHub Actions page.

Start from what you have

Build a scenario from an OpenAPI spec or a browser HAR recording with auto-correlation.

Single static binary

Zero JVM, zero Node runtime. Native on macOS, Linux, Windows, Docker, and Kubernetes.

Safety Guardrails

Your agent can only test hosts you approve.

Localhost is always allowed. Any other target host needs your approval. Every run is written to an audit log on your machine, with strict caps on concurrency and duration.

Host allow-list Audit log (~/.vegaload) RPS & duration caps
  • vegaload run -target https://staging.example.com/health -protocol http1
  • Target "https://staging.example.com/health" is not localhost or allowlisted. Continue? [y/N]
  • vegaload run: refusing to run against "https://staging.example.com/health" without confirmation (pass -yes, or add it with -allow-target)
  • ~/.vegaload/audit.log, one line per run
  • {"trigger":"mcp","scenario":"checkout.vl.js","vus":10,"outcome":"success"}

Visual Artifact

One HTML file. It opens offline.

Latency percentiles, requests per second over time, and step metrics for that run. No dashboard server to operate, no telemetry leaving your machine. Just double-click to view or attach to a PR.

Zero server to host Renders completely offline Self-contained charts
report.html Real run · 3,616 reqs
VegaLoad HTML report preview showing latency percentiles and threshold verdicts

Migration On-Ramp · Coming soon

Bring your existing tests

Soon you can import your JMeter plans, and then your k6 scripts, into VegaLoad scenario files. You get a converted script and a report of what needs a manual fix. Your agent can finish the rest.

Planned command. Not released yet. $ vegaload import jmeter checkout.jmx
  • JMeterPlanned: thread groups, requests and timers
  • ReportPlanned: every element listed, nothing dropped silently

Enterprise Scale

Grow into Harness RT

VegaLoad runs on your laptop, in CI and in CD. When you need distributed cluster load, shared history, and chaos resilience probes, Harness Resilience Testing provides the enterprise continuum.

$ vegaload --move-to-harness checkout

One command moves a test to Harness. Your local test and report stay as they are. Coming soon.

Start

One command. Then ask your agent.

Install the single binary on macOS, Linux, or Windows. Then run vegaload init in any repo.

brew install vegaload/tap/vegaload

Then run vegaload init in your project to set up your agent.

All downloads Cookbook recipes