Celebrimbot


An autonomous AI coding agent embedded directly into your JetBrains IDE.

  View on GitHub

What is Celebrimbot?

Celebrimbot is an IntelliJ Platform plugin that brings a full multi-agent AI system into your IDE. It can read your project files, write and modify code, execute terminal commands, search the web, inspect git history, and hold natural conversations — all from a single chat panel anchored to your IDE window.

Unlike simple autocomplete tools, Celebrimbot operates as an agentic loop: it routes requests intelligently, executes tasks locally when possible, escalates to cloud planning only when needed, and retries failures automatically — without ever leaving your editor.

Celebrimbot chat panel

How It Works

Celebrimbot uses a six-layer architecture — each layer named after a character from Tolkien's legendarium:

User Message | v +-------------+ | Gandalf | ← Router: classifies CHAT / EASY_TASK / COMPLEX_TASK +-------------+ | | | CHAT EASY_TASK COMPLEX_TASK | | | Galadriel Aragorn Elrond (chat reply) easy planner enriches context | | Samwise / Frodo / Celebrimbor (Cloud) Legolas & Gimli master planner execute tasks | | Samwise / Frodo / Treebeard Legolas & Gimli Ent Reviewer execute tasks reflection loop | | Treebeard Bilbo Ent Reviewer summary reflection loop | Bilbo summary

The Fellowship

Character Role Default Provider
🧙 GandalfRouter: CHAT / EASY_TASK / COMPLEX_TASKLocal Qwen
🧝 GaladrielConversational AI, addresses user as "Mellow"Local Qwen
⚔️ AragornEasy-task plannerLocal Qwen
📜 ElrondComplex pre-planner, enriches contextLocal Qwen
💎 CelebrimborMaster planner, produces atomic tasksAlibaba → Gemini → Local
🌿 SamwiseMechanical worker (delete, terminal, git)Local Qwen
🍃 FrodoCode worker (write_code)Local Qwen → Alibaba
🏹 Legolas & GimliExpert worker duo for complex tasksAlibaba → Gemini → Local
🌳 TreebeardEnt Reviewer, up to 2 reflection cyclesAlibaba → Local
📖 BilboSession chroniclerLocal Qwen

Key Features

🖥️ Offline-first

Embedded Qwen 2.5 Coder 1.5B runs entirely on your machine via java-llama.cpp. No internet required for most operations. The model (~1.2 GB) is downloaded automatically on first use.

☁️ Multi-provider fallback

Each Fellowship character has its own configurable provider chain: Local Qwen → Alibaba Cloud (Qwen Plus) → Google Gemini → Amazon Q Developer → Kiro (AWS). If one fails, the next kicks in automatically.

🔧 Autonomous code editing

Reads files, applies changes directly in the editor via PSI/VFS. Creates, edits, and deletes files via natural language. Runs shell commands from within the IDE.

🏛️ Council's Review (Validation Loop)

After every write_code, the build command runs automatically. Up to 3 self-correction cycles with error feedback before escalating to a stronger model.

🌿 Treebeard (Ent Reviewer)

A critic agent between Workers and Bilbo. Reviews completed work against the original request and triggers up to 2 re-planning cycles if the result is incomplete. Never hasty, never satisfied with placeholder code.

🔮 Elrond's Palantír (BM25 Index)

Lightweight local semantic index. COMPLEX_TASK planning retrieves the top-8 relevant files instead of sending the full project skeleton to the planner.

🕯️ MCP Bridge (Beacons of Gondor)

MCP-compliant JSON-RPC 2.0 server over HTTP (POST /mcp) and Stdio (celebrimbot mcp-stdio) for Claude Desktop and other MCP hosts.

↩️ Shadow Log (Auto-Undo)

Every file is backed up before being written or deleted. celebrimbot undo restores the last session from .celebrimbot/shadow_log/.


AI Providers

All five providers are selectable per-character in Settings → Celebrimbot → Fellowship AI Configuration.

ProviderAuthInternet
🖥️ Local (embedded Qwen)None❌ Fully offline
🟠 Alibaba Qwen CloudAPI key
🔵 Google GeminiAPI key
🟡 Amazon Q DeveloperSSO token from ~/.aws/sso/cache/
🟢 Kiro (AWS)SSO token from ~/.aws/sso/cache/kiro-auth-token.json

Provider Benchmark

Evaluated across 12 provider configurations and 8 test cases using Claude Sonnet 4.6 (via Amazon Q Developer) as the judge.

RankConfigurationAvg ScorePass Rate
🥇Claude Sonnet 4.6 planners + Qwen 2.5 Coder 7B workers9.0/10100%
🥈Claude Sonnet 4.6 planners + Phi-3.5 Mini workers9.0/10100%
🥉Claude Sonnet 4.6 planners + Llama 3.1 8B workers8.9/10100%
#4All Claude Sonnet 4.6 (baseline)8.6/1087.5%
#7All Local — Qwen 2.5 Coder 7B7.8/1087.5%
#11All Local — Qwen 2.5 Coder 1.5B4.9/1037.5%

Key finding: Claude Sonnet 4.6 for planners + a 7B+ local model for workers outperforms using Claude for everything (9.0 vs 8.6) while minimising cloud API calls.


Installation

From JetBrains Marketplace:
SettingsPluginsMarketplace → search CelebrimbotInstall

Requirements: IntelliJ IDEA 2025.2+ · Java 21+ · ~1.2 GB disk for local model