Use Xantly with Roo Code

Roo Code is a Cline fork with custom per-mode models. Pair it with Xantly to route Code mode to Claude, Architect to GPT, and commit-message to fast speed-tier models.

Roo Code is a Cline fork with extra features for multi-agent orchestration, custom modes, and enterprise workflows. Same Xantly setup as Cline, with one additional pattern: per-mode model assignment.

Prerequisites

VS Code with the Roo Code extension installed
A Xantly API key, create one

Setup

Open Roo's settings (gear icon in the Roo sidebar).
Under API Provider, select OpenAI Compatible.
Configure:

Field	Value
Base URL	`https://api.xantly.com/v1`
API Key	`xantly_sk_...`
Model ID	`xantly/auto-quality`

Save.

Custom modes per model (Roo-specific win)

Roo's killer feature is custom modes, configure different personalities (Code, Architect, Reviewer, Debugger) with different models. Xantly pairs perfectly with this:

Mode	Recommended Model	Why
Code	`xantly/auto-quality`	Best code generation
Architect	`anthropic/claude-sonnet-4.6`	Strong planning/design reasoning
Reviewer	`openai/gpt-5.4`	Good at structured output + critique
Debugger	`xantly/auto-quality`	Needs tool calling reliability
Docs	`xantly/auto-value`	Cheap, fine quality
Commit Message	`xantly/auto-speed`	Fast, trivial task

Configure each mode's model in Roo's settings → Modes → pick mode → API Configuration → override model.

Verify tool-calling

Same as Cline. Load a project, ask:

Read package.json, check which deps are out of date, propose a patch.

Roo should fire tool calls (read_file, think, write_to_file) and land a patch. If tool-calling fails, swap the model to xantly/auto-quality or anthropic/claude-sonnet-4.6.

What you get

Multi-agent savings. Roo's multi-agent mode fans out work across 3-5 sub-agents. L2 semantic cache is disproportionately valuable here, sub-agents often re-ask similar questions.
Waterfall fallback. Mid-run provider outages don't break Roo's agent loop.
Mode-specific memory. Each mode can have its own X-Xantly-Memory-Mode setting, Code mode uses recall, Architect uses full, Commit Message uses none.
Cost by mode. Roo's dashboard + Xantly's per-key breakdown lets you see exactly which mode spent what.

Gotchas

Model ID must include provider prefix. anthropic/claude-sonnet-4.6, not claude-sonnet-4.6. Bare slugs that don't match the catalog get rejected.

Custom modes cache their model choice. After changing a mode's model, restart VS Code to force Roo to reload.

Roo is the #1 IDE Extension on OpenRouter by token volume. Heavy users can blow through T1 credit budgets fast, consider setting a daily USD cap on your Xantly API key (dashboard → API Keys → Edit → Daily Budget USD).

Next steps

Cline, the upstream Roo was forked from.
Kilo Code, another Cline fork, similar pattern.
Multi-Agent Orchestration, Xantly's optimizations for multi-agent workloads.