AI-POWERED INTELLIGENCE FOR APACHE JMETER

Feather Wand

AI inside JMeter. Workload design coming in 4.0.

Feather Wand brings AI chat, embedded AI CLIs, and Agent Mode into Apache JMeter. Read and edit your live test plan, correlate dynamic values, and refactor JSR223 scripts without switching tools. The upcoming 4.0 release adds Workload Designer, Workload Preflight, and plan-aware chat starters: design your workload, check your plan, and prepare commands for local or distributed execution.

900+ INSTALLS

Trusted by 900+ performance engineering teams worldwide

Installed via JMeter Plugins Manager across CI/CD pipelines, QA labs, and production environments.

Install now →

AI AGENT OPEN SOURCE · MIT AGENT MODE 9 LLM PROVIDERS

Intelligent Features

Supercharge your performance testing workflow with AI-driven capabilities built natively into JMeter.

Upcoming in 4.0

Design your workload. Check your plan. Run it where you need.

Feather Wand started with chat, brought AI CLIs into JMeter, then added Agent Mode to act on your test plan. The next release brings workload design into that same workflow.

Workload Designer

Start with a sample or message rate, iteration rate, or concurrent user target. Preview load, endurance, step, spike, stress, or custom shapes; estimate threads and load-generator needs; then apply the workload to your plan. Copy a local or distributed JMeter command for Bash, cmd, or PowerShell. Validate sizing estimates with a calibration run.

Workload Preflight

Check the open plan for heavy listeners, missing local CSV files, risky EOF settings, and inherited timers that may affect the workload. Each finding explains why it matters and lets you select the affected element. Remote files and generator settings remain Not verified. Preflight does not execute a test or call AI.

Plan-aware chat starters

Once a provider is connected, an empty chat offers three suggestions based on the open plan and selected Thread Group, directly above the composer. Create a group, add a request or CSV data, review response checks, or plan the workload. Clicking fills an editable draft without sending, overwriting existing text, or changing the plan. Suggestions return on New Chat.

Core workload calculations and suggestion generation run locally without an AI call. Ask AI is optional; sending a chat draft uses your chosen provider. These features are coming in 4.0, not the current published release.

🤖

Agent Mode

AI autonomously edits your live test plan through 21 tools: add elements, set properties, run tests, and auto-correlate dynamic values, with confirmation-gated destructive ops. Works with Claude, OpenAI, Gemini, DeepSeek, Grok, and Meta Muse.

🔗

9 LLM Providers

Chat with Claude, OpenAI, Gemini, DeepSeek, Ollama, Grok (xAI), Meta Muse, AWS Bedrock, or Apple Foundation Models on-device. Apple requires macOS 26+ on Apple silicon, supports chat only, and needs no API key.

🔍

Context-Aware @Commands

@this, @testplan, @optimize, @lint, @wrap, @code, and @usage. Each command understands your currently selected test-plan element.

💻

Multi-AI CLI Terminal

Run Claude Code, OpenAI Codex, OpenCode, Antigravity, or Grok CLI in an embedded terminal that receives your open .jmx as context.

⚡

Real-Time Streaming

Watch AI responses appear token-by-token, with a Stop button to cancel mid-response and an optional chime when the AI finishes.

🧹

Smart JSR223 Refactoring

Right-click in the JSR223 editor to refactor, format, or inject functions with AI: no copy-pasting between tools.

🔒

Local & Private via Ollama

Run fully offline with local models: zero data egress, with configurable thinking modes for reasoning models like DeepSeek R1.

🐾

Companion Pets

A draggable animated pet (quill, glim, peacock, or monkey) that cheers on clean runs and frowns on sampler failures.

🔥 Killer Feature

Agent Mode | AI with Tool Calling

Feather Wand doesn't just chat about your test plan, it acts on it. The agent reads your tree, calls tools to mutate elements, verifies each change, and iterates until the task is done. Every tool call streams to the chat in real time.

✓ Anthropic Claude ✓ OpenAI ✓ Google Gemini ✓ DeepSeek ✓ Grok ✓ Meta Muse On by default · opt out with jmeter.ai.agent.enabled=false

Tool calling works with Anthropic Claude, OpenAI, Google Gemini, DeepSeek, Grok, and Meta Muse models, plus the ChatGPT/Codex and Claude Code CLI providers. Ollama, AWS Bedrock, and Apple on-device fall back to plain chat.

