From 2736ab68c1c8335d2918216aaefe7f7cf21c8ba3 Mon Sep 17 00:00:00 2001
From: kirillk
Date: Mon, 31 Aug 2026 14:39:53 -0400
Subject: [PATCH] docs(jetbrains): rewrite plugin marketplace description
The plugin.xml description advertised inline autocomplete and browser
automation, neither of which the JetBrains plugin ships, and undersold
the actual differentiators (remote-dev-native split-mode architecture,
Agent Manager's per-task git worktrees, run/debug without new windows,
GitHub-native PR/diff visibility). Replace it with an accurate,
feature-complete listing covering all six pillars, sourced from the
plugin's feature/pitch doc.
---
.../src/main/resources/META-INF/plugin.xml | 81 ++++++++++++++++++-
1 file changed, 77 insertions(+), 4 deletions(-)
diff --git a/packages/kilo-jetbrains/src/main/resources/META-INF/plugin.xml b/packages/kilo-jetbrains/src/main/resources/META-INF/plugin.xml
index 96e28986525..7fb364cd6db 100644
--- a/packages/kilo-jetbrains/src/main/resources/META-INF/plugin.xml
+++ b/packages/kilo-jetbrains/src/main/resources/META-INF/plugin.xml
@@ -3,10 +3,83 @@
Kilo Code
+ Kilo Code is an open-source AI coding agent for JetBrains IDEs, built for both local
+ and remote development. Run multiple agents on different tasks at the same time — each in its
+ own isolated git worktree — without opening a new IDE window per task. The entire UI is native
+ Swing (no embedded browser, no Node.js), so it works identically locally and in JetBrains
+ Remote Development split mode over SSH, VMs, or Dev Containers.
+
+
+ Built for remote development
+
+ - No JCEF embedded browser anywhere in the UI — renders the same locally and through JetBrains Gateway.
+ - True split-mode architecture: the frontend runs on your machine, the agent server runs on the host next to your code.
+ - No Node.js dependency — a native
kilo binary is downloaded per OS and architecture.
+ - Dev Container aware: if a project is opened so the backend can't see container files, Kilo tells you to reopen via Gateway instead of failing silently.
+ - Resilient to flaky connections — long-running calls survive backend restarts and reconnects.
+
+
+ Agent Manager — many tasks, one IDE frame
+
+ - Every task gets its own git worktree: own branch, own working directory, own uncommitted changes.
+ - Start from a new worktree off a base branch, an existing branch, or directly from a GitHub PR URL.
+ - Continue in Worktree — fork an in-progress session's uncommitted changes into a fresh isolated worktree mid-conversation.
+ - One panel showing every task in flight with live PR status, diff stats, and run state.
+ - Optional per-worktree setup scripts (POSIX / PowerShell / CMD) that run automatically in their own terminal tab.
+
+
+ Run and debug without juggling windows
+
+ - Your existing Run Configurations (Gradle, external system, program-based) are transplanted onto each worktree with the correct working directory.
+ - Live run indicators per worktree, plus Build and Rebuild from the same popup.
+ - Need real breakpoint debugging? Open in New Frame opens one worktree as its own project window — reusing an open frame when there is one, and working correctly in remote sessions too.
+
+
+ GitHub-native workflow
+
+ - PR number, title, and state in the session header, clickable through to the browser.
+ - Diff versus the PR base branch with file and line counts, plus a full diff viewer.
+ - Uncommitted local changes tracked separately from the base diff.
+ - Import any existing PR into an isolated worktree to review, continue, or debug it.
+ - Guided setup when
git or the GitHub CLI is missing or unauthenticated, instead of a silent failure.
+
+
+ 600+ models, your keys or ours
+
+ - Claude, Gemini, GPT, Grok, GLM, and hundreds more from every major provider, switchable without leaving the IDE.
+ - Bring your own API keys or use Kilo credits.
+ - Pin a specific model to a specific custom agent so it never drifts between sessions.
+ - A separate fast, cheap model tier for high-frequency operations like subagent orchestration.
+ - Data-sensitivity badges flag models that may train on your prompts.
+ - Enterprise model access controls: org admins can block models or providers account-wide by training policy, retention policy, or datacenter location.
+
+
+ Powered by Kilo CLI
+
+ - The same open-source agent runtime behind the Kilo CLI, VS Code extension, and Kilo Cloud.
+ - One shared
kilo.jsonc config — set up agents, providers, and preferences once and use them on every surface.
+ - Built-in tools, MCP servers, semantic search, and LSP integration come from the shared runtime.
+ - One agent server per IDE instance serving your main workspace and every open worktree.
+
+
+ Also included
+
+ - Custom agents and modes with full create, edit, import, and export support.
+ - MCP server management with live connection status.
+ - Custom slash-command workflows backed by editable Markdown files.
+ - Skills management, including custom local and remote skill sources.
+ - Session checkpoints — revert to any prior message and redo forward, with per-file diff stats.
+ - Granular auto-approve permissions with per-tool Allow / Ask / Deny rules and queued approval requests.
+ - Session sharing via a public link.
+ - Smart context management: auto-compaction, stale tool-output pruning, and ignore patterns for large repos.
+ - Automatic settings, MCP config, and state migration from the older Roo/Kilo JetBrains plugin.
+
+
+
+ kilo.ai ·
+ Source on GitHub
+
]]>