ABS is available as a CLI (npm/Python) and as a .NET library. The CLI provides commands to initialize, run, and report on ABS sessions; the .NET library embeds the same parser, runner and evaluators directly in your C# application.

Prerequisites

Node.js 18+ or Python 3.11+ for the CLI
.NET 10 SDK for the C# library
A package manager: npm, yarn, pnpm, pip, or NuGet

Install via npm

npm install -g abslang

Verify the installation:

abslang --version
# abs v0.1.0

Install via pip

pip install abslang

Install the .NET library

dotnet add package AbsLang

Optional cloud adapters:

dotnet add package AbsLang.Adapters.Aws
dotnet add package AbsLang.Adapters.Google

using AbsLang;

var session = AbsParser.Parse("sessions/order-status.abs.yaml");
var result = await new AbsRunner().RunAsync(
    session,
    new AgentConfig { Url = "http://localhost:8080/chat" });

Console.WriteLine(AbsReports.FormatTable(result));

The .NET package supports openai, responses, claude and gemini agent protocols, tool round-trips, optional behaviors, LLM judges, cloud adapters, reference metrics, and structured RunLogger events (including duration_ms) for Datadog/OpenTelemetry. See the C# guide for the full API. The CLI and chat remain available only through npm and pip.

Initialize a new project

abslang init my-agent-tests
cd my-agent-tests

This creates:

my-agent-tests/
├── abs.config.yaml    # Configuration
├── sessions/           # Your .abs.yaml files go here
└── README.md

Don't want to write YAML? Use the chatbot

Describe the behavior in plain language and abslang generates the .abs.yaml for you:

abslang chat
# Uses OPENAI_API_KEY, ANTHROPIC_API_KEY, or DEEPSEEK_API_KEY — whichever is set
# You: A customer asks for a refund. The agent should verify the order, process it, and confirm.
# → generates a complete .abs.yaml with evaluations, datasets, and chain checks

Run a session

abslang run sessions/order-status.abs.yaml

For more details on the CLI, see the CLI Reference.

Next steps

Tutorial — a 20-minute step-by-step guide
CLI Reference — full command documentation
Runner — executing sessions against real agents

Installation