JetBrains Context Indexes Multi-Repo Code to Speed Up AI Agents

Autonomous coding agents burn a remarkable amount of tokens and compute time simply searching directories, resolving definitions, and crawling dependencies. To rein in that recursive wandering, JetBrains launched an early access program for JetBrains Context, a semantic repository-indexing service built for agents like Claude Code, Codex CLI, and Junie CLI.

Instead of forcing an agent to run blind text searches or making developers clone every related project locally, the service builds a semantic representation of code structure across an organization's repositories. Agents can query implementations, APIs, and dependencies by meaning rather than exact keywords, even across projects housed entirely on remote company servers. The workflow runs through a terminal utility, jbcontext, with hooks to trigger indexing automatically or inspect estimated compute savings.

According to figures published by JetBrains from tests on SWE-bench and production monorepos, pre-computed semantic indexes cut agent steps by up to 68%, latency by up to 59%, and execution cost by up to 48%. These reflect best-case measurements; the company did not publish granular per-task distributions or exact agent setups in the preview.

The convenience comes with an enterprise architectural trade-off. While JetBrains notes that raw source code is parsed locally and not stored in plaintext on company servers, the resulting vector index—including code structure summaries, embeddings, extracted comments, and metadata—is synced to an internal database hosted on Google Cloud Platform. Using the preview requires an active JetBrains AI license, though the company is not currently billing extra or deducting credits for index queries.

Source: habr

Primary source: JetBrains Blog, JetBrains Context EAP Terms