Pricing

You control
the LLM costs.

Unlike cloud testing platforms with opaque per-execution pricing, Karate Agent's costs are transparent: you pay for your LLM tokens — or nothing, if you use scripted flows.

Token budgets are enforced per-job via maxIterations. The dashboard shows input/output token counts per session.

No per-execution fees No per-seat licensing Self-hosted
report — j_8c3b1d
$ karate-agent report j_8c3b1d ──────────────────────────────────── Status completed Iterations 8 of 20 max LLM calls 1 (recovery only) Duration 18s total ──────────────────────────────────── Tokens in 12,482 Tokens out 1,203 Cache hit 85% ──────────────────────────────────── COST ~$0.02 ↑ on your bill

What You Run

Four ways to run. Four cost profiles.

Pay nothing for scripted flows. Pay pennies for flows + LLM recovery. Pay a bit more for prompt-only exploration. You choose per job.

Full Flows

JS scripts via Flow.run(), native speed

$0
Guided + Flows

JS fills forms, LLM verifies

~$0.01–0.05
per run
Hints

Plain English .md task file, no code

~$0.05–0.15
per run
Explore

Just a prompt, LLM figures it out

~$0.15–0.50
per run

Ranges reflect typical enterprise web workflows. Your cost depends on page size, model choice, and how many steps need LLM recovery.

Teams graduate to the cheap tiers fast.

Start in Explore to build coverage. As each flow gets codified, cost drops. Most production regression suites run in Guided + Flows (~$0.02/run) or Full Flows ($0/run). A 12-step Guidewire submission went from $0.50 to $0 once flows were in place.

See the Guidewire case study

Cost Controls

Four levers that keep costs near zero

Every mechanism is documented, configurable, and visible in the dashboard. No surprises on the token bill.

Flows consume zero tokens

Scripted .js flows run at native JavaScript speed via Flow.run(). Zero LLM calls for known paths.

Per-job token budget

maxIterations caps LLM calls per job. The agent stops when it hits the ceiling — your budget is never exceeded.

High cache hit rate

Prompt caching gives a high hit rate on the system prompt — billed at a fraction of regular input cost on supported providers.

Self-hosted LLM = $0 API

Run Gemma 26B (4B active, MoE) on one consumer GPU via Ollama. Passes all page-automation benchmarks. Zero cloud API cost.

Pricing Model

Karate Agent vs cloud testing SaaS

The usual model bundles infrastructure, per-seat licensing, and per-execution fees into a single opaque bill. Karate Agent unbundles it.

Cloud testing SaaS

  • Opaque per-execution pricing
  • Per-seat licensing for dashboards
  • Data leaves your network
  • LLM vendor locked in
  • Cost scales with test count

Karate Agent

  • Transparent: token counts per session
  • Shared dashboard — no per-seat fees
  • 100% self-hosted, air-gap ready
  • BYO LLM — Claude, GPT, Gemini, Ollama
  • Cost scales with LLM usage, not test count

Enterprise Ready

What every deployment ships with

No feature gating. No premium tier. No paid add-ons. The full platform — both modes, the flow system, every integration — is in one jar and one Docker image.

Self-Hosted Deployment

1 jar. 1 image.

Your infra · Your data · Your rules

Runs where your data already lives

No SaaS dependency, no telemetry, no data egress. Pair with Ollama for a fully air-gapped stack — the LLM, the browser, and the grid all run on your hardware.

Self-hosted Air-gap ready Docker-native
Token-Only Pricing

$0 platform fee

Token bill only · no markup

No per-run, no per-seat, no surprises

Your LLM costs are your own — billed direct by Anthropic, OpenAI, or whichever provider. Scripted flows cost $0 per run. Dashboards are shared — no per-seat licensing.

No execution fees No per-seat Flows = $0
Both Modes Platform

Interactive + Autonomous

Both modes · Flow system · Shared library

The full agent, not a crippled tier

Explore live in the dashboard, then submit unattended jobs. Flows compose into orchestrators, stored in git, shared across teams. Every feature is in the free build.

Interactive Autonomous Flow.run()
Audit-Ready Governance

SSO + Audit + CI

REST · MCP · RBAC · session trails

Slots into the stack you already have

Standard REST API and MCP server. SSO against your IdP, role-based access, full audit trails with H.264 session recordings. A single curl triggers jobs from any CI pipeline.

SSO + RBAC REST + MCP Audit trail

Let's talk

Tell us about your team and testing needs. We'll find the right fit.