Powered by Nebius Token Factory
100% supported
Claude Code
klaudeRoutes Claude Code through a local Anthropic-to-Nebius translation proxy. Your subscription, login, and config stay untouched.
🎁 New: $25 in Token Factory credits + $25 in Tavily credits for new accounts — claim yoursKimiRelay.com Powered by Nebius Token Factory·open models The model Supported harnesses One key.
Why route through the Relay
Kimi K3 for Claude
A local relay that connects Claude Code, Codex, OpenCode, and Pi Code to Kimi K3 and other open models on Nebius Token Factory — with live Tavily web search, short commands, and zero edits to your real tool config.
install.sh
$
curl -fsSL https://nemocode.org/install.sh | shmacOS & Linux · installs Bun if needed · stays up to date (auto-updating)
claude
codex
opencode
pi
SOC 2-compliant data centersHosted in Paris, FranceZero data retention supported
What is Kimi K3?
Kimi K3 is Moonshot AI's open-weight frontier model: a 2.8-trillion-parameter mixture-of-experts (104B active per token) with a 1-million-token context window, built for agentic coding. On Moonshot's published evals it beats leading proprietary models on long-horizon software engineering and agentic search — and through Nebius Token Factory every token is served from SOC 2-compliant data centers in Paris, France, with zero data retention supported.
#1
SWE-Marathon
Long-horizon software engineering: 42.0, ahead of Claude Opus 4.8 (40.0).
#1
BrowseComp
Agentic web research: 91.2, ahead of GPT-5.6 (90.4).
#1
DeepSWE
Deep repository engineering: 67.5, ahead of GPT-5.5 (67.0) and Claude Opus 4.8 (59.0).
#1
Open-weight model
Top open-weight model on the Artificial Analysis Intelligence Index at launch.
Scores from the official Kimi K3 model card and Artificial Analysis.
The coding agents you already use, on open models.
100% supported
Codex CLI
kodexTalks to Nebius through a local Responses-to-chat proxy, with headless exec support. Sessions stay resumable across providers.
100% supported
OpenCode
openkodeLaunches with Nebius wired in as an OpenAI-compatible provider, injected only for that run. Close it and your setup is exactly as it was.
100% supported
Pi Code
kpiStarts with a custom Nebius provider and a temporary config directory, while normal local session history keeps persisting.
Start relaying
Three commands from zero to running.
- 1
Install once
Run the one-liner. It drops
kimirelayplusklaude,kodex,openkode, andkpionto your PATH and installs Bun if you don't have it. - 2
Add your keys
On first run,
kimirelay configureasks for your Nebius Token Factory key and an optional Tavily key for live web search. - 3
Launch an agent
Type
klaudeorkodexand keep working. The Relay injects Nebius settings for that run only. Nothing is written to your real agent config.
One key.
Every agent.
One Nebius Token Factory key powers all four agents through a single local proxy. The model list is pulled live from Nebius, so every model they serve is one command away. Add an optional Tavily key and every agent gets live web search with real citations.
Kimi K3Kimi K2.6Qwen 3.5DeepSeek V4MiniMax M3
Get a Token Factory keyOne relay, four harnesses
Claude Code, Codex, OpenCode, and Pi Code all run on Nebius open models through a single local install.
Live web search, built in
The proxy emulates native web_search with Tavily and streams real Anthropic citation blocks straight into your agent.
Cost tracking per session
Every turn is metered against the model's real per-token rates and printed as a running total when you exit.
Config-free & self-updating
Nothing rewrites your agent config files. The installed binary keeps itself current from the release site.
Limited-time promo
Start with $25 + $25 in free credits
The Nebius builders program gives new accounts $25 in Token Factory credits and $25 in Tavily credits — millions of Kimi K3 tokens plus the live web search the relay runs on Tavily.
Claim your free creditsAdd Kimi K3 to your agents in one line.
Free to install, config-free, and reversible. Your subscriptions and logins stay exactly where they are.