Tura
Tura is an open-source agent runtime harness that improves LLM agent performance by reducing token usage and model round trips. It solves the inefficiency of ReAct-style tool-calling loops by executing multi-step workflows as a single runtime-managed command graph. It is for developers building or using coding agents that need better performance and lower costs.
✨ Key features
- Macro CLI command run: executes multi-step workflows in one LLM turn.
- Backward reasoning: guides LLM to reason from goal backward.
- Runtime context and prompt manager: scopes context to current task.
- Supports CLI, TUI, GUI, and gateway interfaces.
- Benchmark-verified: 77.5% fewer tokens and 16.7% better success rate.
- Open-source with npm and source installation.
🎯 Use cases
- Run coding agent workflows with fewer tokens and turns.
- Improve success rate on long-horizon development tasks.
- Manage context and compaction in agent sessions.
- Execute shell commands and apply patches in a single macro call.
- Build custom agents with runtime prompts and task status.
📦 Installation
🧰 Requirements: Requires Node.js for npm install, or Rust toolchain for source build; no bundled provider credentials, must configure an LLM provider on first launch.
NPM release
Global install (macOS, Linux, and Windows):
npm install -g tura-ai
tura
Local project install:
npm install tura-ai
npx --no-install tura
Source checkout
Windows PowerShell:
git clone https://github.com/Tura-AI/tura.git
cd tura
.\scripts\install.ps1
tura
macOS or Linux shell:
git clone https://github.com/Tura-AI/tura.git
cd tura
./scripts/install.sh
tura
🚀 Usage
tura "prompt"
For direct execution without TUI:
tura exec "prompt"
Example of a macro command_run tool call:
{
"name": "command_run",
"arguments": {
"commands": [
{
"step": 1,
"command_type": "shell_command",
"command_line": "rg -n \"TODO|command_run|handler\" crates/"
},
{
"step": 2,
"command_type": "apply_patch",
"command_line": "*** Begin Patch\n*** Update File: crates/tools/src/command_run/handler.rs\n@@\n- // old command handler logic\n+ // patched command handler logic\n*** End Patch"
}
]
}
}
⚠️ Good to know
Published benchmark results do not establish equivalent quality for every provider; broader measurements are part of the roadmap and known evidence gaps.
❓ FAQ
How does Tura reduce token usage?
Tura executes multi-step workflows as a single runtime-managed command graph, avoiding repeated model round trips and context re-entry, leading to fewer tokens.
What is backward reasoning?
Tura guides the LLM to reason backward from the goal state to the current state, which helps avoid statistically common but mediocre code.
Does Tura bundle provider credentials?
No, Tura does not bundle provider credentials. You must configure an LLM provider on first launch.
What interfaces does Tura provide?
Tura offers a CLI, TUI, GUI, and a gateway for HTTP/SSE streaming.
📊 Repository
🤖 Overview, features, install steps and FAQ were generated from the project's README on Sep 4, 2026. Always check the original source before running commands.