Tura

Tura is an agent runtime and optimization platform that aims to cut the language model costs of AI agents while keeping completed outcomes and execution evidence auditable. It installs through npm and runs as a command line agent, replacing individual tool calls with what the vendor calls macro commands so that a task takes fewer model turns. It also helps define success benchmarks and automates evaluation of agent success rates, development effort, iteration cost and runtime cost on representative workflows. It is aimed at enterprise teams running long horizon agent work who want measurable savings. The vendor publishes a benchmark covering 558 sessions across DeepSWE, repository rewrite and design tasks, reporting an 83.1 percent reduction in model turns and token use on the tested benchmark alongside a 16.7 percent performance improvement. It invites teams to contribute their own representative tasks and evaluation reports, and design audits are reported separately from the other results. Pricing is published as two tiers. The community tier is free for individuals and open source work, with a self hosted runtime, the standard configuration and best effort community support. The commercial license costs 22 percent of verified savings, with no setup fee, so the customer keeps 78 percent. It adds workflow optimization, a guaranteed savings floor based on an agreed baseline of tokens per task, and a 24 hour technical response on a pinned runtime version. Commercial terms are set through a booked call and a signed agreement.