Pulumi Blog: Pulumi esc

Fully Automated AI Inference on AWS, Azure, and Google Cloud with Pulumi

Engin Diri — Tue, 30 Jun 2026 00:00:00 +0000

Putting Ollama on a cloud GPU is something I keep coming back to. A while ago I wrote up running open-source LLMs on an AWS EC2 box with Ollama and Pulumi, and the shape never really changes: a GPU instance, a model server, and a firewall rule in front. Infrastructure as code earned its place by making that kind of setup predictable and repeatable, and AI infrastructure is no exception. A GPU box serving a model is still a VM, a disk, and a firewall rule, and it should be declared like one.

Thorsten Hans made exactly that case in his Akamai post, Fully Automated AI Infrastructures with Terraform and Akamai Cloud, which stands up a single GPU instance on Linode, installs the drivers, runs Ollama, and pulls a model, with no manual steps after terraform apply.

I liked the shape of it, so this post ports the same idea to Pulumi and runs it across AWS, Azure, and Google Cloud instead of one. The result is one program shape per cloud: a single pulumi up brings up a GPU box that installs its own driver, runs Ollama, and pulls a model with no manual steps, and a single pulumi destroy takes it back down. Along the way it drops the two imperative bits the Terraform version leans on: a static access token sitting in an environment variable, and a null_resource running a shell loop to wait for the model. The first becomes an OIDC login from a Pulumi ESC environment, so no long-lived key lives anywhere. The second turns out not to be a resource at all.

What you are building

Strip away the per-cloud naming and every version of this is the same three things: a GPU virtual machine, a firewall in front of it, and a cloud-init script that turns a bare Ubuntu box into a running inference server. The model serving runs on Ollama, which exposes an HTTP API on port 11434 and keeps the model resident in GPU memory between requests.

flowchart LR
Dev([Your machine / curl]) -->|"HTTP :11434"| FW["Firewall / security group<br/>(allow 11434, optional 22)"]
FW --> VM["GPU VM (Ubuntu 24.04)<br/>NVIDIA driver + Ollama"]
VM -->|"resident in GPU memory"| Model["qwen2.5:14b"]
ESC["Pulumi ESC<br/>pulumi-idp/auth"] -.->|"OIDC login (short-lived creds)"| Pulumi["pulumi up"]
Pulumi -.->|"AWS / Azure / GCP"| VM

Component	Port	Description
GPU VM	-	Ubuntu 24.04 with an NVIDIA T4-class GPU. Installs the driver and Ollama from cloud-init, then pulls the model.
Ollama	11434	Serves the model over an HTTP API and keeps it in GPU memory between calls.
Firewall / security group	-	Allows inbound `11434` (and `22` when you ask for it). Open to the world for a demo; lock the CIDR down for anything real.
`pulumi-idp/auth` (ESC)	-	One environment that brokers an OIDC login into AWS, Azure, and GCP. No static keys in code or CI.

One detail here is worth pausing on, and it explains why the Pulumi version comes out shorter than the Terraform one. The Akamai project ends with a null_resource that runs a curl loop in local-exec to block terraform apply until the model finishes downloading. That is not infrastructure but a runtime check wearing a resource costume. Pulumi has a Command provider that would let you reproduce it line for line, and this post deliberately does not. The program provisions the box, the box pulls the model on its own, and the program prints the endpoint. Whether the model has finished downloading yet is a question you answer with a curl, not a question your IaC tool should be holding a deployment open to ask.

Prerequisites

Before getting started, ensure you have:

Pulumi CLI installed and configured
A Pulumi Cloud account (ESC and OIDC live here)
An account on at least one of AWS, Azure, or Google Cloud, with an OIDC trust set up for Pulumi (covered below)
Enough GPU quota in your target region for one T4-class instance; a fresh account often starts at zero, which is the most common reason a first deploy fails, so check it before you run pulumi up
Node.js 18+ for the TypeScript program
curl to talk to the inference endpoint once it comes up

This post serves the qwen2.5:14b model, the same one the Akamai article uses, because the 4-bit quant fits comfortably on a single 16 GB T4. The model is one config value (model), so swap in anything Ollama supports; match the GPU to the model’s memory footprint, because a model that overflows VRAM still runs but spills onto the CPU and crawls.

Credentials without the copy-paste

The Terraform version authenticates the only way a single-cloud demo can: you mint a personal access token, export it as LINODE_TOKEN, and the provider reads it from the environment. It works, but that token is long-lived, it sits in your shell history and your CI secrets, and you mint one per cloud. Across three clouds that is three static keys to rotate and worry about.

Pulumi ESC (Environments, Secrets, and Configuration) replaces it all with an OIDC login. The idea: instead of storing a cloud key, you store a trust relationship. At deploy time, ESC presents a short-lived OIDC token to AWS, Azure, or Google Cloud, and each one hands back temporary credentials scoped to a role you control. Nothing long-lived is ever written down.

I keep that wiring in one environment, pulumi-idp/auth, and every stack imports it. Here is the whole thing:

# pulumi-idp/auth: one ESC environment that brokers an OIDC login into all three
# clouds. Every stack imports it. No static cloud key lives here or anywhere else.
values:
 aws:
 login:
 fn::open::aws-login:
 oidc:
 roleArn: arn:aws:iam::123456789012:role/pulumi-esc
 sessionName: pulumi-esc
 azure:
 login:
 fn::open::azure-login:
 clientId: aaaaaaaa-bbbb-cccc-dddd-eeeeeeeeeeee
 tenantId: aaaaaaaa-bbbb-cccc-dddd-eeeeeeeeeeee
 subscriptionId: /subscriptions/00000000-0000-0000-0000-000000000000
 oidc: true
 gcp:
 login:
 fn::open::gcp-login:
 project: 123456789012 # numeric project number, not the project ID
 oidc:
 workloadPoolId: pulumi-esc
 providerId: pulumi-esc
 serviceAccount: pulumi-esc@my-project.iam.gserviceaccount.com
 environmentVariables:
 # AWS: read by the Pulumi AWS provider, the AWS SDKs, and the aws CLI
 AWS_ACCESS_KEY_ID: ${aws.login.accessKeyId}
 AWS_SECRET_ACCESS_KEY: ${aws.login.secretAccessKey}
 AWS_SESSION_TOKEN: ${aws.login.sessionToken}
 # Azure: read by the azure-native provider
 ARM_USE_OIDC: "true"
 ARM_CLIENT_ID: ${azure.login.clientId}
 ARM_TENANT_ID: ${azure.login.tenantId}
 ARM_SUBSCRIPTION_ID: ${azure.login.subscriptionId}
 ARM_OIDC_TOKEN: ${azure.login.oidc.token}
 # Google Cloud: read by the Pulumi Google Cloud provider
 GOOGLE_CLOUD_PROJECT: ${gcp.login.project}
 GOOGLE_OAUTH_ACCESS_TOKEN: ${gcp.login.accessToken}

You only need the blocks for the clouds you actually deploy to. Opening this environment makes ESC perform a live OIDC login for each cloud listed, so trim it to the one you use, or keep all three if, like me, you bounce between them.

Each cloud needs a one-time trust setup so this login is allowed at all: an IAM OIDC identity provider and role on AWS, a federated credential on an Azure app registration, and a Workload Identity Pool on Google Cloud. You do that once; after that, pulumi-idp/auth is the only thing any stack references for credentials.

A stack opts into it with one block in its stack config. The AWS stack’s Pulumi.aws.yaml:

# Pulumi.aws.yaml
environment:
 - pulumi-idp/auth

That is the entire credential story. No keys in the program, no keys in CI, nothing to rotate. The program below never mentions a secret; it only creates resources, and the ambient credentials from pulumi-idp/auth carry the request.

One cloud-init, three clouds

The part that turns a bare Ubuntu box into an inference server is identical on every cloud, so it lives in one file, cloud-init.yaml, that all three programs read. It follows the Akamai cloud-config closely, with a couple of robustness tweaks for a multi-cloud run, and it runs in a deliberate order because the GPU driver needs a reboot before Ollama can see the card:

Update packages and install the kernel headers and ubuntu-drivers.
Install the NVIDIA driver, then reboot so the kernel loads it.
On the next boot, a one-shot systemd service installs Ollama, binds it to 0.0.0.0:11434, and pulls the model.
The service disables itself, so it never runs again.

