Tura-AI/tura
Tura says its runtime-managed command graph cuts repeated agent turns while improving benchmark pass rates.
The project describes an open-source agent harness that replaces repeated ReAct-style tool loops with a single `command_run` workflow. Its published DeepSWE comparison says Tura Balanced reached 80.0% verifier success while using 31.1% fewer tokens than Codex CLI. Tura Direct is framed as the lower-cost mode, using 77.5% fewer aggregate tokens with a similar success rate. The README also flags evidence gaps, including broader provider, UI latency, parsing, and cross-OS measurements. GitHub · LLM repos' note
The project describes an open-source agent harness that replaces repeated ReAct-style tool loops with a single `command_run` workflow. Its published DeepSWE comparison says Tura Balanced reached 80.0% verifier success while using 31.1% fewer tokens than Codex CLI. Tura Direct is framed as the lower-cost mode, using 77.5% fewer aggregate tokens with a similar success rate. The README also flags evidence gaps, including broader provider, UI latency, parsing, and cross-OS measurements. GitHub · LLM repos' note
score 5