Use Xantly with Aide (codestory.ai)

Aide from codestory.ai uses a local Rust sidecar for LLM dispatch. Its provider-config block accepts OpenAI-compatible endpoints, point it at Xantly for routing on chat, Cmd+K, and agent mode.

Aide from codestory.ai is an AI-native editor with a sidecar architecture, a local Rust binary handles LLM calls and tool orchestration while the editor UI stays snappy. Aide exposes a provider-config block that accepts OpenAI-compatible endpoints. Point it at Xantly and every sidecar-mediated call routes through smart routing + cache + memory.

Prerequisites

Aide installed, download from aide.dev
A Xantly API key, create one

Setup

Open Aide → Settings (Cmd+, / Ctrl+,).
Search for AI Providers → add a new OpenAI Compatible provider.
Configure:

Field	Value
Name	`Xantly`
Base URL	`https://api.xantly.com/v1`
API Key	`xantly_sk_...`
Available Models	`xantly/auto-quality`, `xantly/auto-value`, `anthropic/claude-sonnet-4.6`, `openai/gpt-5.4`

Save. Under Default Model → select xantly/auto-quality.
Reload the window.

Setup via sidecar config

Aide's sidecar reads a config file at ~/.aide/sidecar.toml:

[providers.xantly]
kind = "openai_compatible"
base_url = "https://api.xantly.com/v1"
api_key = "xantly_sk_..."

[defaults]
model = "xantly/auto-quality"
provider = "xantly"

Restart the sidecar (aide-sidecar restart or restart Aide) for changes to apply.

Using Aide

Cmd+K → inline edit.
Cmd+I → agent panel.
/ commands in the chat → references files, commits, diffs.

All three surfaces route through Xantly.

Agent mode

Aide's agent panel supports Plan / Act / Verify cycles, similar to Cline but with codestory's specialized Sota (state-of-the-art task agent) model handoffs.

Use xantly/auto-quality for the main agent model and anthropic/claude-sonnet-4.6 for tool-heavy sub-tasks.

Model choice

Model ID	When
`xantly/auto-quality`	Main agent, Cmd+K complex edits.
`xantly/auto-value`	High-volume chat / reviews.
`anthropic/claude-sonnet-4.6`	Long-context agent runs, tool calls.
`openai/gpt-5.4`	Schema-tight edits.
`xantly/auto-speed`	Inline quick completions.

Verify

In the Aide chat panel:

Reply with just the word "pong".

Check your Xantly dashboard, call logged with routed-to model + cost.

What you get

Sidecar performance + Xantly routing. Aide's Rust sidecar dispatches fast; Xantly adds 12ms p50 overhead for smart routing.
Waterfall fallback through the sidecar, mid-agent provider outages don't derail Aide's Act loop.
Semantic cache on Cmd+K repeats.
Memory across Aide projects via Xantly's L3 cascade.
Cost visibility: the sidecar's request logs align with your Xantly dashboard entries.

Gotchas

Sidecar version matters. Older Aide builds ship an older sidecar that might not support custom openai_compatible providers. Update to the latest Aide release before filing issues.

base_url with /v1.

The sidecar runs as a separate process. If you change the config via Settings, give it a few seconds to pick up the new values, or manually restart with aide-sidecar restart.

Proprietary codestory models. Aide's "Sota" internal model (for plan-act handoffs) is hosted by codestory, that one call stays on their infra. Everything else (chat, edit, agent step models) respects your provider config.

MCP + Aide. Aide is adding MCP client support; Xantly passes MCP tool calls through normally.

Next steps

Cline, closest open-source equivalent.
Claude Code, Anthropic's CLI with MCP.
Zed, fast native editor alternative.
Bring Your Own Key, route Aide through your own provider credits.