The best protein design agent is already on your computer.
The coding agent you already use can read scientific literature, reason about protein structures, write analysis code, and work through a complex experiment with you. Give it access to the right tools and scientific guidance, and it can help you design proteins, too.
You don't need a separate proprietary protein-design agent. Whether you work with Claude Opus 5.5, GPT-6, or another capable model, you should be able to bring that intelligence into your research with the agent you prefer.
The Ariax CLI equips your existing coding agent for protein design in a single package: access to hosted design tools through Ariax's API, engine-specific skills, and a shared workspace where you and your agent can prepare campaigns and review results together.
CLI/Agent jobs can save up to 40% on GPU compute during October. See the pricing update for the rates.
The capability is already here
Recent results show why this approach is compelling. In Anthropic's protein-design study, Claude orchestrated publicly available design and structure-prediction tools. Independent labs confirmed binders against 14 of 15 targets, with overall design hit rates of 22.6–35.1% across the tested models and campaign setups. These experiments used Opus 4.8 and Mythos Preview in Claude Science.
More recently, the Claude Opus 5.5 system card reported 82.6% on an autonomous binder-design benchmark: 24-hour campaigns using protein-design tools across 15 targets. That is a computational score based on predicted interfaces, pose agreement, and structural diversity—not an experimental binding rate.
Together, these results support a powerful idea: a general-purpose agent, equipped with specialist tools, can do serious protein-design work. They don't establish a hit rate for Ariax or every model. They show the opportunity to put increasingly capable agents to work on your scientific questions.
Give your agent the tools and the know-how
Access to an API is only part of the job. A useful collaborator also needs to understand how a design engine works: which inputs it expects, which settings matter for the experiment, and what its outputs actually mean.
The Ariax CLI bundles that guidance alongside the commands. Engine-specific skills cover input preparation, configuration, and output interpretation. Shared guides cover campaign planning, compute selection, and decisions about what to run next. Your agent reads the live configuration schema, validates the proposed setup, and uses the same project API throughout the campaign.
That turns a broad request into concrete work. Ask your agent to inspect a target, prepare a design plan, and explain its choices. Review the scientific settings and compute scope together. Once authorized, it can submit the campaign, follow progress, and analyze the results while Ariax provisions the GPUs and runs the design tools.
Choose the workflow that fits the experiment. Your agent can consult the workflow comparison and bundled skills to help make that decision.
Your science. Your agent. One workspace.
Agentic protein design should fit the way scientists work. Sometimes you want to inspect a structure yourself. Sometimes you want your agent to work through a candidate table, compare outputs, or prepare the next experiment. Often you want both.
Ariax connects those activities through the same projects. Start a campaign on the website and bring your agent in later. Have your agent prepare a project, then review its structures and progress in the browser. Give it an existing project URL or ID and ask what happened, which candidates deserve a closer look, or what evidence would justify another round. Your account's access permissions apply throughout.
You set the scientific direction. The agent helps carry the work forward, keeping its recommendations connected to settings, structures, scores, and observed compute costs. Computational results guide the next experiment; binding and biological activity still need to be tested in the lab.
Connect your agent
Install the CLI through either channel. Both require Node.js 20 or newer and npm.
npm — stable release
npm install --global ariax-cli@latest
GitHub — latest source from main
curl -fsSL https://raw.githubusercontent.com/cytokineking/ariax-cli/main/install.sh | sh
The GitHub channel downloads the source directly and uses npm to install it locally; it may include unreleased changes. You can review the installer before running it.
Create a Full access key in Settings → API keys, then run ariax login and paste it into the hidden terminal prompt. Keep the key out of agent conversations.
Give your agent a goal and this starting instruction:
Use the Ariax CLI for this project. Run ariax skills --read --json and read
data.content, then read the matching protocol and outputs guides. Fetch the
live schema, prepare and validate inputs, and show me the scientific plan and
compute scope before launching. Carry forward any authorization I have already
given. Review progress, candidates, and observed costs before recommending
the next step. If login is needed, ask me to run ariax login in my terminal;
never ask for an API key in this chat.
The skills ship with the CLI; no separate skill installation or MCP server is required. Use Claude Code, Codex, Cursor, or another agent with terminal access. Bring your scientific question. Equip the agent you already know.
