# AgentCompass Documentation

## Docs

- [Run GUI Grounding](https://agent-compass.mintlify.app/cookbooks/benchmark_patterns/run_gui_grounding.md): Evaluate screenshot-to-click behavior with ScreenSpot.
- [Run A Judge-Model Benchmark](https://agent-compass.mintlify.app/cookbooks/benchmark_patterns/run_judge_model_benchmark.md): Pass an LLM-as-judge configuration through benchmark params.
- [Run A Service Benchmark](https://agent-compass.mintlify.app/cookbooks/benchmark_patterns/run_service_benchmark.md): Run an evaluation whose harness delegates the agent loop to an external service.
- [Environment Selection](https://agent-compass.mintlify.app/cookbooks/first_runs/environment_selection.md): Choose host, Docker, Modal, or Daytona execution.
- [First Docker Run](https://agent-compass.mintlify.app/cookbooks/first_runs/first_local_run.md): Run one SWE-bench Verified task in Docker and inspect the result files.
- [Model API Setup](https://agent-compass.mintlify.app/cookbooks/first_runs/model_api_setup.md): Configure run-local model endpoint, protocol, and generation parameters.
- [Concurrency](https://agent-compass.mintlify.app/cookbooks/operations/concurrency.md): Control task parallelism without overrunning model or provider limits.
- [Post-Analysis](https://agent-compass.mintlify.app/cookbooks/operations/post_analysis.md): Run analyzers on completed result directories.
- [Resume And Results](https://agent-compass.mintlify.app/cookbooks/operations/resume_and_results.md): Resume interrupted runs and inspect durable artifacts.
- [Run SWE-bench With Daytona](https://agent-compass.mintlify.app/cookbooks/remote_benchmarks/run_swebench_daytona.md): Run one or more SWE-bench tasks in Daytona sandboxes.
- [Run SWE-bench With Modal](https://agent-compass.mintlify.app/cookbooks/remote_benchmarks/run_swebench_modal.md): Run SWE-bench Verified through Modal cloud sandboxes.
- [Run Terminal-Bench Remotely](https://agent-compass.mintlify.app/cookbooks/remote_benchmarks/run_terminalbench_remote.md): Run Terminal-Bench 2 with a remote sandbox provider.
- [Agentic Coding](https://agent-compass.mintlify.app/cookbooks/use_cases/agentic_coding.md): Evaluate repository repair and scientific coding agents.
- [Deep Research](https://agent-compass.mintlify.app/cookbooks/use_cases/deep_research.md): Run search-heavy, judge-scored, and scientific research benchmarks.
- [GUI Grounding](https://agent-compass.mintlify.app/cookbooks/use_cases/gui_grounding.md): Evaluate whether a VLM can locate UI targets from screenshots.
- [Productivity](https://agent-compass.mintlify.app/cookbooks/use_cases/productivity.md): Evaluate long-horizon productivity tasks such as GDPval AC, PinchBench, and SkillsBench.
- [Terminal Agents](https://agent-compass.mintlify.app/cookbooks/use_cases/terminal_agents.md): Run agents that operate through shell sessions and task workspaces.
- [Tool Calling](https://agent-compass.mintlify.app/cookbooks/use_cases/tool_calling.md): Evaluate model protocol compatibility and tool-use style workflows.
- [Architecture](https://agent-compass.mintlify.app/developer/architecture.md): Understand the AgentCompass runtime architecture.
- [Data Protocol](https://agent-compass.mintlify.app/developer/data_protocol.md): Shared objects between benchmarks and harnesses.
- [Developer Guide](https://agent-compass.mintlify.app/developer/develop_overview.md): Extend AgentCompass with new components.
- [Runtime Extensions](https://agent-compass.mintlify.app/developer/runtime_extensions.md): Add runtime components safely.
- [Daytona](https://agent-compass.mintlify.app/environments/daytona.md): Run remote sandbox sessions with Daytona recipes.
- [Docker](https://agent-compass.mintlify.app/environments/docker.md): Run reproducible local container environments.
- [Modal](https://agent-compass.mintlify.app/environments/modal.md): Run benchmark tasks in Modal cloud sandboxes.
- [Overview](https://agent-compass.mintlify.app/environments/overview.md): Choose local, container, and cloud execution providers.
- [Installation](https://agent-compass.mintlify.app/get_started/installation.md): Install AgentCompass from source with pip, uv, or conda.
- [Introduction](https://agent-compass.mintlify.app/get_started/introduction.md): AgentCompass is a unified evaluation runtime for LLM and VLM agents.
- [Setup](https://agent-compass.mintlify.app/get_started/setup.md): Configure model access, then choose a local or remote environment for your first runs.
- [Analyzers](https://agent-compass.mintlify.app/key_modules/analyzers.md): Post-execution badcase detection, statistics, latency checks, and qualitative trajectory diagnosis.
- [Benchmarks](https://agent-compass.mintlify.app/key_modules/benchmarks.md): Dataset loading, task selection, task preparation, scoring, and aggregation.
- [Configuration](https://agent-compass.mintlify.app/key_modules/configuration.md): How AgentCompass turns YAML defaults, CLI flags, recipes, and component params into one run request.
