Heroa  ·  agent-first hosting & execution

Run agents.
Let agents run anything.

Heroa is agent-first hosting and execution infrastructure. Spin up agents for minutes or months. Let those agents spin up their own instances — ephemeral scratch, long-lived sub-agents, even production apps — without ever touching a Dockerfile.

01
you → agent
Run agents, short or long.
One-shot tasks, persistent agents, nightly batches. Any lifetime.
02
agent → ephemeral
Agents spin up scratch instances.
Sandboxes, tool containers, sub-agents — live for a few seconds of work and gone.
03
agent → long-lived
Agents ship production apps.
A demo URL, a hosted API, a real production service — deployed mid-run via one tool call.
Regions
5 · 1 sovereign
Templates
12 pre-configured
Hosting
URL · API · service
Under the hood
R1 · Apache 2.0
parent_task · t_01HX2E9C
queued
Research agent → building a client dashboard
templateresearch-agent
regionus-east
step
cost $0.00
agent callheroa.deploy
await heroa.deploy("dashboard", { region: "us-east" })
nyt-q2.heroa.app200 OK
§ demo 01 / 05 Live substrate · 01 / 05

Watch the substrate breathe.

Five portfolio products. Five real workload shapes. Pick a scenario and watch Heroa spin instances up, sustain them, burst them under load, and decay them down. Events stream live.

LIVE heroa.ai / substrate / r1-tooluse +0.0s
R1 Heroa tool.run
Agent requests an ephemeral tool container
A research agent realizes it needs to exec a Python numeric toolchain for a minute.
load signal12%
idle1 instance
fleet · 0 instances cpu0% rps0 regionus-east spend / min$0.000 modeburst
events
scenario ledger
elapsed
0.0 s
peak instances
0
compute-min
0.00
billed mode
burst
net cost
$0.00
discount
illustrative workloads · every event is a real product mechanic
The agent deploys. You don't.

Hosting is a tool call.

Your agent doesn't hand you a zip to deploy — it deploys directly. heroa.deploy(…) is an SDK primitive the agent invokes mid-run. Heroa spins up the service, assigns a URL, applies the region and policy of the parent task, and returns the endpoint before the agent's next turn.

agent → hosted app

"Ship this dashboard."

The agent generates a frontend and calls deploy("app"). Heroa serves it at a subdomain on a CDN, TLS already on. The agent gets the URL back as a tool result and can paste it into its own response.

t-dashboard-q2.heroa.app200
agent → hosted api

"Expose this as an API."

The agent writes and packages a service, calls deploy("service"). Heroa hosts it long-running with autoscaling and per-request billing. The agent hands you a callable URL, not a git repo.

api.heroa.app/svc/t_4Pz/v1live
agent → hosted agent

"Keep me around."

The agent promotes itself to a warm endpoint via deploy("agent"). Future requests route back into the same R1 session. Heroa keeps it hot, applies policy per call, honors sovereign region pinning.

agents.heroa.app/supportbotwarm
§ demo 02 / 05 Start from running · 02 / 05

Twelve templates. Pick one.

Common agent shapes are pre-configured: research, coding, content, customer support, data analysis, review, and more. Every template ships with its stance, skills, and sensible resource policies. Start from running, not from a blank config.

12 / 12 templates shown
research

Research Agent

Stance

Preloaded skills

Typical run

duration
median cost
default region
§ demo 03 / 05 Where Heroa fits · 03 / 05

Three paths. Pick yours.

R1 is open source and free to self-host. CloudSwarm is the full control plane for fleet-scale operations. Heroa sits in the middle: the managed runtime when you want R1 without running infrastructure yourself. Three valid choices, different use cases.

Self-host R1
free · OSS
Heroa
managed · per-task
CloudSwarm
managed · fleet-scale
verdict
You are the platform team.
You ship product, not infrastructure.
You manage many agents, not one.
§ demo 04 / 05 Where it runs · 04 / 05

Five regions, your cloud, or your datacenter.

Submit a workload. Declare a region — or a deployment mode. Your execution never leaves the boundary. The Sovereign Canadian region runs on a single-tenant Canadian-resident substrate (GCP Montréal, northamerica-northeast1) with no cross-border data flow. BYOC GCP is on the Q3 2026 roadmap. On-prem installs ship as an air-gapped appliance.

Sovereignty tier STANDARD

Region metadata

Allowed inference providers

Region is a per-task argument. Your workload declares where it runs at submission time. us-e
§ demo 05 / 05 What it costs · 05 / 05

Transparent pricing. Drag to calculate.

Execution is priced per minute. Prompt caching credits apply automatically. Sovereign-region workloads have their own rate. Drag the sliders. Watch the estimate update. No surprises on the invoice.

heroa.ai / pricing / calculator illustrative rates
Execution minutes per month2,400 min
Total minutes your agents will run this month across all tasks.
Prompt cache hit rate72 %
Percentage of requests where R1's native prompt cache hits. Typical: 60-85%.
Sovereign Canadian fraction0 %
Percentage of your execution minutes that run in the sovereign region.
Standard @ $0.0028/min × 2,400$6.72
Sovereign @ $0.0035/min × 0$0.00
Cache credits applied−$1.34
Net estimated monthly cost$5.38
Recommended tier
Workspace
Estimate only. Real costs vary with task complexity, model mix, and prompt size. See /pricing for the complete rate schedule.
What's rare here

Six things that don't exist in the rest of the managed-runtime market.

Not features. Specific decisions that make Heroa different from cloud compute, different from agent frameworks, different from fleet orchestrators.

