Junie

Name: Junie
Author: JetBrains

Junie is a coding agent from JetBrains that plans, edits code, runs commands and tests, and reviews pull requests. It runs as a plugin in JetBrains IDEs and Android Studio, as a standalone CLI (which can also act as an ACP server), and in GitHub Actions and GitLab pipelines. It left beta on 17 June 2026. Usage is paid with JetBrains AI credits, with your own API keys, or with local models, including a free on-device mode (Junie Local) announced in August 2026.

junie.jetbrains.com Source repository Docs Pricing Add to compare

What reviewers and users praise

Model choice: users like switching between Google, OpenAI and Anthropic models, or using BYOK and local models (HN, Reddit, Éderson Fernandes)
Tight integration with JetBrains IDEs for developers already on that toolchain (Mike Lundahl, AltAI, HN)
Some users report low monthly cost when paired with cheaper models such as Gemini Flash (HN)
Mid-table Terminal-Bench 2.0 result (rank 18 of 43 harnesses)

What they criticise

Credit-based pricing: many users say credits run out quickly and top-ups add up (Reddit, HN, AltAI)
Seen by many users as less capable or less polished than Claude Code or Codex (Reddit, HN)
Reported as slower than competing agents (Mike Lundahl, AltAI)
Local-model setup reported as unreliable by some users; confusion over how Junie, AI Assistant and Air relate (Reddit, HN)

Rating certificate · AW/26-09/56426 Sept 2026

55of 100

Junie

JetBrains · Coding agents

No. 15 of 16 in categoryhigh confidenceTrust B

Adoption37

Experts78

Crowd44

Experts vs the crowdcrowd cooler by 34

Experts78

Crowd44

Price	Free, paid from $8.33
Use it via	CLI, IDE extension, Web app
Runs on	macos, linux, windows
Licence	proprietary
Updated	Evidence collected 26 Sept 2026

How the rating was worked out

Weights for Coding agents: adoption 35%, experts 40%, crowd 25%. Pillars without enough evidence are left out and the others re-weighted. The method.

Adoption & momentumweight 35%	37	GitHub stars `460` → 20 · npm weekly downloads `633` → 0 · `32` releases in 90 days → 100 reach × 0.7 + momentum × 0.3
Expertsindependent reviews · weight 40%	78	6 independent reviews (5 positive · 1 mixed · 0 negative), averaged toward 50 for small samples. Benchmarks, shown not scored: Terminal-Bench 2.0 No. 18 of 43 (Junie CLI (JetBrains) + Multiple models)
Crowdcommunity sample · weight 25%	44	18 dated posts from hackernews, reddit: 5 positive · 5 mixed · 8 negative (positive + ½ mixed) ÷ n, averaged toward 50 for small samples
Rating	55	high confidence (3 of 3 pillars, 18 community posts).

Evidence checks. 1 review from tool directories, AI-written pages or unnamed authors not counted. 2 posts over the three-per-thread limit not counted. 1 item could not be checked at the source (the page could not be fetched, or the quote is from a comment under the linked post); kept and marked “unverified”. How evidence is checked.

Editorial adjustment. Not independent: the author received free Junie access from JetBrains, as disclosed in the post (gapfill-1).

Trust & safety

Seven dimensions graded A–D from documented facts. Overall: B.

Permission model

Brave mode has three documented levels (Off asks approval for every potentially sensitive action outside the allowlist; Auto approves commands judged safe; On approves everything), plus a per-user Action Allowlist. The CLI sandbox (writes limited to project/git/temp/tool caches, ~/.ssh reads blocked) is off by default, macOS/Linux only, and documented as available in development, nightly and experimental builds. source

Data access scope

Junie respects a project .aiignore file and asks for explicit approval before viewing or editing listed files; command allowlist is documented; opt-in sandbox limits writes to project directories. YouTrack has user reports of .aiignore regressions (LLM-23365, LLM-27523). source

Data storage

Agent runs locally in the IDE or terminal; can be used with BYOK providers or local models, and Junie Local runs fully on-device with no account. Default JetBrains AI credit usage routes prompts through JetBrains' cloud to model providers; AI Service Terms say data may be stored for a limited period for security monitoring. source

Data retention & training

JetBrains AI Service Terms (v2.1): JetBrains and its AI subcontractors will not use inputs/outputs to train language models. Separately, detailed code-related data (full LLM inputs/outputs) may be used to train JetBrains' own models with a 1-year retention; it is opt-in for commercial and organisational licences, but for non-commercial individual licences sharing is on by default with an opt-out (Product Data Collection notice v1.6; The Register, Oct 2025). Graded C for the opt-out cohort; commercial users alone would be B. source

Incident history (24 months)