Read 4 tools

get_tree_state : Full test-plan tree with names, types, enabled state
get_element_config : All properties of a specific element
get_element_children : Children of a specific element
get_element_schema : Property schema and allowed values for an element type

Write 10 tools

add_element : Add a new element as a child of a parent
update_element_property : Set a scalar property on an element
set_property_list : Set a flat string-list property
set_structured_property_list : Set structured lists like headers and arguments
delete_element : Delete an element and its subtree (confirmation gated)
toggle_element : Enable or disable an element
move_element : Reparent an element (confirmation gated)
duplicate_element : Deep-clone an element's subtree
rename_element : Rename an element
reorder_element : Reposition an element among its siblings

Run 3 tools

run_test : Start the test plan
stop_test : Stop the running test
get_test_results : Run in a private engine and report pass/fail with failure details

Correlation 2 tools

find_correlation_candidates : Probe the plan and detect dynamic values needing correlation
apply_correlation : Add extractors and rewrite values to ${variable} (confirmation gated)

File 2 tools

save_plan : Save the test plan to a .jmx file
open_plan : Open a .jmx file (confirmation gated)

Safety Gates

Destructive operations (delete, move, open plan, apply correlation) ask for confirmation before executing.
Bounded iterations: the agent stops after 8 reason-act cycles by default, configurable via jmeter.ai.agent.max.iterations.
Graceful degradation: if the agent loop fails, you get a plain-text answer describing what it attempted.
Every mutation is undoable with Ctrl+Z when undo.history.size > 0 in user.properties.

See it work: natural language to tool calls

"Add an HTTP Request under the Thread Group and set its path to /login"

get_tree_state add_element update_element_property get_element_config

"Add a Response Assertion that checks for 200"

get_tree_state add_element set_property_list

"Find dynamic values that need correlation"

find_correlation_candidates

"Apply correlation for candidates 1 and 3"

apply_correlation

"Run the test and tell me if it passed"

run_test get_test_results

"Move the JSON Extractor under the first HTTP Request"

get_tree_state move_element

⚔️ Head to Head

Feather Wand vs AI Coding CLIs

Codex and Claude Code are general-purpose coding agents that edit files. Feather Wand is purpose-built to drive Apache JMeter from the inside, and it even ships both CLIs in an embedded terminal. Here is how they stack up.

Capability	Feather Wand	Codex	Claude Code
Runs inside JMeter	Yes, native plugin	No (available via Feather Wand's embedded terminal)	No (available via Feather Wand's embedded terminal)
JMeter-aware correlation	Yes, purpose-built tool call	General reasoning only	General reasoning only
Undo	Ctrl+Z, native JMeter undo stack	Git / manual backup	Git / manual backup
Autonomous test plan edits	Agent Mode, 21 tools (add, update, delete, move, run, correlate)	Yes, general file edits	Yes, general file edits
Model support	Claude, OpenAI, Gemini, DeepSeek, Ollama, Grok, Meta, Bedrock, Apple Foundation Models	GPT	Claude models
Subagents / parallel orchestration	Not built in	Yes	Yes
Setup	Install plugin, add your API key	Install CLI, authenticate	Install CLI, authenticate
Cost	Free and open source (MIT), bring your own API key	Subscription or API usage	Subscription or API usage
Best for	Fast, scoped JMeter edits and correlation without leaving the GUI	Deep reasoning, parallel subagent tasks across a repo	Deep reasoning across a whole codebase

Feather Wand is not a replacement for a general coding agent: it embeds Codex, Claude Code, and other CLIs so you can reach for them when the job outgrows the JMeter GUI.

Installation

Get Feather Wand up and running in minutes.

Plugins Manager (Recommended)

Open JMeter → Options → Plugins Manager → Available Plugins, search "Feather Wand", select it, then Apply Changes and Restart JMeter.

# JMeter → Options → Plugins Manager → Available Plugins
# Search "Feather Wand" → Apply Changes and Restart JMeter

Plugins Manager CLI

Headless and CI-friendly install via the Plugins Manager command line.

