Fast and Focus Modes

When interacting with Olly, you can choose how Olly answers your questions by selecting a mode before sending your prompt. Each mode is optimized for a different type of task, balancing speed and depth of reasoning.

You can change the mode at any time while chatting with Olly.

Supported modes

Olly currently supports two modes:

• Fast mode: optimized for speed and simplicity
• Focus mode (default): optimized for deeper reasoning and investigation

You can choose a mode from the input bar each time you send a prompt, and it takes effect immediately.

Switching modes is possible at any time during the conversation. This allows you to start fast and switch to deeper analysis only when needed.

Fast mode

Fast mode is designed for quick responses and lightweight tasks.

• Model: GPT-5 mini (when GPT is selected) or Haiku 4.5 (when Claude is selected)
• Best for: simple questions, quick lookups, basic data queries
• Very fast response time with simplified reasoning optimized for speed

Use Fast mode when you need a quick answer to simple questions, fast lookups, or basic data queries where speed matters more than deep reasoning.

Examples:

• "Show error rate for checkout service in the last hour"
• "What is the current CPU usage of node-3?"
• "List alerts fired in the last 10 minutes"

Focus mode (default)

Focus mode is Olly's default and most powerful mode.

• Model: GPT-5.1 (when GPT is selected) or Sonnet 4.5 (when Claude is selected)
• Best for: complex investigations, root-cause analysis, exploratory observability questions
• Deeper analysis and reasoning
• Takes longer to respond than Fast mode
• Uses multiple specialized sub-agents, each acting as an expert in a specific domain (for example, logs agent)

Choose Focus mode for in-depth analysis tasks, such as investigating issues, uncovering root causes, or exploring observability data that requires a deeper level of analysis.

Examples:

• "Why did latency spike after the last deployment?"
• "Investigate the root cause of intermittent 5xx errors"
• "Correlate error logs with recent infrastructure changes"

When to use each mode