Three Junie CVEs in NVD within 24 months, all fixed in named versions: one medium information disclosure and two command/code-execution issues rated up to 9.8 CRITICAL. Graded C (serious CVEs, resolved); a strict reading of 'repeated' could give D. source

Compliance

JetBrains holds a SOC 2 Type II report (new 2024/25 report published 20 Nov 2025, available via the Trust Center); not ISO 27001 certified. Junie homepage states SOC 2 certification. source

Transparency

The experts

6 named independent reviews found; dated reviews from the last 12 months are scored.

positive
For me, Junie has been great and I use it every day.Mike Lundahl (mikelundahl.com) · 11 Dec 2025 · mikelundahl.com
positive
If you're a developer looking for an AI coding agent that's flexible, integrable, and doesn't lock you into a single ecosystem, Junie CLI deserves your attention.Éderson Fernandes (blog.edersonfernandes.com.br) · 23 Apr 2026 · blog.edersonfernandes.com.br
positive
In conclusion, Junie proves to be an exceptional tool for conducting code reviews.Miguel Teheran (DEV Community) · 21 Oct 2025 · dev.to
positive
Junie rarely surprises me — and that’s high praise.Alessandro Giacomella (Medium) · 21 Nov 2025 · unverified · medium.com
positive
It didn't just chat—it read my file structure, summarized my UI architecture, and even proposed a "Quick Win" refactor for my Home Screen.Somasundaram Mahesh (YouTube) · 27 Nov 2025 · youtube.com
mixed
the credit system burns through your monthly quota in days once Junie gets involvedJovan Chan (AICoderScope, republished on DEV Community) · 1 Jun 2026 · dev.to

The crowd

18 coded posts from hackernews, reddit.

5 positive5 mixed8 negative

positive
Junie is good. Needs a few UI tweaks, but the code it generates is state of the art.HN: Google Antigravity · 19 Nov 2025 · news.ycombinator.com
positive
I really enjoy Junie, I find it working better out of the box than Claude code.HN: Claude Code gets native LSP support · 22 Dec 2025 · news.ycombinator.com
negative
Mainly, they're pushing Junie and it just isn't that good or compelling, when faced off against the competition.HN: Show HN: Sweep, open-weights 1.5B model · 22 Jan 2026 · news.ycombinator.com
negative
it also has frequent agent crashes (bye tokens) and sometimes is unreachableHN: Anthropic Claude Max $200/mo thread · 4 Feb 2026 · news.ycombinator.com
positive
IDK how Junie does it, but I spend less than $50 USD per month and I'm on it 30 hours per week.HN: SpaceX / Cursor thread · 22 Apr 2026 · news.ycombinator.com
mixed
While Junie is nowhere near as good as Claude Code, it is way ahead of the current Google tooling.HN: Our eighth generation TPUs · 22 Apr 2026 · news.ycombinator.com
mixed
but it burns thru credits leaving me topping up $5 every dayHN: Gemini 3.5 Flash · 20 May 2026 · news.ycombinator.com
negative
I use Junie at work and I have to say I'm not amazed by it.HN: JetBrains Air · 22 Sept 2026 · news.ycombinator.com

Pricing

As published, checked 26 Sept 2026.

Plan	Price	Notes
Free	Free	5 AI credits included; BYOK at provider rates with zero markup; supports locally running models
AI Pro	$8.33 / monthly	Per user; 10 AI credits per 30 days; anytime top-ups. Price shown is the annual-billing equivalent (buy link is the yearly SKU)
AI Ultimate	$25 / monthly	Per user; 35 AI credits per 30 days; 'Recommended for Junie'; anytime top-ups. Annual-billing equivalent; an HN user reports $30 when paid monthly
Junie Local	Free	Runs entirely on an Apple Silicon Mac with a local model; no registration, subscription or credits

Recent changes

17 Jun 2026
Junie leaves beta; adds plan mode, agentic debugging, remote control and code-review integration source
24 Aug 2026
Junie Local: free fully on-device mode for Apple Silicon Macs, no account or credits source
7 Sept 2026
CLI release 26.9.7: streaming responses, /branch conversation branching, live session cost display source
14 Sept 2026
CLI release 26.9.14: 'Apply suggestion' for the /review command source
22 Sept 2026
CLI release 26.9.22: adds Grok 4.7, Claude Opus 5.5 and GPT-6 Sol/Luna source

Ranked alongside

Other ranked entries in Coding agents.

Compare with the top two

OpenCodeAnomaly75

Claude CodeAnthropic73

Zed (Agent Panel)Zed Industries73

GitHub CopilotGitHub (Microsoft)72

Codex (CLI and app)OpenAI71

Sources

Every figure above traces to one of these, observed up to 26 Sept 2026.