Problem

Currently, switching between models requires either:

• CLI commands (openclaw models set)
• The /model slash command
• Editing config

There's no way to say "use local for this" or "switch to the cheap model" in natural conversation and have the agent actually switch.

Use Case

Users with both a paid API model (e.g., Claude Opus) and a local model (e.g., Ollama) want to:

• Use the expensive model for complex tasks
• Switch to local/cheap model for simple queries to save money
• Do this fluidly in conversation without leaving the chat

Proposed Solution

Allow the agent to recognize model-switching intents and either:

1. Self-invoke a model switch (if given that capability)
2. Expose a tool the agent can call to switch models mid-session
3. Honor inline directives like [model:ollama/gpt-oss:20b] in user messages

Ideally this would be per-message or per-turn granularity, not just session-wide.

Current Workaround

The session_status tool has an optional model parameter for per-session override, but:

• The agent can't invoke it based on user intent ("use local for this")
• It's session-wide, not per-message
• Fallbacks only trigger on failure, not user choice

Environment

• OpenClaw 2026.1.30
• Primary: anthropic/claude-opus-4-5
• Local: ollama/gpt-oss:20b