IDE & Developer Tool Integration

Candela acts as a transparent LLM proxy — any tool that lets you configure a custom base_url or API endpoint can route through Candela for full observability.

How It Works

Most AI-powered IDEs and tools talk to LLMs via the OpenAI-compatible API (/v1/chat/completions, /v1/models). candela exposes this on port :1234 — just point your tool there.

IDE / Tool                   candela                          Provider
┌──────────┐   `:1234`   ┌──────────────────────┐   ADC/key   ┌──────────┐
│ Cline    │ ──────────▶ │ /v1/chat/completions │ ──────────▶ │ Gemini   │
│ Zed      │             │ /v1/models           │             │ OpenAI   │
│ IntelliJ │             │                      │             │ Ollama   │
│ OpenCode │             │  trace + cost        │             │ Claude   │
└──────────┘             └──────────────────────┘             └──────────┘

What you get automatically:

⏱️ Latency and TTFB per request
📊 Token counts (input / output / total)
💰 Cost calculation per request
🔗 Trace correlation (if the tool supports traceparent)
📋 Full request/response content capture

Generic Setup

For any tool that supports a custom OpenAI-compatible endpoint:

Setting	Value
Base URL	`http://localhost:1234/v1`
API Key	`candela` (or any string — candela handles real auth)
Model	Any model from `GET /v1/models`

# Verify it works
curl http://localhost:1234/v1/models | jq '.data[].id'

This returns all available models — local (Ollama) and cloud (Gemini, Claude) merged.

IDE-Specific Guides

VS Code Extension Display live cost, token usage, and budget status in your status bar.

OpenCode Terminal-native AI coding agent with official Candela integration.

Zed High-performance multiplayer editor with AI assistant.

Cline Autonomous AI coding agent for VS Code with official plugin support.

JetBrains Official plugin for IntelliJ IDEA, WebStorm, GoLand, and all JetBrains IDEs.

Claude Code Anthropic's terminal-based agentic coding tool.

Cursor / Windsurf AI-native code editors with custom API endpoint support.

Verifying the Connection

After configuring your IDE, check the management UI:

Open http://localhost:8181/_local/
Navigate to the Traces card
Send a prompt from your IDE
You should see the trace appear with model, tokens, cost, and duration