Kerno vs QA Wolf (2026)

The agent calls the customer's running endpoints, triggers background tasks through the customer's own message broker, introspects MCP servers and runs against real dependencies such as databases and caches. It works with coding agents over MCP, GitHub and CI. All of these actions reach the customer's own app under test. There is no connector, OAuth authorized action or custom tool that lets an agent act on other business systems.

SourceKerno, kerno.gitbook.io/docs security testing, environment setup and integrations, and kerno.io/llms.txtread 2026-09-21

Bug reports flow into Jira and Linear, where QA Wolf automatically creates, syncs and closes issue tickets, and results sync to test management tools (TestRail, Xray, Zephyr, Qase and Testmo). Deployments are heard from GitHub and GitLab, notices go to Slack and Microsoft Teams, and webhooks reach any other bug tracker. Its tests can call APIs, seed databases, toggle feature flags and use email inboxes. Writes into the customer's issue trackers mean access runs both ways, and webhooks and helper libraries extend it to other tools.

SourceQA Wolf, docs.qawolf.com Jira, Linear, test management integrations and webhooksread 2026-09-21

A fixed pipeline drives Kerno's agent, mixing deterministic steps with the agent's own planning. The agent indexes code, builds a Docker Compose environment, plans and writes scenarios, runs each one repeatedly with critique and repair until it passes, and updates the suite. These steps run as async jobs that can be polled or canceled. Customers do not define or reuse their own workflows.

SourceKerno, kerno.gitbook.io/docs how Kerno works, testing modes and Kerno MCPread 2026-09-21

Run Rules sequence the order of flows and their dependencies in a suite, and flows pass authentication tokens, user IDs and other values between each other in a coordinated run. Tags build custom suites, runs execute in parallel with automatic reruns of failures, and triggers choose which flows run on each deploy or schedule.

SourceQA Wolf, docs.qawolf.com Run Rules, pass data between flows, tags and triggers, and qawolf.com/pricingread 2026-09-21

A customer's coding agent starts Kerno's agent by calling it over MCP inside a coding session, and a GitHub Action replays the committed tests on every pull request without an account or agent. No event or schedule starts the agent on its own without someone's session asking for it.

SourceKerno, kerno.gitbook.io/docs Kerno MCP and running tests in CIread 2026-09-21

Triggers run chosen flows whenever an application deploys, reported through GitHub or GitLab deployments or a webhook, or on a schedule. Smart Smoke Suites automatically select the flows affected by each pull request, and the automation agent diagnoses broken tests and refactors test code after failures. Work starts on a deploy, a schedule or a pull request without a person asking.

SourceQA Wolf, docs.qawolf.com triggers, Smart Smoke Suites, and qawolf.com/llms.txtread 2026-09-21

Underneath the agent sits a deterministic index of the customer's codebase that resolves every function, class, model, endpoint, import and reference to a precise location. Kerno keeps it up to date as the code changes by recomputing only what a change affects, and queries it to find each change's blast radius and the endpoints to test. Custom rules add the team's own testing guidance. New code enters the index without retraining, and the agent reads it to decide what to test and to show which code a change touches.

SourceKerno, kerno.gitbook.io/docs codebase indexing, how Kerno works and custom rulesread 2026-09-21

Before it builds or explores, the agent reads documentation linked in a prompt, test plans, product requirements and help docs, and it applies team conventions written into custom skill files in every session. Company knowledge enters as linked pages and skill files the agent reads per session. There is no index, retrieval layer or refresh of the customer's sources.

SourceQA Wolf, docs.qawolf.com test automation, coverage mapping and custom skillsread 2026-09-21

What Kerno's agent learns from the code, from running tests and from the team's feedback goes into long term memory, such as how an endpoint authenticates or what must be seeded before a request succeeds. The agent reuses those lessons when planning later tests. People can review each lesson and agree, amend its wording or disagree so Kerno stops using it, and unchecked lessons are labeled as such. Each entry is stored as a git commit stamped with the code revision it was learned at, so stale analysis is derived again when code changes.

SourceKerno, kerno.gitbook.io/docs memory and learningread 2026-09-21

The coverage map, the test code and run history persist, and skill files the team writes apply in every agent session. There is no session, workflow or long term memory that the agent itself writes and reuses.

SourceQA Wolf, docs.qawolf.com custom skills and coverage mappingread 2026-09-21

Before Kerno's agent acts, it pauses for approval. It shows the full environment build plan and cannot start the environment until the plan is explicitly approved. It shows the scenario plan before writing any test and lets the person approve it, reject individual scenarios or give feedback.

It also opens feedback requests for plan approvals, planner questions or missing credentials, which are answered with an approval or a rejection with a reason. Generate runs always pause for approval, update runs pause only in interactive mode, and a plan approval setting chooses between gated and automatic runs.

SourceKerno, kerno.gitbook.io/docs testing modes, Kerno MCP feedback gate and changelogread 2026-09-21

Tests can be created unattended from a prompt, or the agent can work with a person who guides it, and a person can take over the browser during mapping while the agent watches. The generated tests are standard code the team reviews and versions. Choosing guided or unattended working sets an autonomy mode, but there is no approval step that holds the agent's action for a person.

SourceQA Wolf, docs.qawolf.com test automation and coverage mappingread 2026-09-21

The Enterprise plan adds SSO and SAML, audit logs, data controls and a custom data processing agreement. Kerno's portal gives organizations owner, admin and member roles, and API keys can be revoked or expire. A SOC 2 Type 2 audit is in progress.