- [Environments](https://agent-compass.mintlify.app/key_modules/environments.md): Execution and file primitives for local, container, and remote sandbox backends.
- [Harnesses](https://agent-compass.mintlify.app/key_modules/harnesses.md): Execution adapters for model calls, CLI agents, terminal agents, and external frameworks.
- [Models](https://agent-compass.mintlify.app/key_modules/models.md): Run-local model endpoint, protocol, API key, and generation parameters.
- [Overview](https://agent-compass.mintlify.app/key_modules/overview.md): A map of the AgentCompass runtime modules and how they work together.
- [Recipes](https://agent-compass.mintlify.app/key_modules/recipes.md): Provider-aware execution plan rewrites before environment startup.
- [Results](https://agent-compass.mintlify.app/key_modules/results.md): Run directories, per-task details, summaries, progress, logs, and analysis artifacts.
- [Runtime](https://agent-compass.mintlify.app/key_modules/runtime.md): The orchestration layer that turns one request into a bounded evaluation run.
- [Analyzers](https://agent-compass.mintlify.app/reference/analyzers.md): Configure and run post-execution analysis.
- [BrowseComp](https://agent-compass.mintlify.app/reference/benchmarks/browsecomp.md): Browse-heavy research benchmark.
- [BrowseComp-ZH](https://agent-compass.mintlify.app/reference/benchmarks/browsecomp_zh.md): Chinese web-browsing research benchmark.
- [DeepSearchQA](https://agent-compass.mintlify.app/reference/benchmarks/deepsearchqa.md): Search QA benchmark.
- [FrontierScience](https://agent-compass.mintlify.app/reference/benchmarks/frontierscience.md): Expert-level scientific task benchmark.
- [GAIA](https://agent-compass.mintlify.app/reference/benchmarks/gaia.md): General assistant benchmark.
- [GDPval AC](https://agent-compass.mintlify.app/reference/benchmarks/gdpval_ac.md): Economically valuable task benchmark adapted for AgentCompass.
- [HLE](https://agent-compass.mintlify.app/reference/benchmarks/hle.md): Human-level evaluation benchmark.
- [HLE Verified](https://agent-compass.mintlify.app/reference/benchmarks/hle_verified.md): Verified subset of Humanity's Last Exam.
- [Benchmarks Overview](https://agent-compass.mintlify.app/reference/benchmarks/overview.md): Benchmark ids, tags, and execution shapes.
- [PinchBench](https://agent-compass.mintlify.app/reference/benchmarks/pinchbench.md): OpenClaw-style productivity benchmark.
- [ResearchClawBench](https://agent-compass.mintlify.app/reference/benchmarks/researchclawbench.md): End-to-end autonomous scientific research benchmark.
- [SciCode](https://agent-compass.mintlify.app/reference/benchmarks/scicode.md): Scientific coding tasks.
- [ScreenSpot](https://agent-compass.mintlify.app/reference/benchmarks/screenspot.md): GUI grounding benchmark.
- [SGI Deep Research](https://agent-compass.mintlify.app/reference/benchmarks/sgi_deep_research.md): Scientific general intelligence deep-research subset.
- [SkillsBench](https://agent-compass.mintlify.app/reference/benchmarks/skillsbench.md): Terminal tasks with on-demand skill loading.
- [SWE-bench Multilingual](https://agent-compass.mintlify.app/reference/benchmarks/swebench_multilingual.md): Multilingual repository repair tasks.
- [SWE-bench Pro](https://agent-compass.mintlify.app/reference/benchmarks/swebench_pro.md): Harder professional repository tasks.
- [SWE-bench Verified](https://agent-compass.mintlify.app/reference/benchmarks/swebench_verified.md): Repository repair benchmark.
- [Terminal-Bench 2](https://agent-compass.mintlify.app/reference/benchmarks/terminal_bench_2.md): Terminal-agent tasks in remote shell environments.
- [Terminal-Bench 2.1](https://agent-compass.mintlify.app/reference/benchmarks/terminal_bench_2_1.md): Terminal-Bench 2.1 task set.
- [Terminal-Bench 2 Verified](https://agent-compass.mintlify.app/reference/benchmarks/terminal_bench_2_verified.md): Verified subset of Terminal-Bench 2 tasks.
- [WildClawBench](https://agent-compass.mintlify.app/reference/benchmarks/wildclawbench.md): Real-world long-horizon agent benchmark for OpenClaw-style runtimes.
- [CLI](https://agent-compass.mintlify.app/reference/cli.md): Run evaluations, regenerate summaries, re-run analysis, and list registered components.
- [Configuration](https://agent-compass.mintlify.app/reference/configuration.md): Field-by-field reference for built-in config defaults.
- [API Reference Overview](https://agent-compass.mintlify.app/reference/overview.md): Runtime, benchmark, environment, analyzer, and developer reference for AgentCompass.
- [Python API](https://agent-compass.mintlify.app/reference/python_api.md): Call AgentCompass from Python.
- [Results](https://agent-compass.mintlify.app/reference/results.md): Understand details, summaries, logs, progress, and analysis artifacts.
- [Supported Components](https://agent-compass.mintlify.app/reference/supported_components.md): Registered benchmarks, harnesses, and model API protocols supported by the current AgentCompass runtime.