# Linux / macOS
bin/PluginsManagerCMD.sh install feather-wand-jmeter-ai-agent

# Windows
bin\PluginsManagerCMD.bat install feather-wand-jmeter-ai-agent

Manual JAR Install

Download the latest JAR from the GitHub Releases page, copy it into JMeter's lib/ext directory, and restart JMeter. Feather Wand appears in the Tools menu.

# Download from https://github.com/QAInsights/jmeter-ai/releases
cp <downloaded>.jar $JMETER_HOME/lib/ext/
$JMETER_HOME/bin/jmeter

Configure AI Provider

Append jmeter-ai-quickstart.properties to your user.properties so existing settings are preserved. Uncomment one provider and add its key, then restart JMeter. Ollama and Apple on-device need no key.

❓ FAQ

Frequently Asked Questions

What is coming in Feather Wand 4.0?

The upcoming release adds Workload Designer for load targets, shapes, thread and generator sizing, and local or distributed run commands; Workload Preflight for plan-specific checks; and three contextual chat starters above the composer. Core calculations, Preflight, and suggestion generation do not require an AI call. Installation links still point to the latest published release, not an unreleased 4.0 build.

Does Workload Preflight prove my plan is ready to run remotely?

No. In the upcoming 4.0 release, Preflight inspects the open plan and local CSV availability without executing a test. It flags issues such as heavy listeners, CSV EOF settings, and inherited timers, but marks remote files and generator settings as Not verified. Validate remote resources separately and calibrate your workload on the intended generators.

What is Feather Wand for Apache JMeter?

Feather Wand is a free, MIT-licensed AI agent plugin for JMeter. Chat with 9 LLM providers to generate and analyze test plans, use context-aware @commands, run embedded AI CLI terminals (Claude Code, OpenAI Codex, and more), or enable Agent Mode to let the AI autonomously edit your test plan.

Which LLMs does Feather Wand support?

Nine providers: Anthropic Claude, OpenAI, Google Gemini, DeepSeek, Ollama (local), Grok (xAI), Meta Muse, AWS Bedrock via the Converse API, and Apple Foundation Models on-device. Apple requires macOS 26+ on Apple silicon, supports chat only, and needs no API key.

What is Agent Mode in Feather Wand?

Agent Mode lets the AI autonomously edit your live test plan through 21 tools: reading the tree, adding and configuring elements, running tests, and detecting plus applying correlations. Destructive operations ask for confirmation, iterations are bounded, and everything is undoable with Ctrl+Z. It works with Claude, OpenAI, Gemini, DeepSeek, Grok, and Meta Muse models and is on by default; set jmeter.ai.agent.enabled=false to keep it off.

How do I install Feather Wand?

The easiest way is the JMeter Plugins Manager: Options → Plugins Manager → Available Plugins → search "Feather Wand" → Apply Changes and Restart JMeter. For CI, use the Plugins Manager CLI: PluginsManagerCMD install feather-wand-jmeter-ai-agent. Append jmeter-ai-quickstart.properties to user.properties so existing settings are preserved, then uncomment one provider and add its key. Ollama and Apple on-device need no key.

Is Feather Wand free?

Yes, completely. Feather Wand is open source under the MIT license, including Agent Mode, all 9 LLM providers, the CLI terminal, and every @command. You only pay for your own LLM API usage (or nothing at all with local Ollama models or Apple on-device).

Does my test data leave my machine?

Not if you don't want it to. Feather Wand supports fully local deployments via Ollama, so test plans and results can stay on your machine with zero data egress.

GitHub Repository

Explore the source, report issues, and contribute.

QAInsights/jmeter-ai

AI inside JMeter. Workload design coming in 4.0.

Java Stars Forks MIT License

Report an Issue

Found a bug? Let us know.

Discussions

Join the community conversation.

Pull Requests

Contribute to development.

Releases

Download the latest version.

Connect

Follow QAInsights for updates, tutorials, and community events.

GitHub

https://github.com/QAInsights

X

https://x.com/QAInsights

YouTube

https://youtube.com/@QAInsights

https://linkedin.com/in/naveenkumarn

Website

https://qainsights.com