Skip to content

Releases

v4.0.0

Choose a tag to compare

@QAInsights QAInsights released this 04 Nov 23:52

Feather Wand - JMeter AI Agent v4.0.0

What's new

  • Workload Designer. A new dialog to build workload models visually: deterministic JMeter core workload shapes, AI workload suggestions, non-HTTP sampler support, distributed generator capacity checks, shell-specific run commands, a toolbar and Run action, field validation with themed feedback, and a remembered window size.
  • Workload Preflight. Plan-aware checks before you run warns about blank CSV filenames for remote workloads and flags out-of-bounds values.
  • Make it pass. A guarded agent workflow that runs your test plan, observes failures, and iterates fixes in rounds capturing plan diffs, adding result listeners and deterministic page assertions, stopping immediately on Stop, and showing the report inside JMeter.
  • OpenRouter provider. Available for chat and Agent Mode, with featured models first and reasoning-model handling.
  • OpenCode provider. New subscription CLI provider with a setup flow, install hint when the CLI is missing, and configured model ordering.
  • Connect a provider. A single setup dialog for all providers with live reload, provider logos, saved-key reuse, and Smart routing + Apple on-device panes.
  • Contextual chat starters. Empty chats offer up to three contextual starter cards above the composer, scoped to CSV data where applicable.
  • Slim jar. The plugin jar no longer bundles the ONNX runtime it is downloaded on demand for the current platform, verified by hash, with progress and failure reasons shown in the route card.
  • Broader anonymous telemetry. More privacy-safe feature counters with bounded payloads; the telemetry worker now forwards daily pings to PostHog with retries.

Improvements and fixes

  • IntelliSense. Fixed Shift+Enter handling, caret placement, active-token replacement, the prompt picker flow for bare prompts, and command Tab/Enter completion.
  • Ask AI. Disabled while a test is running; Workload Designer supports zoom.
  • Model picker. Faster filtering and first-fill sizing; the OpenAI list now includes o-series and chat-latest models while hiding image and Responses-only models.
  • Recording hardening. Origin checks restricted to http(s); orphaned Playwright MCP processes are swept after the launcher exits.
  • Routing and logs. Download URLs and embedded userinfo redacted from logs and route-card reasons; OpenAI Decisions routing failures now surface in the route card.
  • Miscellaneous. Duration limit errors list all remaining units; slimmer pet sprites; the welcome screen treats an available Apple on-device model as a ready provider; timing-flaky tests hardened.

v3.9.0

Choose a tag to compare

@QAInsights QAInsights released this 23 Oct 03:13

Feather Wand - JMeter AI Agent v3.9.0

What's new

  • Apple Foundation Models on-device provider. New provider chats with Apple's on-device Foundation Models framework — fully local, no API key or cloud required. Agent mode is disabled while the Apple on-device provider is selected.
  • Quick-start properties file. A short jmeter-ai-quickstart.properties gets first-time users running quickly: key-only quick start enables JSR223 refactor, shows the plain quick-start URL, and now offers a Gemini option (and notes Ollama needs no key).

Improvements and fixes

  • Apple provider hardening. Verifies the native helper binary hash, reprobes transient status, surfaces context-too-long guidance in the chat error, and no longer retries after partial output.
  • Telemetry accuracy. Agent tool calls with a missing required parameter or an unknown tool name now count as tool failures, declined confirmations count as outcome_tool_declined, per-tool failure counters (toolfail_<name>) are reported and shown on the stats page, and long tool names match correctly when computing failure share.
  • Header buttons. Record and Donate buttons now size to fit their text.

v3.8.9

Choose a tag to compare

@QAInsights QAInsights released this 22 Oct 03:05

Feather Wand - JMeter AI Agent v3.8.9

What's new

  • Feather Wand as an MCP Server. Run Feather Wand as a Model Context Protocol (MCP) server so external AI tools (such as Claude Code, Cursor, and OpenAI Codex) can inspect, navigate, and edit active JMeter test plans via standard tool calls.
  • Local Smart Routing with ONNX Runtime & EmbeddingGemma. Zero-dependency local embedding-based routing runs directly inside the plugin using ONNX Runtime and EmbeddingGemma without requiring Ollama or external API keys. Smart routing with Ollama embeddings is also supported.
  • Agent Mode enabled by default & Stop Button. Agent Mode is now pre-selected and enabled by default (jmeter.ai.agent.enabled=false keeps chat only). A dedicated Stop button halts Agent Mode runs safely between turns and tool calls.
  • "Extract this" in View Results Tree. Right-click selections in View Results Tree response panes (and JSyntaxTextArea popups) to automatically construct and configure Regular Expression, JSON Path, or Boundary Extractors directly under the target sampler.
  • Natural language @sla command & Triage Severity. The @sla command parses requirements to generate response time and error rate assertions. Failure triage now includes severity scoring and confidence indicators.

