JETTSON.DEV · AGENTS
Three example agents. Each is a short prompt on top of a Jettson machine — fork one, point it at your data, and bring it online.
Classifies inbound Stripe events, drafts replies, posts a Slack summary.
~64 lines
of prompt — no orchestration code
First-pass review on every PR. Learns your team's style over time.
~80 lines
of prompt — no orchestration code
Drop a company URL, get a sales-ready dossier. Caches in memory.
~90 lines
of prompt — no orchestration code
Each one is a short prompt that runs on a fresh Jettson machine. Pick the closest one to your use case, change a few lines, and run your own.
Two SDKs, one shape. Spawn a production agent from Node or Python in three lines — no orchestration layer to assemble first.
Typed end-to-end. Works with the full Next.js / Express / Fastify lineup. Spawn-and-poll helpers built in.
import { Jettson } from "@jettson/sdk";
const client = new Jettson({ apiKey: process.env.JETTSON_KEY });
const agent = await client.agents.spawn({
task: "Find 10 plumbers in Phoenix",
});Async-friendly with httpx under the hood. Same shape as the Node SDK so your team can move between stacks.
from jettson import Jettson
client = Jettson(api_key=os.environ["JETTSON_KEY"])
agent = client.agents.spawn(
task="Find 10 plumbers in Phoenix",
)The infrastructure is the product. You write the prompt and get back an agent — Jettson handles the container, the memory, the tools, the spawn.
Read the docsEach agent gets its own isolated Linux container with a persistent file system, a browser, an identity, and a tool registry — wired together at spawn time, not by you at install time.
Hybrid recall fires on every spawn. The agent reads the relevant memories from prior runs before it even sees the new task — so the tenth call doesn't repeat the mistakes of the first.
US East · Ashburn
2/3
warm / total
412ms
p50 spawn
Europe · London
3/3
warm / total
488ms
p50 spawn
APAC · Sydney
2/3
warm / total
534ms
p50 spawn
hit rate · example
warm-first
illustrative
Multi-region warm pool. Memory that compounds. Sixteen built-in commands across five tool families. The boring infrastructure work — done.
Warm pools running in iad, lhr, and syd keep containers ready. First request lands in well under a second, warm.
Every run extracts what matters and stores it. The next run starts with relevant recall already in context.
Prefers usage-based billing
Use early returns; avoid == null
Match customer's formality level
browser · shell · files · http · memory — 16 commands, the full kit at spawn. No SDK wiring required.
Connect OpenAI, Anthropic, GitHub, Stripe and more. KMS-encrypted at rest, injected into the container at spawn, never logged. Agents push to your GitHub with a leak-proof token.