01

Run and host, same substrate.

Other runtimes stop at "the agent ran." Heroa keeps going: whatever your agent builds — an API, an app, a service, a generated site — is live at a URL the moment the run completes. Same billing rails, same region, same policies. Execution and hosting are not two products you stitch together. They are the substrate.

02

Purpose-built for R1, not a generic compute host.

Cloud compute runs anything and is good at nothing in particular for agents. Heroa is the runtime substrate designed around R1's stance harness, four-tier memory (called L0-L3 in the harness UI; the underlying R1 storage tiers are working / episodic / semantic / procedural), and STOKE protocol. Cold-start is fast because R1 sessions bootstrap from a common image. Memory layers are pre-provisioned. The substrate knows what an agent needs because it was built for one.

03

No Kubernetes, no container orchestration, on your side.

Most managed-runtime products are a thin wrapper on K8s. Heroa is a substrate. Your side of the boundary is an API and an SDK. You never write a Helm chart. You never read a pod log. You submit a task. You get a result — and if that result is a service, you get a URL.

04

Sovereignty as a workload-level declaration, not a separate deployment.

Most sovereign cloud offerings are a separate deployment you configure once and hope does not drift. Heroa takes a region on every task submission. A Canadian workload goes to the Sovereign Canadian substrate at submit time. An ITAR-flagged workload stays out of non-US regions. The sovereignty boundary is enforced per-task — and per-host.

05

Transparent per-minute pricing, not compute credits.

Most managed-compute products invent a "credit" unit that sounds simpler but makes cost modeling opaque. Heroa bills in execution minutes at a published per-minute rate, applies prompt-cache credits visibly, and produces an invoice that reads by agent, by task, by tenant. Your finance team can reconstruct every line.

06

Tenant-zero for the sovereign Canadian substrate.

Actium is the production customer-zero for the Sovereign Canadian substrate. The substrate's operational maturity was built through real workloads running for real Canadian clients, not through a provisioning exercise. When you sign up for Heroa's sovereign region, you are running on the same substrate that has been in production since day one.

Each claim corresponds to a product mechanic. Pricing details on /pricing.
Integrations

Heroa is the runtime that adjacent agent products run on.

Heroa stands alone as a standalone managed runtime. It also sits underneath agent products that coordinate and govern, providing the execution layer.

Heroa + R1

R1 is the framework. Heroa is the runtime. Every R1 agent running on Heroa gets the same framework guarantees with none of the infrastructure work.

r1.run

Heroa + CloudSwarm

CloudSwarm is the control plane. Heroa is the substrate it runs on. CloudSwarm coordinates; Heroa executes. Use them separately or together.

cloudswarm.app

Heroa + RelayGate

RelayGate ContextWorkers run R1 inline at request time. Heroa is the managed runtime those ContextWorkers execute on.

relaygate.ai

Heroa + RelayOne

Policy and evidence for managed runtimes: RelayOne's governance plane covers every Heroa execution when you compose the two products.

relayone.ai

Heroa + TrueCom

Heroa is a TrueCom-receipt producer: every Heroa execution emits signed receipts per task that TrueCom can settle on its rails. Billing reads the same events as audit.

truecom.ai
Heroa is commercial. The framework under everything (R1) is Apache 2.0.
Pricing

Transparent per-minute pricing. Three tiers plus sovereign.

Full pricing and feature matrix on /pricing. Quick preview below.

Developer
For solo developers building agent features.
free
  • Up to 2,000 execution minutes per month
  • All templates, standard regions
  • Community support, public changelog
Enterprise
For larger organizations with networking and compliance needs.
Custom
  • SSO, SCIM, private networking, custom regions
  • Dedicated success engineer, quarterly review
  • Optional CloudSwarm integration for fleet orchestration
Contact sales
Sovereign Canadian
For Canadian regulated workloads.
Custom sovereign rate
  • Single-tenant Canadian-resident substrate (GCP Montréal today; BC colocation on roadmap)
  • Canadian-entity MSA, enforceable in BC courts
  • No cross-border data flow, contractual no-logging
Book sovereign consultation
See full pricing on /pricing. Sovereign details on /sovereign.

Run an agent in five minutes.

Copy the snippet. Sign up for the free tier. Ship your first task.

import { Heroa } from "@heroa/sdk";

const heroa = new Heroa({ apiKey: process.env.HEROA_API_KEY });

const run = await heroa.tasks.create({
  template: "research-agent",
  region: "us-east",
  input: {
    question: "Summarize Q2 coverage of the New York Times",
    sources: ["nytimes.com"],
  },
});

const result = await run.wait();
console.log(result.summary);
from heroa import Heroa

heroa = Heroa(api_key=os.environ["HEROA_API_KEY"])

run = heroa.tasks.create(
    template="research-agent",
    region="us-east",
    input={
        "question": "Summarize Q2 coverage of the New York Times",
        "sources": ["nytimes.com"],
    },
)

result = run.wait()
print(result.summary)
import "github.com/heroa/heroa-go"

h := heroa.New(os.Getenv("HEROA_API_KEY"))

run, err := h.Tasks.Create(ctx, &heroa.TaskInput{
    Template: "research-agent",
    Region:   "us-east",
    Input: map[string]any{
        "question": "Summarize Q2 coverage of the New York Times",
        "sources":  []string{"nytimes.com"},
    },
})

result, err := run.Wait(ctx)
fmt.Println(result.Summary)
SDKs are MIT licensed. Runtime (R1) is Apache 2.0. Heroa managed service is commercial.