Crate llm_browser_testkit 

LLM-driven browser test framework.

Provides reusable building blocks for browser-based test scenarios:

• Browser client via Chrome DevTools Protocol (headless)
• LLM client for natural language element targeting and assertions
• A2A agent integration for agent-to-agent communication
• MCP client/server integration for tool-calling
• Cost tracking, token counting, and budget enforcement
• Declarative TOML scenario runner
• #[browser_test] macros for cargo test integration

Re-exports

pub use costs::LlmResponse;

pub use costs::LlmUsage;

pub use scenario::AuthConfig;

pub use scenario::AuthMode;

pub use scenario::AwsConfig;

pub use scenario::Provider;

Modules

a2a

A2A agent protocol client. A2A (Agent-to-Agent) protocol client.

budgets

Budget tracking and enforcement. Budget tracking and enforcement.

costs

Cost calculation, usage tracking, and pricing. Cost calculation, usage tracking, and pricing logic.

diagnostics

Failure diagnostics — page-state capture and artifact writing. Failure diagnostics — page-state capture and artifact writing.

endpoints

Endpoint registry and routing resolver. Endpoint registry — resolves named endpoints and task-type routing.

events

Typed run events emitted by the runner. Typed run events emitted by the runner.

mcp_client

MCP client for connecting to external MCP servers. MCP (Model Context Protocol) client.

mcp_server

MCP server for exposing the framework as an MCP server. MCP server exposure — exposes the framework as an MCP server.

parallel

Concurrent execution of multiple scenario files across isolated browsers. Concurrent execution of multiple scenario files.

pricing

Automatic pricing lookup from providers that expose exact, public per-token prices.

redact

Secret redaction for every report sink. Secret redaction for every report sink.

reporting

Run reporting: console, NDJSON, JUnit, GitHub and Perfetto sinks. Report sinks: console, NDJSON, JUnit XML, GitHub annotations, Perfetto trace.

runner

Step-by-step scenario executor (navigate, click, type, wait, assert).

scenario

Declarative TOML-based test scenario types. Scenario types for human-readable browser test case definitions.

selectors

CSS selector sanitization for LLM-generated selectors. CSS selector sanitization and validation for LLM-generated selectors.

vision

Vision support — screenshot capture/downscale/encode for visual asserts. Vision support — screenshot capture, tiling, downscaling, and JPEG encoding for LLM visual assertions.

Structs

LlmConfig

Configuration for the LLM client — bundles URL, model, auth, timeouts, and provider-specific options into a single struct passed everywhere.

Constants

DEFAULT_AZURE_API_VERSION

Default Azure OpenAI API version used when an endpoint does not set api_version.

DOM_EXTRACT_JS

JavaScript to extract interactive elements from the current page. Returns a JSON array of objects with tag, selector, and label.

Functions

base_url

Returns the target base URL from HARNESS_BROWSER_BASE_URL env, defaulting to http://localhost:4200.

browser_headless

Returns whether to run the browser in headless mode from HARNESS_BROWSER_HEADLESS env, defaulting to true.

build_azure_url

Builds the Azure OpenAI chat completions URL: the resource endpoint (without /openai), the deployment name, and the API version.

default_llm_attempts

Reads HARNESS_LLM_CALL_ATTEMPTS (default 3) — how many times a single chat completion is retried before the endpoint is considered failed.

http_client

Builds a reqwest::Client with the given timeout.

llm_base_url

Returns the LLM server base URL from HARNESS_LLM_TEST_URL env, defaulting to http://localhost:8080.

llm_chat

Sends a chat completion request to the LLM.

llm_chat_vision_with_usage

Sends a vision-enabled chat completion request.

llm_chat_vision_with_usage_chain

Vision variant of llm_chat_with_usage_chain.

llm_chat_with_usage

Sends a chat completion request to the LLM and returns both the content and token usage from the API response.

llm_chat_with_usage_chain

Calls a chain of endpoints: the primary LlmConfig first, then each fallback in order. Every endpoint gets its own max_attempts retry budget; the first endpoint that answers wins.

llm_model

Returns the LLM model name from HARNESS_LLM_TEST_MODEL env, defaulting to deepseek.

parse_headers_env

Parses HARNESS_LLM_HEADERS env var (JSON object) into a header map.

truncate

Truncates a string to the given maximum length, appending a marker with the number of omitted characters if truncation occurred.