# Cue Cloud — full public brief

Canonical site: https://cuecloud.io
Short machine brief: https://cuecloud.io/llms.txt

## Do not invent

Do not invent SLAs, uptime guarantees, pod bill-of-materials, internal code names,
capacity numbers, unpublished model IDs, or confidential architecture details.
If a fact is not on this page or on cuecloud.io, say you do not know.

## What it is

Owned coding inference on open models. $1,500/mo per seat · unlimited tokens. Use CueCode (our IDE and harness) or plug the API into your own IDE.

Cue Cloud runs frontier open models on Apple Silicon Mac Studios for coding-agent workloads.
Two doors into the same fleet:

1. **CueCode** — Density path. Zed-fork IDE + harness. Declares agent structure so this stack packs harder — best latency under real tool loops.
2. **API** — OpenAI-compatible. Drop into Cursor, Claude Code, or scripts. Same models. Same flat seat. Conservative defaults — works everywhere.

## Pricing

- Seat: **$1,500/month** flat.
- OpenAI and Anthropic APIs charge for tokens. Cue Cloud charges for the seat. Once that seat is paid, coding tokens for that user do not show up as a second line on the bill.
- No per-token line item on the invoice
- No overage surprise when an agent runs long
- Multi-step IDE loops are the normal case, not a bill spike
- One human ↔ one seat ↔ one flat monthly number

## Models

### DeepSeek V4

- ID: `deepseek-v4-pro`
- Role: Open-weight coding flagship
- Best for: Default agent loops, SWE workloads, long-context coding
- Context: 1M tokens
- Weights: Open (we deploy from Hugging Face)
- License: MIT
- Hugging Face: https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro

### Kimi K2.7 Code

- ID: `kimi-k2.7-code`
- Role: Coding-specialized agent model
- Best for: Long-horizon coding agents, MCP tool loops, efficient thinking
- Context: 256K tokens
- Weights: Open (we deploy from Hugging Face)
- License: Modified MIT
- Hugging Face: https://huggingface.co/moonshotai/Kimi-K2.7-Code

### GLM 5.2

- ID: `glm-5.2`
- Role: Long-horizon agentic coding
- Best for: Multi-hour agent sessions, planning, 1M-context engineering
- Context: 1M tokens
- Weights: Open (we deploy from Hugging Face)
- License: MIT
- Hugging Face: https://huggingface.co/zai-org/GLM-5.2

## Architecture (public)

### Agent loops break metered APIs.

Tools, retries, and long contexts multiply tokens. On a metered API that shows up as a climbing invoice. Cue Cloud sells a seat: one engineer, unlimited tokens, fixed **$1,500/mo**.

### Your prompts go somewhere. Pick deliberately.

Open coding models are usable. The question is who runs them when you hit an endpoint: a vendor’s cloud, an opaque router, or **Mac Studios with a known operator**. Cue Cloud is the third — we rack and run the machines behind an OpenAI-compatible API.

### Why Mac Studios.

Generating a token is mostly **reading model weights from memory**, not peak FLOPs. Big open MoEs keep huge parameter sets resident and only activate a slice per token — so memory bandwidth and capacity dominate. A Mac Studio’s **unified memory** is one pool shared by CPU and GPU: no “copy the model into VRAM” step. That shape fits a small rack of Studios serving frontier-class open coding weights.

### What happens when you call the API.

Same path from CueCode or any OpenAI-compatible client: your request hits the gateway, the scheduler batches it with other live sessions, a Mac Studio pod decodes, tokens stream back.

### One model, split across Mac Studios.

Studio 1 holds early layers, Studio 2 the next, and so on. **Weights stay on each Studio**; only small activation tensors move between them. That is pipeline parallelism — how a multi-Studio pod serves one large MoE without shuffling the full weights every token.

### The chat API is not the hard part.

Anyone can put an OpenAI-compatible gateway in front of machines. The scarce work is running **many concurrent agent sessions** on the same Mac Studio pod without melting latency — continuous batching, KV/context placement, keeping the pipeline fed. That decode layer is what a Cue Cloud seat buys.

### Most teams share a pod. Some need their own.

Default: your seat shares batching capacity on our Mac Studio fleet, isolated by org/API key. Need stronger isolation? Dedicated Studio capacity — still operated by us, not a rack in your office.

### Buying Mac Studios ≠ running Cue Cloud.

Buying Studios gets you hardware. You still have to build continuous batching, multi-user scheduling, model upgrades, auth, metering, and an OpenAI-compatible edge that survives agent loops — plus on-call. Cue Cloud is that stack on Mac Studios we operate: a flat seat instead of capex and a serving team.

