How Karate Agent Works

Two modes, one platform. Interactive for exploration, Autonomous for execution.

Interactive Mode

Your LLM coding agent (Claude Code, Cursor, Copilot) sends JS commands via REST or MCP. The grid proxies them to a live browser session.

  • No LLM needed on the grid — your client-side agent drives
  • Exploratory testing, debugging, live demos
  • Discover locators, build flows interactively
  • Connect anything that can curl
Interactive Mode Demo Video coming soon

Autonomous Mode

A worker-side LLM drives the observe-decide-act loop inside the karate-agent container. Submit a job and walk away.

  • LLM configured on the worker — runs independently
  • CI/CD, scheduled tests, batch jobs
  • Flows run first (fast), LLM handles unknowns
  • Token budget enforced via maxIterations
Autonomous Mode Demo Video coming soon

The Development Workflow

1
Explore
LLM explores app, discovers locators
2
Create
LLM writes .js flow files
3
Test
Flow.run() → {ok: true}
4
Compose
Chain flows into orchestrators
5
Autonomy
Submit as job, LLM handles unknowns
6
Report
Review deviations, fix, repeat

Architecture

One jar, one Docker image. No microservices, no databases, no message queues.

Karate Agent grid architecture — grid server with REST API, proxy, and job runner managing isolated Docker worker containers

Scales Without Changing Anything

Solo Developer

java -jar veriquant.jar grid on your laptop. Test against localhost apps. Flows in a local directory.

Shared Team Server

Grid on a Mac Mini or EC2 instance. Dashboard URL shared with the team. Flows directory backed by git.

CI-Integrated

Grid on dedicated infrastructure. Jobs submitted by CI pipelines. Reports collected as artifacts.

Real-World Proof

Guidewire PolicyCenter

12-step Personal Auto submission — from login through quote creation.

12
Steps in workflow
~30s
Flow execution time
8-19m
Pure LLM time

Each step was developed as an independent flow via Interactive mode, then composed into a single orchestrator. The orchestrator runs all 12 flows sequentially at native JS speed. The LLM is invoked only if a step deviates from the expected path.

Guidewire PolicyCenter is a complex enterprise SPA where standard locator strategies fail. Karate Agent's cursor-pointer discovery handles the non-standard <div onclick> targets that Selenium and Playwright struggle with.

Progressive Adoption

Start with zero AI cost. Graduate as confidence grows.

Stage LLM Cost What You Get
1. Interactive Zero Developer drives browser via curl/Claude Code. Explores app, discovers locators. No AI tokens consumed.
2. Scripted Flows Zero Reusable .js flow files execute at native speed via Flow.run(). Deterministic, repeatable. A 12-step workflow costs $0 per run.
3. Autonomous + Flows Minimal Scripted flows first. LLM invoked only on failure — typically 2-4 recovery iterations at ~$0.02-0.05 per job.
4. Fully Autonomous Moderate LLM drives entire workflow. 15-25 iterations at ~$0.15-0.50 per job. Reserve for exploratory testing or new workflows.

Most teams stabilize at Stage 3: 90%+ of execution is scripted (free), LLM handles the remaining edge cases.

See the Workflow in Action

Let us walk you through Interactive mode, flows, and autonomous mode on your application.

Book a Demo