Improvements and fixes

  • Dependency Isolation & Netty Relocation. Bundled Netty classes are shaded and relocated to avoid runtime clashes with older JMeter plugin JARs.
  • Window & Panel Docking Stability. Safely undock nested Feather Wand and AI CLI panels with divider position preservation.
  • Chat & History Reliability. Fixed user-turn pairing in chat history on stream failures or cancellations, preventing flipped roles; conversational element lists cleanly nest into target Thread Groups or Controllers.
  • Test Plan CLI Context. Improved test plan context handling and process scoping for external AI CLIs.

v3.8.8

Choose a tag to compare

@QAInsights QAInsights released this 12 Oct 00:38
d4380b3

Feather Wand - JMeter AI Agent v3.8.8

What's new

  • New Decisions backends for smart routing and failure triage. jmeter.ai.judgment.provider now accepts openai (OpenAI Decisions API — uses openai.api.key and openai.decisions.model, default gpt-6-luna) and openrouter (OpenRouter Decisions API — uses openrouter.api.key, default microsoft/microsoft-decision-1), alongside the default typesafe (Jev). OpenRouter cards name the model vendor, e.g. "Microsoft Decision".
  • Actionable provider errors. Agent Mode surfaces the provider's real error message instead of a generic apology.
  • Richer anonymous telemetry. The daily anonymous ping now also counts agent outcomes, tool calls, routing/triage judgments and provider errors, and reports the configured Decisions backend and well-known model. The telemetry worker adds retention stats, version adoption, Java/arch/agent/model/Decisions breakdowns and grouped feature tables.

Improvements and fixes

  • Jev stays silent when typesafe.api.key is missing or still the placeholder — no more "Jev unavailable" Smart Route card on every Agent Mode reply.
  • CSV import rejects invalid shareMode values; chat questions are no longer fed to the element parser.
  • Triage advisories sent to the model are labeled with the active backend; the agent logs why routing/triage judgments are unavailable.

Build, docs, and CI

  • run.sh/run.bat resolve JMETER_HOME and the pom version instead of hardcoded paths.
  • Documented previously-missing config properties (DeepSeek, Jev, terminal, correlation, system properties).
  • CI builds and tests PRs, runs Swing tests under xvfb with CJK fonts, and skips chime playback tests when no audio device is present.

v3.8.6

Choose a tag to compare

@QAInsights QAInsights released this 30 Sep 15:40
6f4dc87

Feather Wand - JMeter AI Agent v3.8.6

What's new

  • Anonymous usage telemetry (on by default, opt-out). Feather Wand sends one anonymous usage ping per UTC day — a random install id, plugin/JMeter/Java versions, OS name and CPU count, plus feature counters. No prompts, generated content, or test-plan data is collected. Backend is a Cloudflare Workers + D1 service (telemetry-worker/).

Improvements and fixes

  • Opt out any time via jmeter.ai.telemetry.enabled=false in jmeter-ai.properties or the FEATHER_WAND_TELEMETRY=0 environment variable — see the README "Privacy and telemetry" section and jmeter-ai-sample.properties.
  • Extensive new test coverage: agent-mode dispatch and error paths, recording workflow, correlation, element manager, test-plan reload, think-time clamping, optimize handler, subscription CLI failure paths, and provider streaming via local HTTP fixtures.

v3.8.5

Choose a tag to compare

@QAInsights QAInsights released this 19 Sep 02:44

Feather Wand - JMeter AI Agent v3.8.5

What's new

  • Optional Jev smart routing (TypeSafe). Agent Mode requests can be classified by TypeSafe's Jev model into typed routes that select a focused tool pack, with a visible Jev Smart Route card in the transcript and safe fallback to the full tool set when Jev is uncertain or unavailable.
  • Mid-run tool expansion. Focused packs expose an expand_tools escape hatch: Jev re-classifies the request with the model's stated need and grows the live tool set (re-advertised to the provider on the next request), capped per run, never beyond the full registry. Jev Expanded Tools card shows what was added.
  • Jev failure triage. get_test_results failures are classified into root-cause buckets (connection, timeout, 4xx/5xx, auth/session, assertion, script, config) — deduplicated by signature and capped per run — with a Jev Failure Triage card and an advisory appended to the tool result.