SourceKerno, kerno.io pricing and homepage, and kerno.gitbook.io/docs organizations and teamsread 2026-09-21

QA Wolf is SOC 2 Type II compliant, verified through independent audits, with HIPAA ready infrastructure. Customers can set up single sign-on over SAML 2.0 and OpenID Connect, so people sign in through the customer's identity provider, and credentials are stored securely for use in tests.

SourceQA Wolf, docs.qawolf.com why QA Wolf (security and data protection) and single sign-onread 2026-09-21

An IDE extension shows what Kerno's agent is doing. Run reports upload to a portal that rolls up coverage, issues and runs across repositories, and job status and results can be read over MCP and HTTP. There is no step by step record of the agent's own prompts, tool calls and reasoning.

SourceKerno, kerno.gitbook.io/docs portal overview, reports and dashboards, IDE extension and Kerno MCPread 2026-09-21

Each run of the customer's tests is recorded with results, the verdict reached when evaluating each deployment for a trigger is kept, and a person can watch the agent open a browser and explore during mapping. There is no step by step record of the agent's own prompts, tool calls and decisions.

SourceQA Wolf, docs.qawolf.com triggers and diagnosing a trigger, coverage mapping, and qawolf.com/pricingread 2026-09-21

The Kerno agent runs on the customer's machine. The repository, code index, test environment, scenarios and baselines stay on the local filesystem and in local Docker containers, and no code is stored on Kerno's servers. The Enterprise plan adds self hosting in the customer's own infrastructure.

SourceKerno, kerno.gitbook.io/docs security and privacy, and kerno.io/pricingread 2026-09-21

Tests run on QA Wolf's own managed infrastructure, with a shared, allowlisted or private iOS device pool, local runs through its CLI, and Playwright tests customers can export at any time. The platform cannot be deployed in the customer's environment, there is no choice of data region, and QA Wolf does not say where data is kept.

SourceQA Wolf, docs.qawolf.com iOS device pools, CLI local execution, and qawolf.com/pricingread 2026-09-21

The product is a single testing agent with built in validation categories. There is no catalog of ready made agents, templates or packaged workflows for customers to adopt for their own work.

SourceKerno, kerno.io homepage and kerno.io/llms.txtread 2026-09-21

QA Wolf offers solution recipes and helper libraries for testing scenarios such as emails, barcodes, video injection and web performance. There is no catalog of ready made agents, templates or packaged workflows that a customer adopts for its own work.

SourceQA Wolf, docs.qawolf.com solutionsread 2026-09-21

To plan and reason about tests, Kerno sends code excerpts to Anthropic and OpenAI through its own proxy. Customers cannot choose the model or provider or bring their own keys, and there is no routing policy or fallback.

SourceKerno, kerno.gitbook.io/docs security and privacyread 2026-09-21

The judge model is the customer's choice in the non-deterministic AI assertions, and ChatGPT works as a drop in replacement with the same signature and return shape. They are designed for apps that use different AI providers, so the customer makes the model choice in its own tests.

SourceQA Wolf, docs.qawolf.com non-deterministic AI assertionsread 2026-09-21

Outside tools drive Kerno through an MCP server with a named set of tools for workspaces, environments, endpoint tests, jobs and feedback. There is also a CLI installed from npm, an HTTP jobs endpoint and an API key for non interactive sign in during CI. The local agent serves a REST API on localhost for the IDE extension and CLI, but Kerno publishes no reference for it and no SDK.

SourceKerno, kerno.gitbook.io/docs Kerno MCP, Kerno CLI and integrations and API keysread 2026-09-21

A REST API gives direct HTTP access to core platform actions, alongside a TypeScript CI SDK for triggering and reading runs, a CLI for running flows locally, flow and test libraries with API references, and deployment webhooks. An MCP server lets coding agents such as Claude Code and Codex build and run the suite. Together these give a stable API and SDK, an MCP interface and a fit into CI/CD.

SourceQA Wolf, docs.qawolf.com REST API overview, CI SDK reference, CLI and QA Wolf MCPread 2026-09-21

For each endpoint, Kerno generates functional, edge case and security tests and runs them against the customer's running app to capture a baseline. Whenever the customer's coding agent changes code, Kerno runs the affected tests again, reports each difference from the baseline and updates the suite when a change is intended. It also tests MCP servers' tools against the running server and replays the committed tests on every pull request as a CI check.

SourceKerno, kerno.gitbook.io/docs how Kerno works, change validation and running tests in CI, and kerno.io/llms.txtread 2026-09-21

An application's end to end workflows are mapped into a coverage outline, and the agent generates deterministic Playwright and Appium tests from plain language prompts and runs them in parallel on deploy, on a schedule or on pull requests. Judge model assertions return a structured verdict with hard and soft thresholds on a customer's AI generated responses, and tests validate MCP connections and tool execution. Testing happens before release, as a gate in CI and continuously after.

SourceQA Wolf, docs.qawolf.com why QA Wolf, test automation, non-deterministic AI assertions and triggersread 2026-09-21

Kerno tests backends, APIs, MCP servers and background workers over HTTP and message brokers. Its agent does not control a browser, desktop or computer.

SourceKerno, kerno.gitbook.io/docs overview and kerno.io/llms.txtread 2026-09-21

Computer use models let the agent open a browser and explore each feature of the customer's application, documenting the workflows it finds. A person can take control of the browser while the agent watches, and tests run in real browsers and on Android emulators and real iPhones and iPads. So the agent controls a real interface.

SourceQA Wolf, docs.qawolf.com coverage mapping and why QA Wolf, and qawolf.com/pricingread 2026-09-21