Skip to content

Building the #1 Open Source Terminal-Use Agent Recent

Published: August 5, 2025 | Original Post


Letta achieved a 42.5% score on Terminal-Bench, ranking as the #1 open-source agent for terminal tasks.

Technical Architecture

The terminal agent uses two memory blocks:

  • A read-only task description section
  • A modifiable to-do list for planning

Specialized tools include send_keys for command execution and task_completed for signaling completion. When context reaches approximately 40,000 tokens, Letta performs recursive summarization.

Performance

Ranked 4th overall and 2nd among Claude 4 Sonnet-based solutions on Terminal-Bench, covering 100+ challenging tasks including kernel compilation, ML model training, and server configuration.

Design Philosophy

Agents with effective context management can achieve significant gains on extended tasks. By structuring memory into discrete functional units and enabling dynamic updates, the architecture enables consistent performance without manual intervention.

Resources