Our take

The default coding agent for JetBrains users — speaks IntelliJ's run configs, debugger, and test runner. Plan Mode with test-driven verification sets it apart.

What it is

JetBrains Junie is the vendor-native AI coding agent built by JetBrains. It installs as a plugin into IntelliJ IDEA, PyCharm, WebStorm, GoLand, CLion, RustRover, Rider, RubyMine, and DataSpell — all of which already have 10M+ developers as a combined base. Junie runs Claude (Sonnet 4 / Opus 4) in the cloud and offers two modes: Ask (read-only Q&A about code) and Plan (multi-step task with explicit verification at each step, including running tests). It integrates with existing JetBrains run configurations, the debugger, and the test runner — a notable difference from generic IDE plugins.

Best for

JetBrains users who want a vendor-native coding agent that knows their IDE
Multi-step feature implementation with test-driven verification
Polyglot teams (Java, Kotlin, Python, Go, Rust, JS/TS) on one JetBrains stack
Enterprise teams that already pay JetBrains for IDE licenses

Pros

Plan Mode with test-driven verification — not just LLM-generated code, but proof of passing tests
Vendor-native: talks to your existing run configs, debugger, test runner
10M+ JetBrains IDE developer base — instant onboarding
Multi-language across the full JetBrains fleet

Cons

Won't run on VS Code / Cursor / Vim — JetBrains only
AI free tier is steeply limited (6 Junie chats/month)
Cloud-only runs (no fully-local execution path for sensitive code)

Quick start

Open IntelliJ IDEA / PyCharm / WebStorm / GoLand (2024.3 or newer) → Tools menu → install the "Junie" plugin.
Sign in with your JetBrains account — free tier includes 6 Junie chat interactions per month.
AI Pro $10/mo or AI Pro+ $20/mo unlocks unlimited Junie plus inline completion across all JetBrains IDEs.
In the editor: highlight code → right-click → "Ask Junie". Or open the Junie tool window on the right side for a chat panel.
For multi-step tasks: type /plan or click "Plan Mode" → Junie lists steps → approve → it runs them with test-driven verification at each.
Junie understands your Git state, existing run configurations, and tests — it uses what you already have, not generic scaffolding.

Sample input / output

Input

Project: PyCharm 2025.3, Python 3.12, FastAPI backend

In Junie Plan Mode:
"Add a /v1/billing endpoint that:
  - Accepts POST with customer_id (UUID) + amount_usd (Decimal)
  - Loads customer from Postgres (sqlalchemy async session)
  - Strips a 2.9% + 30¢ fee, deposits the net to Stripe via existing stripe_service.create_transfer
  - Returns 201 with txn_id, or 4xx with error code
Write the test cases that cover happy path, insufficient-funds, and stripe-timeout. Run pytest at the end and ensure all pass."

Output

Junie execution log (3m 42s on AI Pro tier):
  Step 1 — Plan: 6 sub-tasks identified; showed diff preview.
  Step 2 — Created app/api/v1/billing.py (62 lines, type-safe with Decimal handling).
  Step 3 — Added SQLAlchemy async loader pattern.
  Step 4 — Reused stripe_service.create_transfer (no duplication).
  Step 5 — tests/v1/test_billing.py (8 tests, 3 categories).
  Step 6 — Ran pytest:
    8 passed in 1.7s
    Coverage of billing.py: 96%
Final summary: "Implemented /v1/billing with full test pass. No existing tests were broken (12/12 in unrelated tests remain green)."
No auto-commit (Junie pauses for git commit approval by default).

Benchmarks

Plans AI Free (6 chats/mo); AI Pro $10/mo; AI Pro+ $20/mo — JetBrains pricing 2026

IDEs supported IntelliJ, PyCharm, WebStorm, GoLand, CLion, RustRover, Rider, RubyMine, DataSpell — JetBrains Junie docs

Mode Plan mode + Ask mode (read-only Q&A) — JetBrains Junie docs

Engines Claude Sonnet 4 / Opus 4 (Anthropic) — JetBrains Junie

JetBrains IDE user base 10M+ developers — JetBrains 2025 disclosure

SWE-bench Verified Top 10 tier (Junie AI Pro) — JetBrains 2026 update

Pricing

AI Free (6 Junie chats/month on JetBrains account); AI Pro $10/mo or AI Pro+ $20/mo (Junie + inline completion + JetBrains AI Premium features).

Underlying models

Claude Sonnet 4Claude Opus 4

Should you pick this?

Pick it if You live in JetBrains IDEs (IntelliJ / PyCharm / WebStorm / GoLand), hate IDE-switching, and want a vendor-native coding agent that talks to your existing run configs, debugger, and test runner.

Skip it if Your editor is VS Code / Cursor / Vim — Junie doesn't run there. For open-source terminal-coding agents, OpenCode or Aider.

JetBrains Junie