#cloud-config
# Zero-touch Ollama GPU box. Cloud-agnostic: no provider metadata, no static creds.
# The Pulumi program substitutes the model name on the `ollama pull` line below
# before this is passed as user-data (AWS/GCP) or custom-data (Azure).

write_files:
 # Make Ollama listen on every interface and never unload the model from VRAM.
 - path: /etc/systemd/system/ollama.service.d/override.conf
 content: |
 [Service]
 Environment="OLLAMA_HOST=0.0.0.0:11434"
 Environment="OLLAMA_KEEP_ALIVE=-1"

 # Runs once on the post-reboot boot, after the GPU driver is loaded.
 - path: /usr/local/bin/ollama-setup.sh
 permissions: '0755'
 content: |
 #!/usr/bin/env bash
 set -euxo pipefail

 # Install Ollama; the installer creates and starts the ollama systemd unit.
 curl -fsSL https://ollama.com/install.sh | sh

 # Pick up the OLLAMA_HOST / OLLAMA_KEEP_ALIVE override written above.
 systemctl daemon-reload
 systemctl enable ollama.service
 systemctl restart ollama.service

 # Wait for the daemon to bind its socket before pulling. This is the box
 # waiting on its OWN local daemon, not an external readiness gate.
 until curl -fsS http://127.0.0.1:11434/api/tags >/dev/null 2>&1; do
 sleep 2
 done

 # Pre-pull the model so the endpoint answers on the very first request.
 ollama pull __MODEL__

 # One-shot: never run again on future boots.
 systemctl disable ollama-setup.service

 - path: /etc/systemd/system/ollama-setup.service
 content: |
 [Unit]
 Description=One-time Ollama install and model pull
 After=network-online.target
 Wants=network-online.target

 [Service]
 Type=oneshot
 ExecStart=/usr/local/bin/ollama-setup.sh
 RemainAfterExit=true
 # The model pull runs for several minutes; without this, systemd's default
 # 90s start timeout kills the service mid-download and the model never lands.
 TimeoutStartSec=0

 [Install]
 WantedBy=multi-user.target

runcmd:
 - apt-get update
 - DEBIAN_FRONTEND=noninteractive apt-get -y upgrade
 # linux-headers-generic also covers the kernel the upgrade above may have pulled
 # in, so DKMS builds the NVIDIA module for the kernel that boots next.
 - DEBIAN_FRONTEND=noninteractive apt-get install -y "linux-headers-$(uname -r)" linux-headers-generic ubuntu-drivers-common
 # --gpgpu selects the headless server driver branch, the right one for a compute
 # GPU like the T4; bare `ubuntu-drivers install` would pull the desktop stack.
 - ubuntu-drivers install --gpgpu
 - systemctl enable ollama-setup.service

power_state:
 mode: reboot
 message: Rebooting to load the NVIDIA driver before Ollama setup
 condition: true

The model name is the literal token __MODEL__. The Pulumi program reads this file at deploy time and substitutes your model config value before handing it to the instance. Two environment settings are doing quiet but important work: OLLAMA_HOST=0.0.0.0:11434 makes Ollama listen on every interface instead of localhost alone, and OLLAMA_KEEP_ALIVE=-1 keeps the model pinned in GPU memory so only the first request pays the load cost.

The programs

Every program has the same five beats: read the config, template the cloud-init.yaml, open the inbound ports, launch the GPU instance with that cloud-init as its user data, and export the endpoint. What differs is only the dialect each cloud speaks for “GPU instance” and “firewall rule.”

The shared contract keeps the three listings legible side by side: the same model and allowSsh config keys, the same cloud-init.yaml, and the same four exports (publicIp, ollamaEndpoint, generateEndpoint, tagsEndpoint). Pick your cloud:

import * as pulumi from "@pulumi/pulumi";
import * as aws from "@pulumi/aws";
import * as fs from "fs";

const cfg = new pulumi.Config();
const model = cfg.get("model") ?? "qwen2.5:14b";
const allowSsh = cfg.getBoolean("allowSsh") ?? false;
const region = (cfg.get("region") ?? "us-east-1") as aws.Region;

// Inject the model name into cloud-init at deploy time (a file read, not a shell-out).
const userData = fs.readFileSync("cloud-init.yaml", "utf8").replace(/__MODEL__/g, model);

// Region comes from config; credentials arrive ambiently from pulumi-idp/auth.
const provider = new aws.Provider("aws", { region });

// Most recent Ubuntu 24.04 LTS (amd64, hvm, gp3) published by Canonical.
const ubuntu = aws.ec2.getAmi({
 mostRecent: true,
 owners: ["099720109477"],
 filters: [
 { name: "name", values: ["ubuntu/images/hvm-ssd-gp3/ubuntu-noble-24.04-amd64-server-*"] },
 { name: "virtualization-type", values: ["hvm"] },
 ],
}, { provider });

const ingress: aws.types.input.ec2.SecurityGroupIngress[] = [{
 description: "Ollama HTTP API. NOTE: restrict cidrBlocks to your own range in production.",
 fromPort: 11434,
 toPort: 11434,
 protocol: "tcp",
 cidrBlocks: ["0.0.0.0/0"],
}];

if (allowSsh) {
 ingress.push({
 description: "SSH",
 fromPort: 22,
 toPort: 22,
 protocol: "tcp",
 cidrBlocks: ["0.0.0.0/0"],
 });
}

const sg = new aws.ec2.SecurityGroup("ollama", {
 description: "Ollama inferencing access",
 ingress,
 egress: [{
 description: "Allow all outbound",
 fromPort: 0,
 toPort: 0,
 protocol: "-1",
 cidrBlocks: ["0.0.0.0/0"],
 }],
}, {
 provider,
});

const server = new aws.ec2.Instance("ollama", {
 ami: ubuntu.then(a => a.id),
 instanceType: "g4dn.xlarge", // 1x NVIDIA T4
 vpcSecurityGroupIds: [sg.id],
 associatePublicIpAddress: true,
 userData, // plain text; the AWS provider base64-encodes it for you
 rootBlockDevice: {
 volumeSize: 40,
 volumeType: "gp3",
 },
 tags: { Name: "ollama" },
}, {
 provider,
});

export const publicIp = server.publicIp;
export const ollamaEndpoint = pulumi.interpolate`http://${publicIp}:11434`;
export const generateEndpoint = pulumi.interpolate`http://${publicIp}:11434/api/generate`;
export const tagsEndpoint = pulumi.interpolate`http://${publicIp}:11434/api/tags`;

import * as pulumi from "@pulumi/pulumi";
import * as gcp from "@pulumi/gcp";
import * as fs from "fs";

const cfg = new pulumi.Config();
const model = cfg.get("model") ?? "qwen2.5:14b";
const allowSsh = cfg.getBoolean("allowSsh") ?? false;
const project = cfg.get("project");
const zone = cfg.get("zone") ?? "us-central1-a";

// Inject the model name into cloud-init at deploy time (a file read, not a shell-out).
const userData = fs.readFileSync("cloud-init.yaml", "utf8").replace(/__MODEL__/g, model);

// Latest Ubuntu 24.04 LTS amd64 image.
const ubuntu = gcp.compute.getImage({
 family: "ubuntu-2404-lts-amd64",
 project: "ubuntu-os-cloud",
});

// Network tag binds the firewall rule to this instance.
const networkTag = "ollama";

const firewall = new gcp.compute.Firewall("ollama-fw", {
 network: "default",
 project: project,
 // Inbound to Ollama from anywhere; restrict this CIDR for production.
 allows: allowSsh
 ? [{ protocol: "tcp", ports: ["11434"] }, { protocol: "tcp", ports: ["22"] }]
 : [{ protocol: "tcp", ports: ["11434"] }],
 sourceRanges: ["0.0.0.0/0"],
 targetTags: [networkTag],
});

const instance = new gcp.compute.Instance("ollama", {
 machineType: "n1-standard-4",
 zone: zone,
 project: project,
 tags: [networkTag],
 bootDisk: {
 initializeParams: {
 image: ubuntu.then(i => i.selfLink),
 size: 40,
 },
 },
 guestAccelerators: [{
 type: "nvidia-tesla-t4",
 count: 1,
 }],
 // GPUs cannot live-migrate, so host maintenance must terminate (and restart) the VM.
 scheduling: {
 onHostMaintenance: "TERMINATE",
 automaticRestart: true,
 },
 networkInterfaces: [{
 network: "default",
 accessConfigs: [{}], // an empty config requests an ephemeral public IP
 }],
 metadata: {
 "user-data": userData, // Ubuntu cloud-init reads this key, not startup-script
 },
}, {
 dependsOn: firewall,
});

export const publicIp = instance.networkInterfaces.apply(nics => nics[0].accessConfigs![0].natIp);
export const ollamaEndpoint = pulumi.interpolate`http://${publicIp}:11434`;
export const generateEndpoint = pulumi.interpolate`http://${publicIp}:11434/api/generate`;
export const tagsEndpoint = pulumi.interpolate`http://${publicIp}:11434/api/tags`;

import * as pulumi from "@pulumi/pulumi";
import * as resources from "@pulumi/azure-native/resources";
import * as network from "@pulumi/azure-native/network";
import * as compute from "@pulumi/azure-native/compute";
import * as random from "@pulumi/random";
import * as fs from "fs";

const cfg = new pulumi.Config();
const model = cfg.get("model") ?? "qwen2.5:14b";
const allowSsh = cfg.getBoolean("allowSsh") ?? false;
const location = cfg.get("location") ?? "eastus";

// Inject the model name into cloud-init at deploy time (a file read, not a shell-out).
const userData = fs.readFileSync("cloud-init.yaml", "utf8").replace(/__MODEL__/g, model);

const resourceGroup = new resources.ResourceGroup("ollama-rg", {
 location,
});

const vnet = new network.VirtualNetwork("ollama-vnet", {
 resourceGroupName: resourceGroup.name,
 addressSpace: { addressPrefixes: ["10.0.0.0/16"] },
});

const subnet = new network.Subnet("ollama-subnet", {
 resourceGroupName: resourceGroup.name,
 virtualNetworkName: vnet.name,
 addressPrefix: "10.0.1.0/24",
});

// Standard SKU public IPs use static allocation.
const publicIpAddress = new network.PublicIPAddress("ollama-pip", {
 resourceGroupName: resourceGroup.name,
 sku: { name: network.PublicIPAddressSkuName.Standard },
 publicIPAllocationMethod: network.IPAllocationMethod.Static,
});

