LunaRoute — Inference that waxes. Never wanes.

Inference that waxes. Never wanes.

One key, a deep roster of frontier models, and a plan priced by the lane. Fixed monthly price, no usage caps, no token meter — and nothing stored.

▸ request bodies stored: never, nowhere

▸ live ledger of every request & charge

▸ web search, document conversion & OCR, image generation & editing — same key, same ledger


;

;;!;;!!!!!!!!!!!;!!;;;;;

!!!!!;!!!!;

;

....

One endpoint. More models than moon phases.

Pick any model on the roster — the router dispatches every request to the fastest healthy provider serving it, and fails over before your agent notices. New models dock regularly; your endpoint never changes.

your agent ------> +---------------------+
| LUNAROUTE CORE |
POST /v1/chat | |- latency probes |
model: glm-5.3 | |- health checks |
| '- auto failover |
+----------+----------+
+-----------+----------+----------+-----------+
v           v          v          v           v
glm-5.3    kimi-k3    OpenAI      Anthropic  deepseek
v4-pro
…and more, docking monthly

Search, read, and draw. On the same key.

A model is a brain. Your agent also needs eyes and hands. LunaRoute serves both over MCP from one endpoint: web search across multiple providers, document conversion with OCR, and multiple image models for generation and editing, on the API key it already holds.

▸one key, one bill — no search-API account, no image-model account, no document-parsing SaaS. Every call lands next to your inference in the same ledger.

▸same privacy posture — LunaRoute keeps no search queries and no documents, only the count, and routes search through providers that offer zero data retention. Images and stored conversions your agent asks us to keep live seven days, then delete.

▸governed, not bolted on — admins decide which tools each teammate's agent may call.

▸works with any MCP client — Claude Code, Codex, Pi, Cursor and friends. One server, one header, every harness.

tool no

Sign up. Connect your agent. Route.

LunaRoute is one OpenAI-compatible endpoint that routes every request to frontier models with zero data retention — no logs, no usage meter, flat monthly plans. Getting started is three steps:

Sign up — pick a plan below, pay, your workspace is ready.
Connect your agent — one command configures Pi, Claude Code, opencode, or any OpenAI-compatible client.
Route — every request lands on your plan's lanes. That's it.

Each plan includes a set of concurrency lanes plus background lanes for batch work, on best-effort or priority-start scheduling (flex / realtime in the response headers). There is no usage allowance to watch and no overage bill — heavy usage won't trigger usage-based blocking or overage charges. NO LOGS, every model, one endpoint.

MONTHLY YEARLY · SAVE UP TO 17%


..........,,:;
       ...,.........,.,::;!
     .    .,,...,.....,,:;!!~
          .,.......,,.,,::;!!~
          .......,.,,..,,:;;!!;
          .,...     .,,,,:;;!!;:
         .,..          .,::;;!;,:
        ....            ,,:;;!;::
     ..,....            .,,:;!!!;
 ...,.....,,            ..,:;;!~~
  .....,,.,.,         ...,,;;!!~=
  .,...,,.....,.  . .....,,:;;!~=
   ,.....    .,,.....,...,,:;;!~=
   ....        ....,,.   .::;;!~
   ....         ....,    ,:;;!~
    .....      ....,.. .,,::;~
     .......,.....,,,.,..::;!
       .....,....,.....,,:;
          ............,,

Every request, accounted for.

Your plan covers LunaRoute private inference by the lane — a fixed monthly price, no meter. Every request still lands in a live ledger: model, tokens, lane-seconds, scheduling class, and every agent-tool call, in one table. No surprise invoices, no "contact sales to understand your bill".

$ lunaroute usage
TIME(UTC)  MODEL            TOK_IN   TOK_OUT  LANE_S  QOS
14:02:11   glm-5.3          12,408    3,211    3.4    realtime
14:02:09   qwen-3.8-max     48,116    1,094   12.1    flex
14:01:55   deepseek-v4-pro   8,733    6,402    7.8    flex
14:01:48   web_search            —        —      —    tool
14:01:32   minimax-m3      402,338    2,114   88.6    background
─────────────────────────────────────────────────
PLAN: Apex · 6 lanes · fixed price · nothing to top up

What happens on LunaRoute stays nowhere.

LunaRoute is zero data retention: request and response bodies are processed in memory and never written to disk — not by us, and not by our providers, contractually. Your prompts, your code, your agent's questionable variable names: gone the moment the response ships.

▸ what we keep: timestamps, model, token counts, cost — your ledger

▸ what we don't: everything else. bodies, headers, your secrets

Your agent forgets. The route remembers.

The one thing we will keep, if you ask. Every request your agent makes already passes through LunaRoute; opt a project in and that stream becomes queryable history — past decisions, dead ends, the why behind the code — with no client-side hooks. Only a router in the path can do this.

▸off by default — strictly opt-in and admin-controlled. NO LOGS stays the default until you say otherwise.

▸scoped per project — shared across the team, or private per API key. Retention window is yours to set.

▸recall from any harness — Claude Code, Codex, Pi, your own — over the CLI or a local MCP server. Private embeddings included.

$ lunaroute skill install # wires memory into Claude Code $ lunaroute memory search "why did we drop per-seat billing"

#a41f 2 weeks ago teams plan: seats -> flat lanes #9c02 3 weeks ago billing model spike, rejected $ lunaroute memory read a41f

Frequently anticipated questions.

What does NO LOGS actually mean?

Which models can I use?

What is a lane, and what are background lanes?

Is there a usage cap, or overage billing?

Do I need separate accounts for web search, document conversion, OCR or image models?

Can I switch plans mid-month?

Is there a free tier?

Ready for liftoff?

Fixed-price private inference from $19.99/mo. No meter, no overage. Cancel anytime.

$ npx lunaroute-cli init


;

;;!;;!!!!!!!!!!!;!!;;;;;

!!!!!;!!!!;

;

....

:;:;;;;;;;;;!;!!!!!!!!!;;;;;