Tura

About Tura

Tura is an open-source agent runtime harness for coding agents, designed to complete development workflows such as patching, compiling, validation, log inspection, and follow-up decisions within a single LLM turn using CLI macros and decision trees. It manages its own context window and runtime prompt to reduce the effect of irrelevant context, and before making a patch it investigates the execution path and reproduces the issue first. The project publishes benchmark comparisons against Codex CLI across DeepSWE, rewrite, and design tasks, reporting fewer agent turns and a higher DeepSWE pass rate; the benchmark cohort is open to community-contributed tasks and evaluation reports.

It is installable via npm (tura-ai) or by cloning its GitHub repository and running an install script (macOS/Linux or Windows), and is aimed at developers building or evaluating coding agents who want an alternative to tool-calling, context-compaction, or skills-based agent frameworks.

Key features

• Open-source agent runtime harness for long-horizon coding workflows
• Executes patching, compiling, validation, and log inspection via CLI macros in a single LLM turn
• Uses backward reasoning from outcomes to root causes
• Manages its own context window and runtime prompt via CLI commands
• Investigates and reproduces issues before making a patch (test-driven approach)
• Community benchmark with published sessions across DeepSWE, rewrite, and design tasks

Frequently asked questions