Improvements and fixes

  • Jev is advisory only: it never generates answers, selects tool arguments, executes tools, or bypasses destructive-operation confirmation. All Jev features default off behind jmeter.ai.typesafe.* flags — see README and jmeter-ai-sample.properties.
  • Extracted AgentCommandRunner from CommandDispatcher; transcript cards for Jev decisions render provider-first model names without truncation.

v3.8.4

Choose a tag to compare

@QAInsights QAInsights released this 17 Sep 01:35

Feather Wand - JMeter AI Agent v3.8.4

What's new

  • Per-turn token usage logging. The agent now logs token usage for each turn so you can track consumption.

Improvements and fixes

  • 429 quota handling. Detects Gemini 429 responses, surfaces clear quota-exceeded explanations in plain-chat streaming and worker error paths, and guards against cause cycles.
  • Configurable SDK retries for AI providers.
  • Added a GUI-testing skill for exercising Feather Wand against a local gateway stub.

Install

Drop jmeter-agent-3.8.4.jar into <JMETER_HOME>/lib/ext and restart JMeter, or install via Plugins Manager.

Full Changelog: v3.8.3...v3.8.4

v3.8.3

Choose a tag to compare

@QAInsights QAInsights released this 14 Sep 00:54

Feather Wand - JMeter AI Agent v3.8.3

What's new

  • Agent Mode for DeepSeek, Grok, and Meta Muse. Agent mode now works with these providers (Meta Muse agent runs use Chat Completions).
  • Higher iteration limits. Default agent max iterations raised to 16; agent and recording iteration caps raised to 500, with the recording cap now configurable.

Improvements and fixes

  • Agent rejects unknown property keys and defaults HTTP samplers to GET.
  • Refreshed the vendored model-capability catalog.
  • Documented corporate LLM gateway configuration and troubleshooting.
  • Fixed the test suite on newer JDKs (Mockito 5.18, JaCoCo 0.8.13).

Install

Drop jmeter-agent-3.8.3.jar into <JMETER_HOME>/lib/ext and restart JMeter, or install via Plugins Manager.

Full Changelog: v3.8.2...v3.8.3

v3.8.2

Choose a tag to compare

@QAInsights QAInsights released this 02 Sep 02:31
105bce5

Feather Wand - JMeter AI Agent v3.8.2

What's new

  • Explicit Chat / Agent mode selector. The composer now defaults to safe Chat mode and only permits agentic JMeter tool calls after the user explicitly selects Agent mode.
  • OpenAI and Anthropic gateway support. Route requests through custom OpenAI-compatible or Anthropic-compatible base URLs, provide gateway-specific headers, and configure model lists for gateways without model-discovery endpoints.

Improvements and fixes

  • Added support for header-only gateway authentication while preserving API-key authentication.
  • Added plaintext HTTP warnings for gateway URLs that could expose prompts or credentials.
  • Improved gateway model discovery and filtering for non-OpenAI model IDs.
  • Refreshed the vendored model-capability catalog and its Linux-based update workflow.
  • Updated provider setup guidance and screenshots.

Install

Drop jmeter-agent-3.8.2.jar into <JMETER_HOME>/lib/ext and restart JMeter, or install via Plugins Manager.

Full Changelog: v3.8.1...v3.8.2

v3.8.1

Choose a tag to compare

@QAInsights QAInsights released this 26 Aug 02:57

Feather Wand - JMeter AI Agent v3.8.1

What's new

  • ChatGPT / Codex and Claude Code providers. Use the authentication already managed by the local CLI for chat and Agent Mode, with sign-in status/actions in the model picker, configurable or custom model ids, bounded history replay, cancellation, and no plugin-managed CLI credentials.
  • Modern chat and model-picker UI. Search by model or provider, inspect context/cost/capability metadata, pin favorites, reuse recent models, and use the responsive composer with integrated reasoning controls, attachments, context status, and Send/Stop actions.
  • New Feather Wand branding. Dedicated chat and AI CLI terminal marks now appear across the README, Run menu, JMeter toolbar, chat/terminal headers, and assistant messages.

Improvements and fixes

  • Hardened CLI request cancellation, timeout, process cleanup, and concurrent request handling.
  • Added persistent custom subscription model ids without requiring a properties edit or restart.
  • Fixed picker-footer layout, custom-model dialog ownership, and duplicate failed-CLI calls.
  • Improved selected-model readability and expanded README/sample-property documentation for CLI providers and the redesigned UI.

Install

Drop jmeter-agent-3.8.1.jar into <JMETER_HOME>/lib/ext and restart JMeter, or install via Plugins Manager.

Full Changelog: v3.8.0...v3.8.1