// Azure permits all outbound by default, so only inbound rules are needed.
const nsg = new network.NetworkSecurityGroup("ollama-nsg", {
 resourceGroupName: resourceGroup.name,
 securityRules: [
 {
 name: "allow-ollama",
 priority: 1000,
 direction: network.SecurityRuleDirection.Inbound,
 access: network.SecurityRuleAccess.Allow,
 protocol: network.SecurityRuleProtocol.Tcp,
 sourcePortRange: "*",
 destinationPortRange: "11434",
 sourceAddressPrefix: "0.0.0.0/0", // prod: restrict to your client CIDR
 destinationAddressPrefix: "*",
 },
 ...(allowSsh ? [{
 name: "allow-ssh",
 priority: 1001,
 direction: network.SecurityRuleDirection.Inbound,
 access: network.SecurityRuleAccess.Allow,
 protocol: network.SecurityRuleProtocol.Tcp,
 sourcePortRange: "*",
 destinationPortRange: "22",
 sourceAddressPrefix: "0.0.0.0/0",
 destinationAddressPrefix: "*",
 }] : []),
 ],
});

const nic = new network.NetworkInterface("ollama-nic", {
 resourceGroupName: resourceGroup.name,
 networkSecurityGroup: { id: nsg.id },
 ipConfigurations: [{
 name: "ipconfig1",
 subnet: { id: subnet.id },
 publicIPAddress: { id: publicIpAddress.id },
 privateIPAllocationMethod: network.IPAllocationMethod.Dynamic,
 primary: true,
 }],
});

// Azure requires an admin credential even though we never log in. Generate one
// instead of hard-coding it; it stays a Pulumi secret and is never exported.
const adminPassword = new random.RandomPassword("ollama-admin-password", {
 length: 24,
 special: true,
 overrideSpecial: "!#$%*",
 minLower: 1,
 minUpper: 1,
 minNumeric: 1,
 minSpecial: 1, // Azure requires 3 of 4 character classes
});

const vm = new compute.VirtualMachine("ollama-vm", {
 resourceGroupName: resourceGroup.name,
 hardwareProfile: { vmSize: "Standard_NC4as_T4_v3" }, // 1x NVIDIA T4
 networkProfile: {
 networkInterfaces: [{ id: nic.id, primary: true }],
 },
 osProfile: {
 computerName: "ollama",
 adminUsername: "azureuser",
 adminPassword: adminPassword.result,
 customData: Buffer.from(userData).toString("base64"), // Azure wants base64
 linuxConfiguration: {
 disablePasswordAuthentication: false,
 },
 },
 storageProfile: {
 imageReference: {
 publisher: "Canonical",
 offer: "ubuntu-24_04-lts",
 sku: "server", // Gen2; NC4as_T4_v3 is a Gen2 size
 version: "latest",
 },
 osDisk: {
 name: "ollama-osdisk",
 createOption: "FromImage",
 diskSizeGB: 40,
 managedDisk: { storageAccountType: "StandardSSD_LRS" },
 },
 },
});

// A Standard/Static public IP is reserved at creation, so its address is known
// once the resource exists; no separate lookup needed.
export const publicIp = publicIpAddress.ipAddress.apply(ip => ip!);
export const ollamaEndpoint = pulumi.interpolate`http://${publicIp}:11434`;
export const generateEndpoint = pulumi.interpolate`http://${publicIp}:11434/api/generate`;
export const tagsEndpoint = pulumi.interpolate`http://${publicIp}:11434/api/tags`;

A few per-cloud details are worth calling out, since they are the places the “same program” abstraction leaks:

AWS is the shortest program, because the GPU comes with the instance shape: a g4dn.xlarge is a T4 box, so there is no separate accelerator to attach. The one thing to do before you deploy is raise the Running On-Demand G and VT instances vCPU quota in your region; a fresh account starts at zero, and pulumi up fails with VcpuLimitExceeded until you do.
Google Cloud attaches the GPU explicitly with guestAccelerators, and that brings the rule that trips people up: a GPU instance cannot live-migrate, so scheduling.onHostMaintenance must be "TERMINATE" or the apply is rejected. The cloud-init also has to ride on the user-data metadata key, not startup-script, and the empty accessConfigs: [{}] is what hands the box a public IP. T4 quota is also zero on a new project.
Azure is the longest listing, because the network is à la carte: the resource group, virtual network, subnet, public IP, security group, and NIC are each their own resource before you reach the VM. One detail surprises people: a Linux VM requires an admin credential even when you never log in, so the program generates a throwaway password with random.RandomPassword rather than committing one. The NC4as_T4_v3 is a compute GPU, so the standard server (CUDA) driver from the cloud-init is the right choice here; the GRID driver is only needed for GPU-accelerated visualization workloads.

The full programs, all three Pulumi.<cloud>.yaml files, and the shared cloud-init.yaml are in the companion repo:

github.com/dirien/fully-automated-ai-inference-pulumi

Run GPU inference across three clouds

Stand up an Ollama GPU box on AWS, Azure, or Google Cloud from one Pulumi program, with short-lived credentials brokered by Pulumi ESC and no static keys to rotate.

Get started

Deploying

Create a project, install the provider for your cloud, point the stack at pulumi-idp/auth, and deploy. For AWS:

mkdir ai-inference && cd ai-inference
pulumi new typescript
npm install @pulumi/aws

Drop the AWS listing into index.ts, copy the cloud-init.yaml shown earlier into the same directory (or grab it from the companion repo), point the stack at the auth environment, and set your region:

pulumi stack init aws
# add `environment: [pulumi-idp/auth]` to Pulumi.aws.yaml (shown above)
pulumi config set region us-east-1
pulumi config set model qwen2.5:14b
pulumi up

The other two clouds are the same flow with a different provider package and a couple of config keys: npm install @pulumi/gcp with pulumi config set project <your-project-id> and pulumi config set zone us-central1-a, or npm install @pulumi/azure-native @pulumi/random with pulumi config set location eastus.

The slow part is the GPU box: it boots, installs the driver, reboots, installs Ollama, and pulls the model, all without you. Depending on the instance, the region, and the model size, plan on roughly 10 to 15 minutes before the endpoint answers. When pulumi up finishes you get the endpoints straight back as stack outputs:

Outputs:
 generateEndpoint: "http://<public-ip>:11434/api/generate"
 ollamaEndpoint : "http://<public-ip>:11434"
 publicIp : "<public-ip>"
 tagsEndpoint : "http://<public-ip>:11434/api/tags"

Testing the inference server

pulumi up returns as soon as the infrastructure exists, which is before the model has finished downloading. This is exactly where Terraform reached for that null_resource loop, and where Pulumi hands you a URL instead. To check whether the model is ready, ask Ollama what it has loaded:

curl -s $(pulumi stack output tagsEndpoint) | grep -q "qwen2.5:14b" && echo "ready" || echo "still pulling"

Once it reports ready, send it a prompt. This is the same request the Akamai post makes, pointed at your generateEndpoint output:

curl -s $(pulumi stack output generateEndpoint) -d '{
 "model": "qwen2.5:14b",
 "system": "Answer every question with a three-line poem.",
 "prompt": "Why is the sky blue?",
 "stream": false
}'

The first call is slower, because Ollama loads the model into GPU memory before it answers. Every call after that is fast, since OLLAMA_KEEP_ALIVE=-1 keeps it resident:

{
 "model": "qwen2.5:14b",
 "response": "Sunlight scatters, short waves fly,\nBlue light paints the open sky,\nViolet fades as day drifts by.",
 "done": true
}

When you are done, take it all down with one command, so the GPU stops billing:

pulumi destroy

Cost

A GPU instance is the whole bill, and you pay by the hour whether the model is busy or idle. The numbers below are rough on-demand rates for a single T4-class box left running 24/7; in practice you spin it up, use it, and pulumi destroy it, so what you actually pay tracks the hours it stays up.

Cloud	Instance	GPU	~Hourly	~Monthly (24/7)
AWS	`g4dn.xlarge`	1× T4 (16 GB)	~$0.53	~$384
Azure	`Standard_NC4as_T4_v3`	1× T4 (16 GB)	~$0.53	~$384
Google Cloud	`n1-standard-4` + 1× T4	1× T4 (16 GB)	~$0.54	~$394

These are on-demand Linux rates in a cheap US region (us-east-1, eastus, us-central1) as of mid-2026, and the GPU dominates the bill. Spot or preemptible capacity cuts it sharply if your workload tolerates interruption (often 60 to 70 percent off), Google Cloud’s sustained-use discount trims an always-on instance on its own, and the 40 GB disk adds only a few dollars a month.

A GPU box left running overnight is the expensive mistake here. Because the whole stack is declarative, the safe habit is cheap: pulumi destroy when you stop using it, pulumi up when you need it back. The state and config are version-controlled, so standing it up again is one command, not a rebuild.

Security considerations

The architecture in this post matches the Akamai original on purpose, which means it carries the same caveat: Ollama is exposed over plain HTTP on a port open to the entire internet. That is fine for a demo on a box you tear down the same day. It is not fine for anything that outlives the afternoon, and self-hosted AI endpoints get found fast: scanners like Shodan enumerate a freshly exposed one within hours, and an open Ollama instance is free compute for whoever finds it first.

The credential side, though, is genuinely better than a static-token setup, and that is the part worth keeping. There is no long-lived cloud key in the program, in your shell, or in CI; every deploy gets short-lived credentials minted through pulumi-idp/auth and thrown away when it finishes.

Concern	Akamai/Terraform demo	This deployment
Cloud credentials	Static PAT in an env var	OIDC login, short-lived, via `pulumi-idp/auth`
Inbound exposure	Port 11434 open to `0.0.0.0/0`	Same by default; one config flag scopes the CIDR
Transport	Plain HTTP	Plain HTTP (front with TLS for real use)
Readiness wait	`null_resource` + `local-exec` shell loop	A runtime `curl`, no resource

My recommendations for taking this past a demo:

Scope the inbound rule to your own CIDR instead of 0.0.0.0/0; the allowSsh flag in the program is the pattern to copy for the Ollama port
Put a reverse proxy with TLS in front of Ollama, or keep the box off the public internet entirely and reach it over a private network or a Tailscale tailnet, the way I did for the Hermes agent
Keep credentials on OIDC through ESC; never fall back to a static key to “just get it working”
Treat the box as disposable, and pulumi destroy it when you are not using it, which shrinks both the bill and the attack window

What’s next?

