# JetBrains Context JetBrains Context is a repository intelligence layer for coding agents, announced in July 2026 and in early access at no extra cost with a JetBrains AI subscription. It builds a semantic index of your repositories and lets agents query it by concept instead of by keyword, so they stop burning turns rediscovering the codebase. The claimed effect, across 2,333 tasks: up to **68% fewer agent turns**, **59% less latency**, **48% lower cost**. ## The problem it targets Watch any coding agent start a task and you see the same opening: grep, read a file, grep again, read three more, build a mental model of the code, then finally begin. That exploration is repeated on every single task, it consumes most of the early context window, and it produces nothing durable. The agent throws the map away and redraws it next time. JetBrains Context makes the map persistent and shared. Index once, incrementally, then answer semantic questions against it. **Multi-repo search is the part that stands out.** It searches across an organization's whole codebase, not only the projects you have checked out locally. That is a capability no amount of clever prompting gives you, because the files are not on the machine. ## How it works - **Backend indexing** builds a semantic index incrementally - **Semantic retrieval** exposes tools the agent calls with concepts rather than keywords - CLI-driven setup (`jbcontext setup-agent`, `jbcontext index`), with agent hooks firing during normal workflows - Works with [[Claude Code]], [[OpenAI Codex]], and JetBrains Junie - Works in [[JetBrains]] IDEs, Air, VS Code, and others ## Why it matters **This is [[Context Engineering]] shipped as infrastructure.** Everyone agrees the context window is the scarce resource. Most responses to that are per-project conventions: an AGENTS.md, a set of rules, a well-written README. Those help the agent know *how* you work. They do not help it find things. Indexing is the other half, and it is the half that needs a server rather than a markdown file. **Cross-agent is the right bet.** Supporting Claude Code and Codex alongside their own Junie says JetBrains reads the index as the durable asset and the agent as swappable. That is almost certainly correct, and it is the opposite of how most vendors are positioning. **Turn reduction compounds with everything else.** A 68% cut in turns is a cut in latency, in cost, and in the number of chances the agent has to go off the rails. In a swarm, where the same exploration is repeated by every worker, the multiplier is larger still. See [[Cursor Agent Swarms]] for what that waste looks like when nobody controls it. ## Caveats - The benchmark spread is wide and the "up to" numbers come from the favourable end. 205 SWE-bench tasks, 175 production monorepo tasks, 1,953 localization tasks; localization is the easiest of the three and dominates the count - JetBrains benchmarked JetBrains - The integration mechanism is not clearly documented as [[Model Context Protocol (MCP)]] or a standard API, which matters if you want to use it outside the supported agents - Early access, free for now, with a JetBrains AI subscription. Pricing after early access is unstated, and index-as-a-service is an easy thing to start charging for - Sending an index of your entire organizational codebase to a vendor backend is a procurement conversation, not a checkbox ## References - Announcement — https://blog.jetbrains.com/ai/2026/07/introducing-jetbrains-context-repository-intelligence-for-coding-agents/ ## Related - [[JetBrains]] - [[Context Engineering]] - [[AI Agents]] - [[AI Agent Harness]] - [[How Coding Agents Work]] - [[Agentic Engineering]] - [[Claude Code]] - [[OpenAI Codex]] - [[Model Context Protocol (MCP)]] - [[Cursor Agent Swarms]] - [[AI Tool Use]] - [[Knowledge Graph (KG)]]