Ship to the mesh in minutes.

Quickstart

This guide deploys a model to the global mesh and serves a request in under five minutes. You'll need a Cyata account and Node.js 20+ or Python 3.11+.

1. Install the CLI

npm install -g @cyata/cli
cyata login

2. Deploy your first model

cyata deploy --model cyata/llama3-70b \
  --regions auto \
  --replicas 0..32

# ✓ resolved weights (142GB)
# ✓ streaming to 38 regions
# ✓ endpoint → https://api.cyata.cloud/v1/llama3

3. Call it

curl https://api.cyata.cloud/v1/llama3/infer \
  -H "Authorization: Bearer $CYATA_KEY" \
  -d '{"prompt":"What is Cyata?","stream":true}'

SDKs

First-class SDKs for Python and TypeScript mirror the REST API and add typed helpers for streaming, agents, and the data mesh.

# Python
pip install cyata

# TypeScript
npm install @cyata/sdk

from cyata import Cyata

cy = Cyata(api_key=os.environ["CYATA_KEY"])

stream = cy.models("cyata/llama3-70b").stream(
    prompt="Explain the mesh in one line.",
    regions="auto",
)
for token in stream:
    print(token, end="", flush=True)

CLI reference

• cyata deploy — deploy a model or agent to the mesh
• cyata agent run <file> — run an agent workload
• cyata logs <id> — tail streaming logs
• cyata scale <id> --replicas 0..N — adjust autoscale bounds
• cyata mesh put <topic> <file> — publish to the data mesh
• cyata status — view regions, health, and spend

Agent runtime

Agents are declarative. Define a model, a sandbox, context streams, and a schedule; the runtime handles placement, checkpoints, and recovery.

agent: support-triage
model: cyata/llama3-70b
runtime:
  sandbox: microvm
  limits: { cpu: 2, mem: 4Gi }
context:
  stream: [tickets.live, kb.vectors]
  policy: tenant-scoped
schedule:
  strategy: closest-healthy
  replicas: 1..24
  checkpoint: every 30s

Data mesh

The mesh exposes streaming topics, a vector store, and a lineage graph under one API. Subscribe an agent to a topic and it receives live context without bespoke plumbing.

cy.mesh.subscribe("tickets.live", handler=on_event)
cy.mesh.vectors.upsert(collection="kb", items=[...])
cy.mesh.lineage.trace(output_id="out_8f3")

Observability

Every request emits a trace with tokens, latency, cost, and carbon — no extra instrumentation. Filter in the dashboard or export to your OTLP endpoint.

Status & SLAs

Live region health is published at status.cyata.cloud. Scale plans include a 99.95% SLA; Enterprise includes 99.99% with a dedicated TAM.