OpenCode

OpenCode is a terminal-native AI coding agent with MCP support. It uses the OpenAI-compatible API and can be pointed at Candela with minimal config.

Configuration

Edit your OpenCode config (~/.config/opencode/config.json or project-level .opencode.json):

{
  "provider": {
    "name": "openai",
    "apiBase": "http://localhost:1234/v1",
    "apiKey": "candela",
    "model": "gemini-3.5-pro"
  }
}

Using Cloud Models

For cloud models routed through candela in Solo + Cloud mode:

{
  "provider": {
    "name": "openai",
    "apiBase": "http://localhost:1234/v1",
    "apiKey": "candela",
    "model": "gemini-3.5-pro"
  }
}

candela handles Vertex AI authentication via ADC — no API key needed.

Using Local Models

For Ollama models running locally:

{
  "provider": {
    "name": "openai",
    "apiBase": "http://localhost:1234/v1",
    "apiKey": "candela",
    "model": "llama3.2:3b"
  }
}

Verifying

# Start candela
candela start

# In another terminal, start OpenCode
opencode

# Check traces at http://localhost:8181/_local/

Every prompt and response flows through Candela with full token counting and cost tracking.

---

Official Candela Plugin (v0.8.0)

The official @candelahq/opencode plugin hooks directly into OpenCode’s execution lifecycle to provide real-time cost visibility, budget guardrails, spending intelligence, and 13 slash commands — all inside your terminal.

➡️ View on npm · 📦 GitHub

Features

Category	What you get
💰 Real-Time Cost Tracking	Per-response cost deltas, session totals, and 24h spend in status bar and sidebar
📊 Budget Monitoring	Threshold toasts at 80/90/100%, budget pacing forecast, reset countdown
🔀 Smart Model Routing	Opt-in suggestions to swap to cheaper models when budget is tight
📏 Context Window Gauge	Token usage tracking with compaction warnings at 80%+
🎯 Daily Cost Goals	Set spending targets, track progress with visual pacing
🛑 Session Cost Alerts	Per-session cost tracking with 80%/100% warning toasts
📈 Cost Forecasting	Extrapolate session cost based on current call rate
🔇 Quiet Mode	Suppress info toasts, keep warnings and errors
🏷️ Session Tagging	Tag sessions by activity (auto-detects git branch)
📂 Repo Attribution	Auto-tracks costs per git repository
📜 Session History	Browse past sessions with cost, duration, and tool usage
⏰ Time-of-Day Patterns	Discover when you spend the most
🛠️ Tool Cost Breakdown	See which tools cost the most per call
📝 Git Commit Annotation	Prepare cost metadata for commit messages
📦 Export	JSON + CSV export of session data
🗄️ Local Analytics	JSONL event log with 90-day auto-rotation and 10MB cap

Intelligence Layer

Beyond basic tracking, the plugin builds a spending intelligence profile:

• Cost Streaks — Track consecutive under-budget days
• Anomaly Detection — Alert when session cost is 2x+ your average
• Budget Pacing — Estimate budget exhaustion time from hourly burn rate
• Model Efficiency — Score models by cost-per-call vs effectiveness
• Weekly Digest — Week-over-week spending comparison
• Time Patterns — Morning vs afternoon vs evening vs night cost analysis

Installation

npm install @candelahq/opencode

Add to your OpenCode config (~/.config/opencode/config.json or .opencode.json):

{
  "plugins": ["@candelahq/opencode"]
}

Zero config

The plugin works out of the box when Candela is running on localhost:8181. No additional configuration needed.

Slash Commands

Command	Aliases	Description
/cost	/spend	Session cost + 24h total breakdown
/budget	/remaining	Budget remaining, grants, reset time
/models	—	Top models by spend and call count
/dashboard	/dash	Open Candela web dashboard
/export	/dump	Export session data to JSON + CSV
/goal	—	Set or view daily cost goal
/quiet	/shh	Toggle quiet mode
/tag	/label	Tag session for cost attribution
/cap	—	Set per-session cost cap
/history	/sessions	Browse recent sessions
/patterns	/when	Time-of-day cost analysis
/annotate	/commit-cost	Git commit cost metadata
/tools	/tool-cost	Tool cost breakdown

Sidebar Dashboard

The plugin renders a live sidebar with real-time metrics:

📊 $4.20 · 24h
🗄️ Cache hit rate: 72%
🏷️ feat/context-gauge
⚡ Session: $1.80 · 12 calls
📈 Forecast: ~$3.30 if 10 more calls
📏 Context: 45k tokens 🟩 ~35%
🎯 Goal: $4.20/$20 🟩 21%
⏱️ Budget exhausted by 4:30 PM
  claude-sonnet: $2.10 (8 calls)
  gpt-4o: $1.30 (4 calls)

Environment Variables

Variable	Type	Default	Description
CANDELA_PROXY_URL	String	http://localhost:8181	Candela proxy URL
CANDELA_CONFIG	String	—	Path to Candela config YAML (for port discovery)
CANDELA_SMART_ROUTING	Boolean	false	Enable cost-conscious model routing
CANDELA_ROUTING_THRESHOLD	Float (0–1)	0.7	Budget fraction to trigger routing
CANDELA_ROUTING_SAVINGS_THRESHOLD	Float (0–1)	0.5	Min savings to suggest model swap
CANDELA_DAILY_GOAL	Number (USD)	—	Daily spending target
CANDELA_QUIET	Boolean	false	Suppress info-level toasts
CANDELA_SESSION_CAP	Number (USD)	—	Per-session cost alert threshold

Settings File

Persistent settings at ~/.config/opencode/candela-settings.json. Resolution priority: env vars > settings file > defaults.

---

Troubleshooting

OpenCode Hangs, Shows Stale Models, or Behaves Unexpectedly

If OpenCode hangs, shows stale models, or behaves unexpectedly after config changes, delete the OpenCode database:

rm -rf ~/.local/share/opencode/opencode.db*

Then restart OpenCode. This clears cached state (model lists, provider connections, etc.) and forces a fresh sync.

Model Not Found Errors

Ensure the model IDs in your .opencode.json match what the provider expects. Model versions get retired (e.g. codestral-2501 → codestral-2). Check the Candela server logs for the exact model ID being sent.

no pricing configured for model X

The model needs a pricing entry in Candela. Contact your admin or add it to the cost calculator.

Connection Refused on localhost:8181

Make sure the Candela server is running (candela server or the Cloud Run instance).

---

Mission Orchestration Plugin

For multi-step autonomous workflows, install the companion @candelahq/missions plugin.

This plugin provides structured mission orchestration, allowing the agent to plan, execute, and validate complex goals across multiple child sessions.

Features

• Breaks down large goals into discrete milestones (mission_plan).
• Dispatches isolated child sessions for focused work (mission_next).
• Validates milestone completion with test commands (mission_validate).
• Tracks progress across sessions and restarts.
• Exposes 5 core tools for the agent: mission_plan, mission_next, mission_validate, mission_status, mission_cancel.

Installation

Install the package in your project:

npm install @candelahq/missions

Then add it to your .opencode.json:

{
  "plugins": [
    "@candelahq/opencode",
    "@candelahq/missions"
  ]
}

Integration with @candelahq/opencode

When run alongside @candelahq/opencode, child sessions spawned by the missions plugin automatically inject the CANDELA_MISSION_ID environment variable. This translates to the X-Mission-Id HTTP header in API requests, enabling grouped cost tracking and budget attribution for the entire multi-step mission.

➡️ View on GitHub