The program is a foundation, and the interesting work is what you layer on once a model is a pulumi up away:

Lock it down. Scope the firewall, add TLS, or move the box behind a private network so the endpoint is not on the public internet at all.
Scale the model to the GPU. qwen2.5:14b on a T4 is the entry point. Bigger models want more memory, which means an A10G, an L4, or an A100, and the only change in the program is the instance type and the model value.
Swap the serving engine. Ollama is the simplest thing that works. For higher throughput, the same shape holds with vLLM in place of Ollama in the cloud-init.
Generate the next one. Pulumi Neo can take a target like “a GPU inference box on Azure behind a private endpoint” and produce a first draft of the program, which you then review in a PR like any other change.

Conclusion

Declaring an AI inference server as code buys you the same thing it buys for any other infrastructure: you can reproduce it on any of the three clouds, version-control it, and tear it down with a single pulumi destroy. Porting the Akamai project made the contrast sharp. The credentials moved from a static token to an OIDC login that leaves nothing behind, and the readiness wait disappeared entirely once you stop treating a runtime check as a resource.

That is the line worth carrying to the next thing you deploy. Not every step in a runbook is a resource. A GPU box and a firewall rule are; waiting for a download to finish is not. Draw that line first, and the program gets shorter on its own.

If you run into issues or have questions, drop by the Pulumi Community Slack or GitHub Discussions. New to Pulumi? Get started here.

Deploy a Private Hermes Agent on Render Securely with Pulumi, Modal, and Tailscale

Engin Diri — Wed, 24 Jun 2026 00:00:00 +0000

Personal AI agents had their breakout this year. OpenClaw crossed 100,000 GitHub stars within months of launching, and self-hosting your own assistant went from a hobbyist trick to something a lot of developers actually do. I wrote up how to deploy that lobster to AWS or Hetzner back when it was everywhere.

The one people are switching to now is Hermes, the open-source runtime from Nous Research, and it caught on just as quickly. The reason shows up in every “I ditched OpenClaw for Hermes” thread: it actually learns, building up memory and writing its own skills as it goes instead of running off a static, human-written list.

Here is the part the launch videos skip. Hermes writes and runs its own code, with no human approving the commands. A model that can write code will eventually write a bad one, and the only thing between that command and your credentials is the sandbox it runs in. That is the box you do not want on the public internet. Researchers found 175,000 exposed Ollama servers sitting open in early 2026, and attackers hijack the ones they find for compute. The fix is not a better lock on the front door. It is to have no front door at all.

So this post deploys a private Hermes agent as one Pulumi program across Render, Modal, and Tailscale. The agent and its chat UI run as Render private services with no public URL, Tailscale puts the UI on your tailnet, and Modal runs the agent’s code in throwaway sandboxes. One pulumi up to stand it up, one pulumi destroy to tear it down, around $50 a month. No Makefile, no CLI escape hatches, and an honest look at the few places “nothing on the public internet” still leaks.

What is Hermes?

Hermes is the open-source, MIT-licensed agent runtime from Nous Research. It runs continuously on a server rather than living in a browser tab, and you talk to it through a chat UI or a messaging platform like Telegram, Discord, Slack, or Signal. A few things make it more than a chatbot:

It is model-agnostic. You point it at any provider. This deployment uses OpenRouter, one key that fronts 400-plus models across providers, with automatic failover when one is down.
It remembers across restarts. Memory persists on disk, so it builds up context over time instead of resetting every session.
It writes its own skills. When it works through a task, it can save that approach as a reusable skill and reach for it next time.
It schedules itself. A built-in cron runs automations on a schedule, driven in plain language.
It runs code in a sandbox. When the agent needs to execute code, it hands that off to an isolated, ephemeral container instead of running it next to its own process.

The difference from a cloud-hosted assistant is the usual self-hosting trade: it runs on your infrastructure with your keys, and your conversation history and memory stay on your disk.

“Hermes” here is the agent runtime from Nous Research, not the Hermes family of language models that share the name. The agent is model-agnostic, and the model it talks to is a separate choice you make through OpenRouter.

Prerequisites

Before getting started, ensure you have:

Pulumi CLI installed and configured
A Pulumi Cloud account
A Render account, an API key, and your workspace owner ID
A Modal account and a token pair (token ID and secret)
An OpenRouter API key with credit on the account
A container registry you can push to; this post uses GitHub Container Registry (a GHCR username and token)
Docker running locally (the image build rides on a local buildx engine), or a Docker-enabled CI runner
A Tailscale account with HTTPS enabled and a reusable auth key
Node.js 18+ (for TypeScript) or Python 3.9+ (for Python), depending on the language you pick

This guide routes models through OpenRouter, but Hermes works with other providers too. The model is just a routing string you pass (gpt-5.5 here), so any model OpenRouter supports works by changing that one value.

Understanding the Hermes architecture

Strip away the dashboard clicks and the architecture is four pieces. One rule governs all of them: what the public internet is allowed to touch.

flowchart LR
User([Your devices]) -->|"tailnet (WireGuard)"| TS["Tailscale<br/>private mesh"]
TS -->|"open-webui.&lt;tailnet&gt;.ts.net"| WebUI["Open WebUI<br/>(Render private service,<br/>joined to your tailnet)"]
WebUI -->|"private network<br/>&lt;hermes-slug&gt;:8642"| Hermes["Hermes gateway<br/>(Render private service)"]
Hermes -->|"Modal SDK at runtime"| Modal["Modal sandbox<br/>isolated code execution"]
Hermes -->|model calls| OR["OpenRouter"]

Component	Port	Description
Hermes gateway	8642	Holds secrets and memory, talks to the model, creates Modal sandboxes. The most dangerous box, so it has no public URL.
Open WebUI	8080	The chat front end. Joins your tailnet and reaches Hermes over Render’s private network.
Modal sandbox	-	Isolated, ephemeral container for running the agent’s code.
Tailscale	-	A private WireGuard mesh that makes the UI reachable to your devices only.

The agent that can run code and read secrets is never routable from the internet, and neither is the UI in front of it. There is no inbound ingress on any box (the outbound calls to the model, to Modal, and to the image registry still cross the internet, as they have to). Expressing “which box can reach which” in code is what makes the boundary auditable: the rule lives in the program rather than in a dashboard.

The four pieces split cleanly in two: two you declare as resources, and two you hand a credential and let run.

Layer	How you provision it	Status
Render services, disks, secrets	official `render-oss` Terraform provider, bridged	Declarative, via one `pulumi package add`.
Container images	`@pulumi/docker-build`	Built and pushed during `pulumi up`. No `docker` CLI.
Modal sandboxes	nothing	No Terraform provider exists. A runtime credential.
Tailnet access (Tailscale)	nothing	No provider needed. A runtime credential, like Modal.

Two of these four layers are credentials, not resources. That distinction often gets missed, and it is worth getting right: a provider can only manage what it can provision, and spinning up a Modal sandbox is a runtime act, not a provisioned resource. We come back to it at each layer.

Setting up ESC for secrets management

Deploying Hermes means handling a pile of credentials: a Render API key, registry credentials, the OpenRouter key, the Modal token pair, and the Tailscale auth key. You don’t want these hardcoded or scattered across environment variables. Pulumi ESC (Environments, Secrets, and Configuration) stores them securely and passes them directly to your Pulumi program.

Create a new ESC environment:

pulumi env init <your-org>/hermes-secrets

Add your secrets to the environment:

values:
 renderApiKey:
 fn::secret: "rnd_xxxxx"
 renderOwnerId: "tea-xxxxx"
 ghcrUser: "your-github-username"
 ghcrToken:
 fn::secret: "ghp_xxxxx"
 openrouterApiKey:
 fn::secret: "sk-or-xxxxx"
 modalTokenId:
 fn::secret: "ak-xxxxx"
 modalTokenSecret:
 fn::secret: "as-xxxxx"
 tailscaleAuthKey:
 fn::secret: "tskey-auth-xxxxx"
 tailnetDnsName: "tailxxxxx.ts.net"
 pulumiConfig:
 renderApiKey: ${renderApiKey}
 ownerId: ${renderOwnerId}
 ghcrUser: ${ghcrUser}
 ghcrToken: ${ghcrToken}
 openrouterApiKey: ${openrouterApiKey}
 modalTokenId: ${modalTokenId}
 modalTokenSecret: ${modalTokenSecret}
 tailscaleAuthKey: ${tailscaleAuthKey}
 tailnetDnsName: ${tailnetDnsName}

To find your tailnet DNS name, go to the Tailscale admin console, look under the DNS section, and find your tailnet name (e.g., tailxxxxx.ts.net). This is the domain suffix used for all machines in your Tailscale network.

Then create a Pulumi.dev.yaml file in your project to reference the environment:

environment:
 - <your-org>/hermes-secrets

This keeps your secrets out of your codebase and hands them to the program at deploy time. The Modal token pair and the Tailscale auth key never become resources; they ride straight through as encrypted env vars on the services that need them.

Generating the Render SDK

Render has no first-party Pulumi provider. What it has is an official, maintained Terraform provider, and Pulumi can turn any Terraform provider into a typed local SDK with Any Terraform Provider:

pulumi package add terraform-provider render-oss/render 1.8.0

That one command generates a render SDK you import like any other package, in whatever language your stack is written in. There is no published @pulumi/render package and no vendor to wait on.

