JetBrains' AI coding agent, built into IntelliJ IDEA, PyCharm, WebStorm, and the rest of the JetBrains family, plus a standalone Junie CLI — reached GA on June 17, 2026 with real debugger integration as the headline differentiator.
What It Is
Junie is JetBrains' answer to the Copilot/Cursor wave — an AI coding agent that lives inside the JetBrains IDEs developers already use, rather than asking them to switch to a new editor. The GA release (June 17, 2026) rebuilt the IDE integration on ACP (Agent Communication Protocol) and shipped agentic debugging as its lead feature: Junie can set real breakpoints, step through execution, inspect variables, and iterate on fixes against live runtime state — using the same debugger the developer would use manually.
The Junie CLI extends the same agent to terminal, CI/CD pipelines, and GitHub/GitLab workflows, operating independently of a running IDE.
Why It's in Trial
JetBrains has a large installed base — IntelliJ, PyCharm, and WebStorm together serve the majority of professional Java, Kotlin, and Python developers. Junie gives that base an agentic upgrade path without switching editors. The ACP-based architecture and real debugger access are meaningfully differentiated from Cursor's VS Code base and from Copilot's chat-centric approach.
The debugger integration is the strongest argument for Junie. Instead of the typical AI-coding loop of "add print statements, re-run, ask the agent about the output," Junie can set breakpoints and inspect state mid-execution — a workflow native Java and Kotlin developers expect. This is genuinely novel compared to the rest of the field.
Benchmark scores are now published: JetBrains reports 60.8% on SWE-bench Verified (cited at GA launch), and the independent SWE-Rebench leaderboard (June 2026 run) shows 61.6% resolved with 72.7% pass@5 — placing Junie in the competitive tier alongside other leading coding agents. Trial rather than Adopt because GA is still recent (June 17, 2026), production reliability at scale is unproven, and JetBrains' execution on agentic tools has historically been slower than Cursor or Anthropic. Windsurf's collapse-then-acquisition (into GitHub Copilot) in 2026 is a reminder that even well-funded IDE competitors can fail quickly in this market.
Key Capabilities at GA
- Agentic debugging: Set breakpoints, step through code, inspect variables — not simulated, using the real JetBrains debugger
- ACP-native: Rebuilt on Agent Communication Protocol; more reliable task handoffs between orchestrator and IDE than the beta
- Cross-platform: macOS, Windows, and Linux — no omissions unlike JetBrains Air (which still lacks Windows)
- BYOK or JetBrains cloud: Use JetBrains' hosted model access or bring your own API key from Anthropic, OpenAI, Google, xAI, OpenRouter, or GitHub Copilot
- Local model support: Ollama, LM Studio, LiteLLM — prompts never leave your machine
- Junie CLI: Same agent in terminal and CI/CD without a running IDE
Limitations
- GA is recent (June 17, 2026) — production reliability at scale is unproven
- Requires a JetBrains subscription; the free tier has usage limits
- JetBrains IDE-only — not useful if your team uses VS Code, Neovim, or Emacs
- Model default is configurable but not always the frontier model; heavy users need to set API keys manually for best results
Key Characteristics
| Property | Value |
|---|---|
| Interface | IDE plugin (IntelliJ, PyCharm, WebStorm, etc.) + CLI |
| Provider | JetBrains |
| License | Proprietary |
| Pricing | Free tier; AI Pro $10/mo; AI Ultimate $30/mo |
| Underlying model | GPT-5, Claude Sonnet 4.5, Gemini 2.5 Pro (configurable); BYOK from 6 providers; local via Ollama/LM Studio/LiteLLM |
| Sandbox | Application-layer (runs inside IDE process) |
| Benchmark | 60.8% SWE-bench Verified (JetBrains, GA launch); 61.6% SWE-Rebench resolved, 72.7% pass@5 (independent, June 2026) |
| Website | jetbrains.com/junie |
| Docs | junie.jetbrains.com |