JetBrains at present is launching its Air agentic improvement surroundings into public preview. Air was created to allow builders to construct instruments round brokers, information them and fine-tune their outputs.
In line with the corporate, Air is designed to finish the present state of agent fragmentation, during which every agent runs in a unique software, with completely different setup, context and an absence of structural understanding of the code. It additionally eases the switching of brokers throughout initiatives, the place they aren’t being migrated however as a substitute are part of the workflow.
With Air, builders can outline duties by mentioning “a selected line, commit, class, technique, or different image,” JetBrains wrote in its weblog asserting Air. “In consequence, the agent will get exact context as a substitute of a blob of pasted textual content.” When the duty is finished, Air allows customers to see adjustments not solely on the code diff, however within the context of the codebase. Instruments equivalent to terminal, Git consumer and built-in preview are also enabled.
Air helps the Agent Consumer Protocol (ACP), an open normal that lets AI coding brokers combine with any code editor or IDE, and helps Codex, Claude Agent, Gemini and Junie. JetBrainssaid it can add help for different brokers by the ACP Agent Registry. Customers will be capable of run brokers regionally, or isolate them in containers and Git worktrees for sandboxing and concurrent work, the corporate wrote.
Run brokers regionally by default, or isolate them in Docker containers and Git worktrees for sandboxing and concurrent work.
Air lets you keep away from the mess of getting a number of home windows and terminal tabs open for every activity. You see one activity (that means one agent session) at a time. You’ll get a notification when one other activity wants your consideration, so you may rapidly swap to it whereas the agent retains working. Air then helps deliver your adjustments from a container or worktree to your important copy.