This is also where the provider map shows its edges. The bridge wraps a Terraform provider that already exists, so it works for Render. But Modal has no Terraform provider anywhere, so pulumi package add terraform-provider modal-labs/modal has nothing to fetch. That is not a gap to work around. Modal is not a provisioned resource here; Hermes creates sandboxes at runtime through the Modal SDK, and Pulumi’s only job is to hold the token pair and hand it over. Tailscale lands in the same bucket: a Tailscale provider exists, but it manages your tailnet’s settings, not the act of putting a container on the network, which is a runtime job done inside the image.

Securing with Tailscale

The usual way to make a self-hosted UI reachable is to give it a public hostname and put an identity gate in front: an SSO provider, an access policy, a public DNS record, and a certificate to keep renewing. Tailscale collapses all of it into nothing you declare. There is no gate to bolt in front of a public URL, because there is no public URL. Open WebUI joins your tailnet and answers only to your own devices.

This is a stronger position than a public UI behind a login. Because there is no public URL, there is no login to harden and no certificate to renew, and Open WebUI answers only to your own devices.

Tailscale lives in the Open WebUI image rather than the Pulumi program. The image adds Tailscale and an entrypoint that brings up tailscaled in userspace mode (a container has no /dev/net/tun), joins your tailnet, and runs tailscale serve to expose Open WebUI on it. serve publishes over HTTPS to authenticated members of your tailnet only, where its sibling tailscale funnel would put the same port on the public internet, the one thing this design avoids.

FROM ghcr.io/open-webui/open-webui:main-slim
RUN apt-get update && apt-get install -y --no-install-recommends curl ca-certificates \
 && curl -fsSL https://tailscale.com/install.sh | sh
COPY entrypoint.sh /usr/local/bin/ts-entrypoint
ENTRYPOINT ["/usr/local/bin/ts-entrypoint"]

#!/usr/bin/env bash
set -e
PORT="${PORT:-8080}"
mkdir -p /app/backend/data/tailscale
# Userspace networking: a container has no /dev/net/tun. Keep node state on the disk
# so it survives restarts, and accept-dns=false so Render's DNS still resolves the
# private Hermes hostname.
tailscaled --tun=userspace-networking --statedir=/app/backend/data/tailscale &
n=0
until tailscale up --authkey="${TS_AUTHKEY}" --hostname="${TS_HOSTNAME}" --accept-dns=false; do
 n=$((n + 1)); [ "${n}" -ge 30 ] && { echo "tailnet join failed; check TS_AUTHKEY"; exit 1; }
 sleep 2
done
tailscale serve --bg "${PORT}" # serve Open WebUI on the tailnet over HTTPS
cd /app/backend && exec bash start.sh

The TS_AUTHKEY env var on the service, pulled from ESC, is the only thing Pulumi contributes; everything else is the image doing its own networking.

There is one prerequisite, the same one any Tailscale node needs for serve: enable HTTPS for your tailnet once in the admin console so it can get a certificate. Use a reusable auth key, since the service re-authenticates on every redeploy, and the join loop has a ceiling so a dead key fails the deploy in a minute instead of hanging it.

To generate the reusable auth key:

Go to the Tailscale admin console.
Click Generate auth key.
Enable Reusable, since the service re-authenticates on every redeploy.
Copy the key into your ESC environment as tailscaleAuthKey.

Deploying the agent

This is the part a Makefile usually handles. Two private services, two disks, every secret an encrypted env var, and both images built and pushed during the same pulumi up that creates the services. No CLI runs anywhere, though the build still rides on a local Docker engine through buildx: the docker call is gone, the daemon it talks to is not, so pulumi up needs Docker running.

Create a new Pulumi project (pick the language you prefer; the companion repo has all three):

mkdir hermes-agent-pulumi && cd hermes-agent-pulumi
pulumi new typescript

Install the dependencies and generate the Render SDK:

npm install @pulumi/docker-build @pulumi/random
pulumi package add terraform-provider render-oss/render 1.8.0

Both services need the 2 GB Standard plan. The gateway and Open WebUI each exhaust memory on the 512 MB Starter and OOM on boot.

Two facts make this clean. Render env var values are encrypted at rest, so a secret is only an env var, with no separate resource. And @pulumi/docker-build builds and pushes an image through an embedded buildx during pulumi up, so the image pipeline is declarative too. Render then pulls the pushed image by digest.

The complete program

Hermes comes first: a private Render service with no public URL, a disk for its memory, and every secret as an encrypted env var. The shared bearer key is generated in state with random.RandomBytes, the clean replacement for openssl rand -base64 32; the Modal token pair and the OpenRouter key come from config or Pulumi ESC. Render has no port field, so Hermes binds 0.0.0.0:8642 and Open WebUI reaches it at the service’s read-only slug. Open WebUI is almost the mirror image, private too, except its image joins your tailnet (the tailscaled entrypoint from the previous section) and it carries the operational env vars that keep it lean: no local embedding models, and ENABLE_PERSISTENT_CONFIG=false so your config always wins over a stale copy in its database. One render.Provider carries your renderApiKey and ownerId into every Render resource.

Here is the whole program, copy-paste ready as your index.ts (or __main__.py, or Pulumi.yaml):

import * as pulumi from "@pulumi/pulumi";
import * as dockerbuild from "@pulumi/docker-build";
import * as random from "@pulumi/random";
// Local SDK generated by: pulumi package add terraform-provider render-oss/render 1.8.0
// (no @pulumi/render is published; the SDK is bridged from the official Render provider).
import * as render from "@pulumi/render";

const cfg = new pulumi.Config();

// ---------------------------------------------------------------------------
// Configuration. Non-secret values via `pulumi config set`, secrets via
// `pulumi config set --secret` or a Pulumi ESC environment (see README).
// The Render provider reads renderApiKey/renderOwnerId (or RENDER_API_KEY/RENDER_OWNER_ID).
// ---------------------------------------------------------------------------
const region = cfg.get("renderRegion") ?? "oregon";
const modelDefault = cfg.get("modelDefault") ?? "gpt-5.5";

// Container registry the images are pushed to and Render pulls from (GHCR here).
const ghcrUser = cfg.require("ghcrUser");
const ghcrToken = cfg.requireSecret("ghcrToken"); // GHCR PAT with write:packages

// Long-lived secrets supplied by you (dashboard-minted). Hold in config or ESC.
const modalTokenId = cfg.requireSecret("modalTokenId");
const modalTokenSecret = cfg.requireSecret("modalTokenSecret");
const openrouterApiKey = cfg.requireSecret("openrouterApiKey");

// Tailscale: the only "gateway". Open WebUI joins your tailnet and is reachable
// at https://<hostname>.<tailnet>.ts.net by your devices only — never public.
// Both values are already in your ESC. Enable HTTPS once in the Tailscale admin console.
const tailscaleAuthKey = cfg.requireSecret("tailscaleAuthKey");
const tailnetDnsName = cfg.require("tailnetDnsName"); // e.g. tailc6fb4e.ts.net
const webuiHostname = "open-webui";
const webuiTailnetUrl = `https://${webuiHostname}.${tailnetDnsName}`;

// Render provider auth (from ESC: pulumiConfig.renderApiKey / pulumiConfig.renderOwnerId).
const renderProvider = new render.Provider("render", {
 apiKey: cfg.requireSecret("renderApiKey"),
 ownerId: cfg.require("renderOwnerId"),
});

// ---------------------------------------------------------------------------
// Secrets generated in state (the in-IaC replacement for `openssl rand -base64 32`).
// The shared bearer authenticates Open WebUI -> Hermes; it lands on both services.
// ---------------------------------------------------------------------------
const sharedKey = new random.RandomBytes("hermes-shared-key", { length: 32 });
const webuiSecretKey = new random.RandomBytes("webui-secret-key", { length: 32 });

// ---------------------------------------------------------------------------
// Build and push both images during `pulumi up` (no docker CLI; the provider
// embeds buildx). Render then pulls them by their digest-pinned ref.
// ---------------------------------------------------------------------------
const registryAuth = [{ address: "ghcr.io", username: ghcrUser, password: ghcrToken }];

const hermesImage = new dockerbuild.Image("hermes-image", {
 context: { location: "./hermes" },
 dockerfile: { location: "./hermes/Dockerfile" },
 platforms: ["linux/amd64"],
 tags: [pulumi.interpolate`ghcr.io/${ghcrUser}/hermes-agent:latest`],
 push: true,
 registries: registryAuth,
});

const webuiImage = new dockerbuild.Image("webui-image", {
 context: { location: "./openwebui" },
 dockerfile: { location: "./openwebui/Dockerfile" },
 platforms: ["linux/amd64"],
 tags: [pulumi.interpolate`ghcr.io/${ghcrUser}/open-webui:latest`],
 push: true,
 registries: registryAuth,
});

// Credential Render uses to pull the private GHCR images.
const ghcrCredential = new render.RegistryCredential("ghcr", {
 name: "ghcr",
 registry: "GITHUB",
 username: ghcrUser,
 authToken: ghcrToken,
}, { provider: renderProvider });

