User Guide

AI Assistant

Ask questions about your cloud costs in plain English — powered by OpenAI or Anthropic with 9 read-only tools.

How it works#

The AI Assistant uses a tool/function calling pattern. When you ask a question, the AI model decides which internal tools to call, executes them against your live data, and formats the results into a natural-language response.

  • ✓You type a question in the floating chat panel (bottom-right sparkle button)
  • ✓The AI selects from 9 read-only tools (cost exploration, recommendations, budgets, etc.)
  • ✓Tools query your database and Redis cache directly — no cloud API calls
  • ✓Results stream token-by-token via Server-Sent Events

9 built-in tools#

🔍

Explore Costs

Multi-dimensional cost analysis by service, resource, provider, tag.
💡

Recommendations

Active optimization recommendations with category/severity filtering.
💵

Budget Status

Current spend, forecast, and alert levels for all budgets.
🔔

Alert Summary

Counts by severity, type, and status.
📈

Cost Trend

Monthly history with month-over-month change.
☁️

Subscriptions

Connected subscriptions with provider and status.
💰

Savings

Realized savings with period and category filters.
🏷️

Tag Costs

Tag keys, values, and associated cost breakdowns.
📊

Forecast

Projected month-end costs for all subscriptions.
Tip
All tools respect the user's data scope. An analyst with Azure-only access will only get Azure data in AI responses.

Example questions#

"What are my top 3 cost drivers this month?"

"Show me idle resources in Azure"

"How is my overall budget tracking?"

"Compare AWS spending last month vs this month"

"Which subscriptions have the highest cost growth?"

"Are there any critical alerts I should look at?"

"How much have we saved from resolved recommendations?"

"What services are tagged with environment:production?"

Setup#

The Owner configures a single AI provider for the entire installation from Settings → AI Assistant:

🟢

OpenAI

GPT-4o, GPT-4 Turbo, and other OpenAI models. Default: gpt-4o with temperature 0.3.
🟣

Anthropic

Claude models (Sonnet, Opus, Haiku). Same tool-calling integration via Anthropic's function calling API.

You provide your own API key. CCO calls the AI provider directly from your Docker container — no proxy through our servers.

Demo mode#

In demo mode, the assistant works without an API key using keyword matching and pre-built response templates. This lets you evaluate the assistant's capabilities at zero cost before configuring an AI provider.

Same data, different brain
Demo mode uses the exact same tools and data formatting as live mode. Only the intent detection differs — keywords instead of an LLM.

Conversation management#

  • ✓Conversations persist in PostgreSQL per-user
  • ✓Switch between conversations via the dropdown in the chat header
  • ✓90-day auto-cleanup keeps storage lean
  • ✓Rate limit: 20 messages per minute per user
  • ✓Press Escape to close the panel without losing context