Use Xantly with JetBrains AI Assistant + Junie

JetBrains AI Assistant (IntelliJ, PyCharm, WebStorm, etc.) and agent-mode Junie support custom OpenAI-compatible providers since 2024.3. Drop in Xantly for routing on chat, inline edit, and Junie agent runs.

JetBrains AI Assistant (IntelliJ, PyCharm, WebStorm, GoLand, Rider, etc.) and the agent-mode Junie both support custom OpenAI-compatible providers since 2024.3. Drop in Xantly once and every chat, inline edit, code-complete assist, and Junie agent run routes through smart routing + semantic cache + memory.

Prerequisites

A JetBrains IDE version 2024.3+ (IntelliJ IDEA, PyCharm, WebStorm, etc.)
AI Assistant plugin installed (bundled in most 2024.3+ versions)
A Xantly API key, create one

Setup

Open Settings → Tools → AI Assistant → Providers & API Keys.
Click Add Provider → OpenAI (compatible).
Fill in:

Field	Value
Display name	`Xantly`
Base URL	`https://api.xantly.com/v1`
API Key	`xantly_sk_...`
Available models	`xantly/auto-quality`, `xantly/auto-value`, `anthropic/claude-sonnet-4.6`, `openai/gpt-5.4` (comma-separated)

Click Test connection → you should see a green checkmark.
Under Models → set the default to xantly/auto-quality for chat, and xantly/auto-value for inline edits.
Apply + restart the IDE.

Using AI Assistant

Alt+Enter on a selection → AI actions (Explain, Refactor, Write docs).
Ctrl+\ → inline AI completion.
AI chat panel → conversational mode.

All three route through your Xantly provider now.

Using Junie (agent mode)

Junie is JetBrains' autonomous agent. With the Xantly provider active:

Open the Junie panel (sidebar) or Cmd+J.
Give it a task: "Add tests for the auth module."
Junie plans → edits files → runs tests, all via Xantly.

Junie's multi-step agent runs are cache-friendly, similar sub-prompts across steps hit L2 cache.

Model choice

Model ID	When
`xantly/auto-quality`	Chat + Junie, production default.
`xantly/auto-value`	Inline edits (high frequency).
`anthropic/claude-sonnet-4.6`	Long-context refactors, multi-file agent work.
`openai/gpt-5.4`	Structured output, schema-tight edits.
`xantly/auto-speed`	Quick inline completions / commit messages.

Verify

Open the AI chat panel, send:

Reply with just the word "pong".

Check your Xantly dashboard, call logged with routed-to model, cache status, USD cost.

What you get

Full IDE surface. Chat, inline actions, Junie agent, commit-message generation, all through Xantly.
Waterfall fallback. Junie's long agent runs don't crash on mid-run provider outages.
Semantic cache. Highly cache-friendly for repetitive "write a test for X" / "explain this function" patterns.
Memory. JetBrains passes project + file context with each call; Xantly's L3 memory layers cross-session context on top.
Cost per IDE action logged.

Gotchas

JetBrains AI Assistant used to require their Enterprise plan for custom providers. As of 2024.3 and the free-tier reorg, the custom-provider feature is available on all plans. If your IDE shows it as locked, upgrade to the latest version.

Model availability. The provider config requires you to list available models. If you add new Xantly models to your catalog later, remember to add them to the AI Assistant config too.

Junie context strategy. Junie sends aggressive repo context with each call. Xantly's auto-quality model can handle it; pure speed-tier models may truncate. Stick to quality/value tiers for Junie.

Base URL must include /v1. Same rule as other OpenAI-compatible endpoints.

Plugin vs bundled AI Assistant. In newer JetBrains versions the AI Assistant is bundled; if your version has it as a separate plugin, make sure it's updated.

Next steps

Cline, VS Code agent mode, same pattern.
Kilo Code, Kilo ships a JetBrains plugin with per-mode models.
Continue.dev, also ships a JetBrains plugin with role-split config.
Bring Your Own Key, route JetBrains AI through your own provider credits.