// ---------------------------------------------------------------------------
// Hermes: a PRIVATE Render service. No public URL; reachable only over Render's
// private network. A disk holds its memory and workspace. Env values are
// encrypted at rest, which is how secrets ride on Render (no separate flag).
// ---------------------------------------------------------------------------
const hermes = new render.PrivateService("hermes-agent", {
 name: "hermes-agent",
 plan: "standard", // 2 GB; the gateway OOMs on starter's 512 MB
 region: region,
 runtimeSource: {
 // Render wants the bare repo in imageUrl; the tag/digest go in their own field.
 image: {
 imageUrl: pulumi.interpolate`ghcr.io/${ghcrUser}/hermes-agent`,
 digest: hermesImage.digest,
 registryCredentialId: ghcrCredential.id,
 },
 },
 disk: { name: "hermes-data", mountPath: "/opt/data", sizeGb: 5 },
 envVars: {
 PORT: { value: "8642" },
 API_SERVER_ENABLED: { value: "true" },
 API_SERVER_HOST: { value: "0.0.0.0" },
 API_SERVER_PORT: { value: "8642" },
 API_SERVER_MODEL_NAME: { value: modelDefault },
 // Auto-approve the agent's shell/terminal hooks (no TTY in a container),
 // so it can run code in the Modal sandbox without an interactive prompt.
 HERMES_ACCEPT_HOOKS: { value: "1" },
 // Secrets (encrypted at rest on Render):
 API_SERVER_KEY: { value: sharedKey.base64 },
 OPENROUTER_API_KEY: { value: openrouterApiKey },
 // Modal is a runtime credential, not a provisioned resource: Hermes creates
 // sandboxes at runtime with this token pair. No Modal provider, no Modal CLI.
 MODAL_TOKEN_ID: { value: modalTokenId },
 MODAL_TOKEN_SECRET: { value: modalTokenSecret },
 },
}, { provider: renderProvider });

// Open WebUI reaches Hermes at its Render-assigned internal hostname (the read-only
// slug), not the bare service name. Same region + same workspace share the network.
const hermesApiBaseUrl = pulumi.interpolate`http://${hermes.slug}:8642/v1`;

// ---------------------------------------------------------------------------
// Open WebUI: also a PRIVATE Render service, but the container joins your tailnet
// (tailscaled in userspace + `tailscale serve`), so you reach it at
// https://open-webui.<tailnet>.ts.net from your devices. Never public; the gate
// is tailnet membership. It still reaches Hermes over Render's private network.
// ---------------------------------------------------------------------------
const webui = new render.PrivateService("open-webui", {
 name: "open-webui",
 plan: "standard", // 2 GB; Open WebUI OOMs on starter's 512 MB
 region: region, // must match Hermes for private networking
 runtimeSource: {
 image: {
 imageUrl: pulumi.interpolate`ghcr.io/${ghcrUser}/open-webui`,
 digest: webuiImage.digest,
 registryCredentialId: ghcrCredential.id,
 },
 },
 disk: { name: "openwebui-data", mountPath: "/app/backend/data", sizeGb: 5 },
 envVars: {
 PORT: { value: "8080" }, // tailscale serve points at this; Open WebUI binds it
 OPENAI_API_BASE_URL: { value: hermesApiBaseUrl },
 ENABLE_OPENAI_API: { value: "true" },
 ENABLE_OLLAMA_API: { value: "false" },
 DEFAULT_MODELS: { value: modelDefault },
 WEBUI_AUTH: { value: "true" },
 WEBUI_URL: { value: webuiTailnetUrl },
 // Reached only over the tailnet, so the first sign-up is the admin.
 ENABLE_LOGIN_FORM: { value: "true" },
 ENABLE_SIGNUP: { value: "true" },
 DEFAULT_USER_ROLE: { value: "admin" },
 // Always apply env config; otherwise Open WebUI persists the first run's config
 // to its DB and ignores later changes.
 ENABLE_PERSISTENT_CONFIG: { value: "false" },
 // Don't load local embedding/RAG models (a big memory saver; the agent is the brain).
 RAG_EMBEDDING_ENGINE: { value: "openai" },
 BYPASS_EMBEDDING_AND_RETRIEVAL: { value: "true" },
 OFFLINE_MODE: { value: "true" },
 HF_HUB_OFFLINE: { value: "1" },
 // Secrets (encrypted at rest):
 OPENAI_API_KEY: { value: sharedKey.base64 }, // shared bearer to call Hermes
 WEBUI_SECRET_KEY: { value: webuiSecretKey.base64 },
 // Tailscale joins this container to your tailnet and serves Open WebUI on it.
 TS_AUTHKEY: { value: tailscaleAuthKey },
 TS_HOSTNAME: { value: webuiHostname },
 },
}, { dependsOn: hermes, provider: renderProvider });

export const hermesInternalUrl = hermesApiBaseUrl;
// Open this from any device on your tailnet:
export const openWebUiTailnetUrl = webuiTailnetUrl;

import pulumi
import pulumi_docker_build as docker_build
import pulumi_random as random

# Local SDK generated by: pulumi package add terraform-provider render-oss/render 1.8.0
# (no pulumi_render is published; the SDK is bridged from the official Render provider).
import pulumi_render as render

cfg = pulumi.Config()

# ---------------------------------------------------------------------------
# Configuration. Non-secret values via `pulumi config set`, secrets via
# `pulumi config set --secret` or a Pulumi ESC environment (see README).
# The Render provider reads renderApiKey/renderOwnerId (or RENDER_API_KEY/RENDER_OWNER_ID).
# ---------------------------------------------------------------------------
region = cfg.get("renderRegion") or "oregon"
model_default = cfg.get("modelDefault") or "gpt-5.5"

# Container registry the images are pushed to and Render pulls from (GHCR here).
ghcr_user = cfg.require("ghcrUser")
ghcr_token = cfg.require_secret("ghcrToken") # GHCR PAT with write:packages

# Long-lived secrets supplied by you (dashboard-minted). Hold in config or ESC.
modal_token_id = cfg.require_secret("modalTokenId")
modal_token_secret = cfg.require_secret("modalTokenSecret")
openrouter_api_key = cfg.require_secret("openrouterApiKey")

# Tailscale: the only "gateway". Open WebUI joins your tailnet and is reachable
# at https://<hostname>.<tailnet>.ts.net by your devices only -- never public.
# Enable HTTPS once in the Tailscale admin console so `tailscale serve` can get a cert.
tailscale_auth_key = cfg.require_secret("tailscaleAuthKey")
tailnet_dns_name = cfg.require("tailnetDnsName") # e.g. tailc6fb4e.ts.net
webui_hostname = "open-webui"
webui_tailnet_url = f"https://{webui_hostname}.{tailnet_dns_name}"

# Render provider auth (from ESC: pulumiConfig.renderApiKey / pulumiConfig.renderOwnerId).
render_provider = render.Provider(
 "render",
 api_key=cfg.require_secret("renderApiKey"),
 owner_id=cfg.require("renderOwnerId"),
)

# ---------------------------------------------------------------------------
# Secrets generated in state (the in-IaC replacement for `openssl rand -base64 32`).
# The shared bearer authenticates Open WebUI -> Hermes; it lands on both services.
# ---------------------------------------------------------------------------
shared_key = random.RandomBytes("hermes-shared-key", length=32)
webui_secret_key = random.RandomBytes("webui-secret-key", length=32)

# ---------------------------------------------------------------------------
# Build and push both images during `pulumi up` (no docker CLI; the provider
# embeds buildx). Render then pulls them by their digest-pinned ref.
# ---------------------------------------------------------------------------
registry_auth = [{"address": "ghcr.io", "username": ghcr_user, "password": ghcr_token}]

hermes_image = docker_build.Image(
 "hermes-image",
 context={"location": "./hermes"},
 dockerfile={"location": "./hermes/Dockerfile"},
 platforms=["linux/amd64"],
 tags=[pulumi.Output.concat("ghcr.io/", ghcr_user, "/hermes-agent:latest")],
 push=True,
 registries=registry_auth,
)

webui_image = docker_build.Image(
 "webui-image",
 context={"location": "./openwebui"},
 dockerfile={"location": "./openwebui/Dockerfile"},
 platforms=["linux/amd64"],
 tags=[pulumi.Output.concat("ghcr.io/", ghcr_user, "/open-webui:latest")],
 push=True,
 registries=registry_auth,
)

# Credential Render uses to pull the private GHCR images.
ghcr_credential = render.RegistryCredential(
 "ghcr",
 name="ghcr",
 registry="GITHUB",
 username=ghcr_user,
 auth_token=ghcr_token,
 opts=pulumi.ResourceOptions(provider=render_provider),
)

# ---------------------------------------------------------------------------
# Hermes: a PRIVATE Render service. No public URL; reachable only over Render's
# private network. A disk holds its memory and workspace. Env values are
# encrypted at rest, which is how secrets ride on Render (no separate flag).
# ---------------------------------------------------------------------------
hermes = render.PrivateService(
 "hermes-agent",
 name="hermes-agent",
 plan="standard", # 2 GB; the gateway OOMs on starter's 512 MB
 region=region,
 runtime_source={
 "image": {
 "image_url": pulumi.Output.concat("ghcr.io/", ghcr_user, "/hermes-agent"),
 "digest": hermes_image.digest,
 "registry_credential_id": ghcr_credential.id,
 },
 },
 disk={"name": "hermes-data", "mount_path": "/opt/data", "size_gb": 5},
 env_vars={
 "PORT": {"value": "8642"},
 "API_SERVER_ENABLED": {"value": "true"},
 "API_SERVER_HOST": {"value": "0.0.0.0"},
 "API_SERVER_PORT": {"value": "8642"},
 "API_SERVER_MODEL_NAME": {"value": model_default},
 # Auto-approve the agent's shell/terminal hooks (no TTY in a container).
 "HERMES_ACCEPT_HOOKS": {"value": "1"},
 # Secrets (encrypted at rest on Render):
 "API_SERVER_KEY": {"value": shared_key.base64},
 "OPENROUTER_API_KEY": {"value": openrouter_api_key},
 # Modal is a runtime credential: Hermes creates sandboxes at runtime with this pair.
 "MODAL_TOKEN_ID": {"value": modal_token_id},
 "MODAL_TOKEN_SECRET": {"value": modal_token_secret},
 },
 opts=pulumi.ResourceOptions(provider=render_provider),
)

