llmcloud.ai
Apps · Coding agents

Roo Code

Multi-mode coding agent with architect and debug personas.

Tokens / 7d
268B
Routing directive
auto:coding
On our metal
22%
Licence
Open source

Roo Code splits work across modes — architect, code, debug, ask — and each mode can point at a different model. That maps cleanly onto per-mode routing directives instead of one hardcoded model.

How it connects

  • Separate profiles per mode: auto:reasoning for architect, auto:coding for code.
  • Streaming enabled; tool calls validated against strict JSON schema.
  • Fallback chain configured so a provider outage does not kill a long task.

Traffic pattern

Bursty. Architect turns are small and expensive, code turns are large and cheap. Mode-level routing lowers spend ~30% versus one model everywhere.

claude-sonnet-4.5gpt-5.5qwen3-coder-480b

Point it at the gateway

// Roo Code profile: architect
{ "baseUrl": "https://api.llmcloud.ai/v1", "model": "auto:reasoning" }
// Roo Code profile: code
{ "baseUrl": "https://api.llmcloud.ai/v1", "model": "auto:coding" }