Drift is an innovative intent-based programming language tailored for building agentic systems. With Drift, developers can create agents using intuitive English-like syntax, which is then transpiled into asynchronous Python code that operates seamlessly on Drift's efficient runtime.

The structure and features of Drift support complex automation and intelligent decision-making, allowing for streamlined workflows with minimal overhead. Below is a succinct overview of what Drift offers:

Key Features

• Agent Creation: Define agents with properties such as model, budget, and procedural steps using a clear, readable format.
• Rich Functionality: Utilize built-in intent verbs like summarize, extract, and classify, making it easy to implement advanced functionalities.
• Efficient Parallel Processing: Leverage asynchronous capabilities with constructs like for each x in xs parallel, enabling optimized performance in handling tasks concurrently.
• Error Handling: Benefit from a robust error management system with features for attempt and recover, ensuring reliability in agent execution.

Example Usage

Create an agent to triage emails as follows:

agent InboxTriage {
  model: "gpt-5.4-nano"
  budget: $0.10 per run

  step triage(emails: list<string>) -> list<EmailAnalysis> {
    let analyses = []
    for each email in emails parallel {
      let analysis = classify email as EmailAnalysis
      analyses.add(analysis)
    }
    for each a in analyses {
      if a.priority == "urgent" {
        respond "URGENT {a.subject}: {a.summary}"
      }
    }
    return analyses
  }
}

This example demonstrates how to define an agent, select a model, manage budgets, and handle email classifications effectively.

Quick Start

Create your first agent in just 30 seconds without requiring an API key:

drift new hello
cd hello
drift run hello.drift --input '{"name":"Riley"}'

Drift will automatically revert to a mock provider, enabling instant testing of the system.

Comprehensive Documentation

Access a variety of resources to explore and understand Drift’s capabilities:

• Coding agents across different platforms in LLM.md
• Learning materials in docs/language.md
• Practical examples in the docs/cookbook.md
• Common pitfalls in docs/gotchas.md

Examples Repository

Explore a collection of functional .drift programs detailed in the examples directory, showcasing various applications of Drift:

• Minimal agents
• Conditional and parallel processing
• Integration of memory via Dendric backend

Drift represents a modern approach to building intelligent agents, blending ease of use with powerful capabilities ideal for developers looking to harness the full potential of agent-based systems.