# Open WebUI reaches Hermes at its Render-assigned internal hostname (the read-only slug).
hermes_api_base_url = pulumi.Output.concat("http://", hermes.slug, ":8642/v1")

# ---------------------------------------------------------------------------
# Open WebUI: also a PRIVATE Render service, but the container joins your tailnet
# (tailscaled in userspace + `tailscale serve`), so you reach it at
# https://open-webui.<tailnet>.ts.net from your devices. Never public.
# ---------------------------------------------------------------------------
webui = render.PrivateService(
 "open-webui",
 name="open-webui",
 plan="standard", # 2 GB; Open WebUI OOMs on starter's 512 MB
 region=region, # must match Hermes for private networking
 runtime_source={
 "image": {
 "image_url": pulumi.Output.concat("ghcr.io/", ghcr_user, "/open-webui"),
 "digest": webui_image.digest,
 "registry_credential_id": ghcr_credential.id,
 },
 },
 disk={"name": "openwebui-data", "mount_path": "/app/backend/data", "size_gb": 5},
 env_vars={
 "PORT": {"value": "8080"}, # tailscale serve points at this; Open WebUI binds it
 "OPENAI_API_BASE_URL": {"value": hermes_api_base_url},
 "ENABLE_OPENAI_API": {"value": "true"},
 "ENABLE_OLLAMA_API": {"value": "false"},
 "DEFAULT_MODELS": {"value": model_default},
 "WEBUI_AUTH": {"value": "true"},
 "WEBUI_URL": {"value": webui_tailnet_url},
 # Reached only over the tailnet, so the first sign-up is the admin.
 "ENABLE_LOGIN_FORM": {"value": "true"},
 "ENABLE_SIGNUP": {"value": "true"},
 "DEFAULT_USER_ROLE": {"value": "admin"},
 # Always apply env config; otherwise Open WebUI persists the first run's config.
 "ENABLE_PERSISTENT_CONFIG": {"value": "false"},
 # Don't load local embedding/RAG models (a big memory saver; the agent is the brain).
 "RAG_EMBEDDING_ENGINE": {"value": "openai"},
 "BYPASS_EMBEDDING_AND_RETRIEVAL": {"value": "true"},
 "OFFLINE_MODE": {"value": "true"},
 "HF_HUB_OFFLINE": {"value": "1"},
 # Secrets (encrypted at rest):
 "OPENAI_API_KEY": {"value": shared_key.base64},
 "WEBUI_SECRET_KEY": {"value": webui_secret_key.base64},
 # Tailscale joins this container to your tailnet and serves Open WebUI on it.
 "TS_AUTHKEY": {"value": tailscale_auth_key},
 "TS_HOSTNAME": {"value": webui_hostname},
 },
 opts=pulumi.ResourceOptions(depends_on=[hermes], provider=render_provider),
)

pulumi.export("hermesInternalUrl", hermes_api_base_url)
# Open this from any device on your tailnet:
pulumi.export("openWebUiTailnetUrl", webui_tailnet_url)

name: hermes-agent-pulumi
runtime: yaml
description: Deploy a Hermes agent (Render + Modal + Tailscale) as one declarative Pulumi program.

# The Render SDK is bridged from the official Terraform provider. Run once:
# pulumi package add terraform-provider render-oss/render 1.8.0
# Then `pulumi install` regenerates it on a fresh checkout.
packages:
 render:
 source: terraform-provider
 version: 1.1.4
 parameters:
 - render-oss/render
 - 1.8.0

config:
 renderRegion:
 type: string
 default: oregon
 modelDefault:
 type: string
 default: gpt-5.5
 ghcrUser:
 type: string
 ghcrToken:
 type: string
 secret: true
 modalTokenId:
 type: string
 secret: true
 modalTokenSecret:
 type: string
 secret: true
 openrouterApiKey:
 type: string
 secret: true
 tailscaleAuthKey:
 type: string
 secret: true
 tailnetDnsName:
 type: string
 renderApiKey:
 type: string
 secret: true
 renderOwnerId:
 type: string

variables:
 registryAuth:
 - address: ghcr.io
 username: ${ghcrUser}
 password: ${ghcrToken}
 # Open WebUI reaches Hermes at its private slug; you reach Open WebUI on the tailnet.
 hermesApiBaseUrl: http://${hermes-agent.slug}:8642/v1
 webuiTailnetUrl: https://open-webui.${tailnetDnsName}

resources:
 # Render provider auth (from config or ESC).
 render:
 type: pulumi:providers:render
 properties:
 apiKey: ${renderApiKey}
 ownerId: ${renderOwnerId}

 # Secrets generated in state: the replacement for `openssl rand -base64 32`.
 hermes-shared-key:
 type: random:RandomBytes
 properties:
 length: 32
 webui-secret-key:
 type: random:RandomBytes
 properties:
 length: 32

 # Build + push both images during `pulumi up` (no docker CLI).
 hermes-image:
 type: docker-build:Image
 properties:
 context:
 location: ./hermes
 dockerfile:
 location: ./hermes/Dockerfile
 platforms: [linux/amd64]
 tags:
 - ghcr.io/${ghcrUser}/hermes-agent:latest
 push: true
 registries: ${registryAuth}
 webui-image:
 type: docker-build:Image
 properties:
 context:
 location: ./openwebui
 dockerfile:
 location: ./openwebui/Dockerfile
 platforms: [linux/amd64]
 tags:
 - ghcr.io/${ghcrUser}/open-webui:latest
 push: true
 registries: ${registryAuth}

 # Credential Render uses to pull the private GHCR images.
 ghcr:
 type: render:RegistryCredential
 properties:
 name: ghcr
 registry: GITHUB
 username: ${ghcrUser}
 authToken: ${ghcrToken}
 options:
 provider: ${render}

 # Hermes: a PRIVATE Render service. No public URL; encrypted env vars are the secret mechanism.
 hermes-agent:
 type: render:PrivateService
 properties:
 name: hermes-agent
 plan: standard  # 2 GB; the gateway OOMs on starter's 512 MB
 region: ${renderRegion}
 runtimeSource:
 image:
 imageUrl: ghcr.io/${ghcrUser}/hermes-agent
 digest: ${hermes-image.digest}
 registryCredentialId: ${ghcr.id}
 disk:
 name: hermes-data
 mountPath: /opt/data
 sizeGb: 5
 envVars:
 PORT: { value: "8642" }
 API_SERVER_ENABLED: { value: "true" }
 API_SERVER_HOST: { value: "0.0.0.0" }
 API_SERVER_PORT: { value: "8642" }
 API_SERVER_MODEL_NAME: { value: "${modelDefault}" }
 HERMES_ACCEPT_HOOKS: { value: "1" }
 API_SERVER_KEY: { value: "${hermes-shared-key.base64}" }
 OPENROUTER_API_KEY: { value: "${openrouterApiKey}" }
 MODAL_TOKEN_ID: { value: "${modalTokenId}" }
 MODAL_TOKEN_SECRET: { value: "${modalTokenSecret}" }
 options:
 provider: ${render}

 # Open WebUI: also a PRIVATE Render service, joined to your tailnet by its image.
 open-webui:
 type: render:PrivateService
 properties:
 name: open-webui
 plan: standard  # 2 GB; Open WebUI OOMs on starter's 512 MB
 region: ${renderRegion}  # must match Hermes for private networking
 runtimeSource:
 image:
 imageUrl: ghcr.io/${ghcrUser}/open-webui
 digest: ${webui-image.digest}
 registryCredentialId: ${ghcr.id}
 disk:
 name: openwebui-data
 mountPath: /app/backend/data
 sizeGb: 5
 envVars:
 PORT: { value: "8080" }
 OPENAI_API_BASE_URL: { value: "${hermesApiBaseUrl}" }
 ENABLE_OPENAI_API: { value: "true" }
 ENABLE_OLLAMA_API: { value: "false" }
 DEFAULT_MODELS: { value: "${modelDefault}" }
 WEBUI_AUTH: { value: "true" }
 WEBUI_URL: { value: "${webuiTailnetUrl}" }
 ENABLE_LOGIN_FORM: { value: "true" }
 ENABLE_SIGNUP: { value: "true" }
 DEFAULT_USER_ROLE: { value: "admin" }
 ENABLE_PERSISTENT_CONFIG: { value: "false" }
 RAG_EMBEDDING_ENGINE: { value: "openai" }
 BYPASS_EMBEDDING_AND_RETRIEVAL: { value: "true" }
 OFFLINE_MODE: { value: "true" }
 HF_HUB_OFFLINE: { value: "1" }
 OPENAI_API_KEY: { value: "${hermes-shared-key.base64}" }
 WEBUI_SECRET_KEY: { value: "${webui-secret-key.base64}" }
 TS_AUTHKEY: { value: "${tailscaleAuthKey}" }
 TS_HOSTNAME: { value: "open-webui" }
 options:
 provider: ${render}
 dependsOn:
 - ${hermes-agent}

outputs:
 hermesInternalUrl: ${hermesApiBaseUrl}
 # Open this from any device on your tailnet:
 openWebUiTailnetUrl: ${webuiTailnetUrl}

Neither box is published to the internet. The same code, plus the Dockerfiles and the Tailscale entrypoint, is in the companion repo:

github.com/dirien/hermes-agent-pulumi

Cost

Two details shape the bill. The host starts on Render’s free Hobby workspace, but a private service with a persistent disk runs on paid compute, so you need a card on file the moment you add them. There is no free-tier version of this architecture. Both services also need the 2 GB Standard plan, since the gateway and Open WebUI each OOM on the 512 MB Starter.

