Providers · one contract, server-side keys

Every provider Zap composes with.

Sandboxes for CPU execution, LLM routes and media generation behind the gateway, and payment rails for live spend. Plan-only runs quote against rate tables and never contact a provider; keys stay server-side.

providers

plan

default mode

Sandboxes

Isolated Zap sandbox VMs and containers where CPU work executes. Same contract everywhere: exec, files, snapshots, ports.

Zap Sandbox

default

box

Zap's default sandbox: full VMs on ascii.dev.

Snapshots + fork
Stop / resume
Hosted ports
Desktop streaming
Docker inside

E2B

e2b

Firecracker microVM sandboxes.

Pause / resume snapshots
Public HTTPS ports

Daytona

daytona

Container sandboxes.

Snapshots
HTTPS preview links

Cloudflare Sandbox

cloudflare

Container sandboxes on Cloudflare Workers.

Backup / restore snapshots
Public preview ports

Docker

docker

Local container sandboxes for development.

Same contract locally
Container isolation
No cloud account required

microsandbox

microVM isolation on self-hosted KVM hosts.

Self-hosted KVM
microsandbox SDK (pinned 0.6.15)

Namespace

namespace

Linux containers and native macOS Apple-silicon instances.

Linux container instances
macOS Apple-silicon instances
Two-token bridge

Modal

GPU lane

modal

The GPU lane target — mounts only when a gpu lane is declared.

On-demand GPU functions
Declared gpu lanes only

LLM routes

Model providers behind the gateway. Deterministic routing, pure quotes, and provider keys that never leave the server side.

OpenRouter

default

openrouter

Default LLM route across hundreds of models.

Multi-model routing
BYOK or managed

OpenAI

openai

GPT model family, including Codex device auth.

GPT models
Codex device-auth login

Anthropic

anthropic

Claude model family, including Claude Code auth.

Claude models
Claude Code device-auth login

xAI

xai

Grok model family.

Grok models

Vercel AI Gateway

gateway

Vercel's multi-provider AI gateway.

Unified model access
Provider fallback

GMI Cloud

gmi

GPU cloud inference for LLM and media models.

LLM inference
Seedance video models

Media generation

Image, video, and audio generation providers. Plan-only never contacts a provider; live runs require a payer and explicit intent.

fal

Fast image and video model inference.

Image generation
Video generation
Webhook result delivery

Prodia

prodia

Low-cost image generation API.

Image generation
Webhook result delivery

Runware

runware

Sub-second image inference network.

Image generation
Webhook result delivery

Replicate

replicate

Open-model marketplace for media generation.

Image / video / audio models
Open-weights catalog

AWS Bedrock

aws

Amazon's managed foundation-model service.

Nova / Titan media models
Enterprise IAM auth

Vertex AI

vertex

Google Cloud's foundation-model platform.

Imagen + Veo models
Google Cloud auth

GMI Cloud

gmi

Seedance video generation, used by the Air iMessage service.

Seedance video
Air video pipeline

Payments

Zap never spends money by default. Every runtime resolves a payer before executing a prompt — quoted, metered, capped, fail-closed.

thirdweb

Wallet auth for Studio and managed payment sessions.

SIWE wallet sign-in
Scoped session keys
Spend caps

Coinbase x402

x402 v2 payment protocol through the Zap Cloud payment gate.

PAYMENT-SIGNATURE settlement
Replay-protected receipts
MPP also accepted