JetBrains Junie Wypróbuj
Overview
JetBrains Junie embeds agentic coding in IntelliJ-based IDEs with project context, refactor awareness, and enterprise admin controls (Junie).
Trial for JVM-heavy shops standardized on JetBrains. Junie policies should be aligned with existing IDE plugin and data export rules. Since 2026 Junie also runs as a standalone CLI in terminals and CI (GitHub Actions, GitLab CI/CD) (Junie CLI beta).
Adoption Signals
- General availability: Junie left beta in June 2026 and is available in all JetBrains IDEs and through Junie CLI, with plan mode, agentic debugging, remote tasks, code review, and DataGrip database access (Junie leaves beta).
- Third-party benchmark, as quoted by JetBrains: in the SWE-rebench cycle cited by Nebius's research lead, Junie was the top model-harness at 61.6% resolved and 72.7% pass@5 (Junie leaves beta).
- CI and terminal reach: Junie CLI (beta since March 2026) is LLM-agnostic, supports MCP servers, and integrates with GitHub and GitLab pipelines (Junie CLI beta).
- Data-residency option: since August 2026 Junie can run fully on-device with a quantized Qwen3.6-27B model on M5 Macs with 64 GB RAM, without credits or cloud calls, at reduced capability on complex reasoning (Junie local launch).
- Published pricing: a free tier with 5 AI credits and BYOK, AI Pro at $8.33 and AI Ultimate at $25 per user and month (annual billing) (Junie).
Risks
- Untrusted project files. CVE-2025-59458 (CVSS 8.3) allowed code execution through improper command validation, and CVE-2026-41153 allowed command execution via a malicious project file; both are fixed, but they show the attack path (CVE-2025-59458, CVE-2026-41153).
- Credit consumption. Agent sessions draw on a fixed monthly credit allowance per seat; long autonomous runs or expensive models can use it up long before the month ends (Junie).
- Governance maturity. In JetBrains Central, AI policies and model management require the Business or Enterprise tier, AI audit logs are Enterprise-only, and SCIM was listed as "coming soon"; the console documentation was still marked Early Access (JetBrains Central plans).
- Ecosystem coupling. The strongest features (refactoring, debugger, DataGrip) depend on JetBrains IDEs; teams with mixed VS Code and JetBrains usage end up running two agent stacks.
Pros & Cons
Advantages
- Generally available since June 2026 across all JetBrains IDEs and as Junie CLI, with plan mode, agentic debugging through the IDE debugger, and remote task execution, so JVM and Kotlin teams get an agent that uses IDE-grade project understanding.
- Model choice is broad: JetBrains-billed frontier models, bring-your-own-key for Anthropic, OpenAI, Google, and xAI at provider rates, and local runtimes (Ollama, LM Studio, LiteLLM), including a fully offline mode on high-end Macs.
- Third-party benchmark result: in the SWE-rebench cycle that JetBrains quoted at GA, Nebius's research lead named Junie the top model-harness with 61.6% resolved and 72.7% pass@5.
Disadvantages
- Two command-injection CVEs in 12 months (CVE-2025-59458, CVSS 8.3, and CVE-2026-41153 via malicious project files) show that opening untrusted repositories with an agent enabled is a real attack path.
- Usage is metered in AI credits (10 per 30 days on AI Pro, 35 on AI Ultimate, with top-ups), so agent-heavy work can exhaust quotas quickly and costs need per-team tracking.
- Organisation-level governance (AI policies, BYOK management, audit logs, SCIM) sits in the JetBrains Central console and its higher tiers, which was still in Early Access when checked.
Recommendation
Trial Junie in teams already standardized on IntelliJ IDEA, Android Studio, or other JetBrains IDEs, especially for JVM and Kotlin codebases where IDE refactoring and debugger access give it an edge over editor-agnostic agents. Before rollout, update Junie to a version that includes the CVE fixes, do not let the agent run terminal commands without confirmation in repositories from outside the organisation, and decide which models are allowed: JetBrains-billed models, BYOK against a company contract, or local models for code under NDA.
Set a credit budget per team and review consumption after the first month. If the JetBrains Central governance features are required (audit logs, SCIM, AI policies), confirm their availability and contract terms first. Teams that do not use JetBrains IDEs will get more from GitHub Copilot, Claude Code, or Cursor.
Sources
Doradztwo
Od radaru do wdrożenia
Ocenili Państwo technologie — pomagamy bezpiecznie i mierzalnie wdrożyć je w systemach produkcyjnych.