	Hermes gateway	Open WebUI
Plan	Standard (2 GB)	Standard (2 GB)
Compute	~$25/mo	~$25/mo
Disk	5 GB	5 GB
Public URL	none	none (tailnet only)

That lands the host around $50/month, in line with what a comparable managed setup runs. Modal and OpenRouter bill by usage on top, so the total tracks how hard the agent works rather than a flat number. Tailscale is free for a personal tailnet, so the private gateway adds nothing to that.

The persistent disk pins each service to a single instance, so there is no autoscaling, and a redeploy drops a few seconds of connectivity while the old instance lets go of the disk. For a single-user agent that is a fine trade, and it is why “stand it up in a second region” is a copy-the-config move and not a live failover.

The OpenRouter Activity dashboard: 56 requests and 738K tokens of gpt-5.5 usage, all routed through one key.

Running the deployment

With your ESC environment referenced in Pulumi.dev.yaml, deploy with:

pulumi up

This generates the Render SDK’s resources, builds and pushes both images through buildx, and creates the two private services with their disks and encrypted env vars, all in one run, with no docker CLI and no secret import hiding in a script. When it finishes, both services show up in the Render dashboard with no public URL.

The hermes-agent and open-webui private services in the Render dashboard, both deployed with no public URL.

Access the agent

Give it a few minutes after pulumi up finishes. The images still have to be pulled and the containers booted, and Open WebUI has to join your tailnet before the URL resolves. Refresh until it loads.

Open WebUI is reachable only from a device on your tailnet, at https://open-webui.<your-tailnet>.ts.net, with no domain to buy, no DNS to point, and no certificate to manage. Open it, sign up so the first account becomes the admin, and ask the agent to run a Python command. It should fire up a Modal sandbox and return the result, with nothing exposed to the public internet at any point.

Open WebUI on the tailnet URL: the gpt-5.5 agent runs sum(i*i for i in range(1, 1001)) in a Modal sandbox and returns 333833500.

When the agent runs that code, Modal builds a throwaway sandbox specifically for it. There is no Modal resource to declare, no Modal CLI to call, and no Modal provider to bridge. The Hermes service carries MODAL_TOKEN_ID and MODAL_TOKEN_SECRET as two encrypted env vars, and that is the whole footprint.

The Modal dashboard: the per-task sandboxes the Hermes agent spun up at runtime to execute code.

Verify the deployment (optional)

If the chat UI does not load, give the services a few minutes to finish their first boot, then check two things:

Render dashboard logs for the hermes-agent and open-webui services, to confirm both started and pulled their images.
Tailscale admin console (Machines), to confirm the open-webui node joined your tailnet. If it did not, the TS_AUTHKEY is likely expired or not reusable.

Keep your AI agent private

Deploy your Hermes agent across Render, Modal, and Tailscale as one Pulumi program, with secrets held in Pulumi ESC and nothing exposed to the public internet.

Get started

Security considerations

Self-hosted AI servers get found fast: scanners like Shodan and Censys enumerate a freshly exposed one within hours to days, and the code-capable ones are already under active attack. Running an always-on agent on a machine you also use for everything else invites prompt injection on top of that. The answer here is to keep the agent unroutable from the internet and to isolate the code it runs. That defeats the scanner: nobody finds an open port to attack. It does less against the agent being turned against you from the inside, which is the harder problem.

One detail makes that isolation essential. In a headless container there is no one at the terminal to approve anything, and the same is true of its cron and CI runs, so the deployment runs Hermes with HERMES_ACCEPT_HOOKS=1 to auto-accept the shell hooks it would otherwise pause on. The practical result is that the agent’s code runs with no human in the loop, which is exactly why the Modal sandbox has to carry the weight: when the model writes a bad command, the throwaway container it runs in is what stands between that command and the gateway’s credentials, not a confirmation prompt.

Concern	Typical self-host	This deployment
Gateway reachability	Public port, or behind a login	Private service, no public URL
Chat UI reachability	Public hostname + auth	Tailnet only (`tailscale serve`)
Code execution	Runs next to the gateway	Isolated Modal sandbox, torn down after
Command approval	Human approves dangerous commands	Headless: isolation replaces the prompt
Secrets in transit	Exposed if accessed over HTTP	Encrypted env vars; tailnet-encrypted access

This is private, which is not the same as fully trusted. A few places the boundary still leaks:

The shared bearer never rotates. Hermes is unroutable from the internet, but it is wide open to Open WebUI, and the two share one static bearer key. Whoever owns the front end owns the gateway behind it. Rotate it periodically.
The tailnet policy is click-ops. The ACLs, the node’s tags, and the one-time HTTPS toggle live in the Tailscale admin console, not in pulumi up, and the reusable key has no tag scoping its blast radius. The Tailscale provider can manage that policy declaratively if you want it in code.
Signup defaults to admin. Fine on a single-user tailnet, the first thing to lock down (ENABLE_SIGNUP, DEFAULT_USER_ROLE) once more than one person can reach the URL.
Every messaging client is another entry point. Each platform you wire up adds another way in, which is why the tailnet is the baseline for securing this, not the full extent of it.
A compromised agent can still reach out. The private core stops external discovery, but it does not contain the agent itself. The gateway holds the Modal and OpenRouter tokens as env vars and can call out freely, so a prompt-injected agent could read its own environment and send them somewhere; the Modal sandbox isolates the code the agent runs, not the credentials it holds. Cap and rotate the OpenRouter key, keep the Modal token off the gateway, and allowlist outbound traffic, which on Render means a forward-proxy sidecar (Render gives you dedicated outbound IPs and inbound rules, but no destination egress filter).

My recommendations:

Keep both services private; never give Hermes a public URL
Run code in Modal sandboxes, since HERMES_ACCEPT_HOOKS=1 means no human approves commands
Rotate the shared bearer key and the Tailscale auth key periodically
Cap and rotate the OpenRouter key; a leaked credential is the likelier failure mode than an open port
Use Pulumi ESC for secrets instead of hardcoding
Lock down Open WebUI signup once more than one person can reach it
Manage Tailscale ACLs and tags deliberately, in the admin console or with the Tailscale provider

What’s next?

The deployment is the foundation the agent builds on. A private agent becomes useful when it remembers earlier context, runs code without a human at the terminal, and initiates contact on its own. The disk is what makes this persist: its memory and its schedule live on that disk, so they survive every restart.

It schedules itself. Hermes ships a first-class cron you drive in plain language, either by asking the agent in the chat window or through its CLI. The schedule takes intervals, five-field cron, or one-shots:

hermes cron create "every 1h" \
 "Scrape <listing-url> for the price, compare it with seen.json in the sandbox, message me on Telegram only if it dropped, and otherwise reply exactly [SILENT]." \
 --deliver telegram

The [SILENT] reply suppresses delivery when nothing changed, so the watcher only speaks up when there is something to say. Scheduled runs skip memory by default, so polling does not pollute what the agent has learned.

It connects where you work, with the same secret pattern. Messaging clients attach to the same gateway, and their tokens ride as encrypted env vars on the Hermes service, the exact shape you used for the Modal pair:

// alongside MODAL_TOKEN_ID on the hermes service:
TELEGRAM_BOT_TOKEN: { value: cfg.requireSecret("telegramBotToken") },

plus a platform block you add to config.yaml (the image ships only the model and terminal blocks; Hermes supports the messaging platforms on top, per its docs):

gateway:
 platforms:
 telegram:
 require_mention: true

With those two pieces, a few things are genuinely doable today:

A price or uptime watcher that only speaks up when something changed, using the [SILENT] and persistent-sandbox dedup pattern above.
A daily briefing that ranks and dedups your feeds and delivers the top items each morning, the documented daily-briefing bot. If you want it to adapt to your taste over time, that is the optional Honcho memory provider’s job; it reads your preferences out of the conversation itself. The built-in file memory has no version of that.
A PR-review teammate that reviews open pull requests against your repo’s AGENTS.md conventions and saves recurring patterns as a skill, so the fifth review needs less hand-holding than the first.

A few limits are worth stating plainly, since demos tend to overstate the capabilities. Hermes does not ship Gmail, Calendar, or market integrations out of the box; those are wired up as MCP servers or skills, which is more work than a launch video implies. It curates its own memory and improves its own skills, but it does not retrain the underlying model or autonomously change its high-level strategy, so “self-learning” here means whatever it writes into memory and skills each run.

Conclusion

Deploying a Hermes agent with infrastructure as code means you can reproduce the setup anytime, version control it, and tear it down with a single pulumi destroy. Keeping both services private, with no public URL and no exposed ports, means the code-executing core never sits on the internet for someone to find.

The part worth carrying to the next service you deploy: not every layer has a provider, and that is fine. The task is to tell a real resource apart from a runtime dependency wearing a costume. Render and the images are resources you declare; Modal and Tailscale are credentials you hand over. Draw that map first, and the program follows from it.

To generate that map instead of writing it by hand, Pulumi Neo can take a target architecture like this one and produce a first draft of the program. The agent side of this shift is covered in How Building AI Agents Has Changed in 2026 and the platform side in Agent Sprawl Is Here. Your IaC Platform Is the Answer..

If you run into issues or have questions, drop by the Pulumi Community Slack or GitHub Discussions. New to Pulumi? Get started here.