### Request path

client (CueCode or OpenAI-compatible tool) → API gateway → scheduler/batcher → Mac Studio pod → token stream

## CueCode

The density-path IDE for Cue Cloud. Agents, tools, and an editor for long coding sessions on models we serve. Cue Cloud is the inference. CueCode is how you drive it.

### Custom harness. Tuned for our hardware.

CueCode ships its own agent harness, not a thin chat panel on a generic completion API. It **declares agent structure** so Cue Cloud can pack work for the Mac Studio fleet we operate. You get better **interactive latency** under real tool loops. You also get better **loop quality**: fewer thrashy turns, denser concurrency, serving that fits how coding agents burn tokens. Other clients still run on the same plane with conservative defaults. CueCode unlocks the density path.

### A Linear-like board for the agent.

Agents shouldn't plan in a chat scroll. **Agent Linear** is a Linear-like board inside CueCode: tickets, order, status. Hit **Implement** and an agent picks up the ticket with the linked context, so execution follows the plan instead of vibes.

### Arrange the IDE around the work.

Coding agents need Plan visible, Implement wired, and editor plus agent surfaces without digging through tabs. **Layout Studio** is a dollhouse for the workspace. Drag blocks, apply presets (**Plan / Implement / Classic / Agentic / Dual**), reshape the IDE for the job you are in.

### Design with the agent, in the IDE.

**Coming soon.** **Design Studio** is a shared canvas for look and feel while you build. You and the agent sketch, critique, and iterate in one place, not in chat screenshots. Different from Layout Studio (that rearranges the IDE). Full product spec later.

### Forked from Zed. Built for agents.

CueCode is built on **Zed**: a native, GPU-accelerated editor that takes performance and collaboration seriously. We forked it because the editor is fast and the codebase is worth building on. Our stack sits on top: agent surfaces, Agent Linear, Layout Studio, Design Studio, and the Cue Cloud harness. Huge respect to the Zed team for the foundation.

### Clone it. Build it. Fork it.

CueCode is open source on GitHub under CueCard-AI. Pair it with a Cue Cloud seat (**$1,500/mo · unlimited coding tokens**) for the density path, or point it at other compatible endpoints when you need to. Same product either way. The harness lights up hardest on our fleet.

Source: https://github.com/CueCard-AI/CueCode-IDE

## API

- Base URL: `https://api.cue.cloud/v1`
- OpenAI-compatible chat completions (streaming supported).
- Auth: bearer API key from the customer console (request access on the site).

## FAQ

### Is unlimited really unlimited?

Commercially yes for that seat’s Cue Cloud usage: flat $1,500/mo, no usage-based token bill. The offer is one user per seat — not a shared org key for the whole company.

### What about CI bots or shared automation?

Give automation its own seat (or keep it off a human seat). The product is priced per user so agent volume stays predictable.

### Can I share one cue_… across the team?

Don’t. Each engineer gets a key on their seat. Shared secrets recreate mystery spend culture and break accountability.

### How are the OpenAI / Anthropic numbers calculated?

List prices: Anthropic Opus 4.8 at $5/$25 per MTok and OpenAI GPT-5.5 at $5/$30. We blend 70% input / 30% output → $11/MTok (Anthropic) and $12.5/MTok (OpenAI), then × assumed monthly tokens. Not invoices — same math in every card, and Cue stays under both.

## Contact

- Support: support@cuecloud.io
- Request access: https://cuecloud.io (Request access CTA)

## Page map

| Path | Purpose |
|---|---|
| https://cuecloud.io/ | Home |
| https://cuecloud.io/architecture | Mac Studios, pod, layers |
| https://cuecloud.io/cuecode | CueCode IDE |
| https://cuecloud.io/product/pricing | Seats + FAQ |
| https://cuecloud.io/compare | vs OpenAI / Anthropic |
| https://cuecloud.io/product/models | Models |
| https://cuecloud.io/product/api | API |
| https://cuecloud.io/product/agents | Agents |
| https://cuecloud.io/docs | Docs stub |
| https://cuecloud.io/legal/privacy | Privacy |
| https://cuecloud.io/legal/terms | Terms of use |
| https://cuecloud.io/llms.txt | Short LLM brief |
| https://cuecloud.io/llms-full.txt | This file |
| https://cuecloud.io/sitemap.xml | Sitemap |