Hermes Agent on Oracle Cloud Free Tier: A UK Guide (2026)

Oracle Cloud Free Tier gives UK businesses a genuinely free server for Hermes Agent — one OCPU of Ampere A1 compute, six gigabytes of RAM, twenty terabytes of egress per month, no credit card charge ever. We have run Hermes on Oracle Cloud since 3 April 2026 (40+ days at time of writing). Honest disclosure: our own production deployment is on a paid Oracle x86 shape (2 vCPU / 12GB) because we co-locate Hermes Workspace + Dashboard + Syncthing — but the Always Free Ampere A1 spec described in this guide is what we recommend for Hermes-only pilot deployments and is genuinely production-grade for that use case. Real measured uptime over the 40 days: 93.5% including a 62-hour incident, 99.7% excluding it. The full numbers are in our 40-day production cost teardown. This guide covers the specific Oracle setup steps, the UK-relevant region choice, the network rules that matter, and the latency reality.

Last updated: May 2026 · Covers Oracle Cloud Always Free + Hermes Agent v0.13 · Based on 40+ days of live production data

• Oracle Cloud Free Tier is genuinely free forever — not a 12-month trial — and the Ampere A1 instance is generous enough to run Hermes
• Pick the UK South (London) region for sub-20ms latency to UK users
• Two firewall rules to open: SSH (port 22, your IP only) and outbound HTTPS (already open by default)
• The free tier has soft limits — Oracle will not bill you, but they may stop your instance if you exceed compute quota over time
• Real performance from 40 days (our paid x86 deployment, similar profile expected on Always Free Ampere A1 for Hermes-only): ~1.5 GB RAM steady state, 360MB peak observed, ~890MB/day network egress

Why Oracle Free Tier specifically

Three other free-tier options exist; Oracle is materially better for this use case.

AWS gives you 12 months of free t2.micro — too small for Hermes (1 GB RAM is tight) and time-limited. Google Cloud gives you 1 small e2-micro instance free forever in select US regions only (no UK option). Azure gives you 12 months of B1S, also time-limited.

Oracle gives you Always Free — meaning no time limit, ever — plus the Ampere A1 instance which is significantly larger than the alternatives:

The 6 GB RAM is the thing that matters for Hermes. Hermes itself runs at ~1.5 GB steady state, but during heavy use (multi-agent sessions, large model context windows, log rotation) it bursts to 3-4 GB. A 1 GB instance OOM-kills under load. The 6 GB Ampere A1 has comfortable headroom.

Oracle has a reputation for aggressive time-limited promotions but Always Free is a separate, well-defined offer. Read their tier definitions to confirm. Multiple Ampliflow client deployments have used Always Free Ampere A1 instances without surprise charges; for our own headquarters deployment we upgraded to paid x86 because we co-locate other services on the same box.

Region choice for UK users

Pick UK South (London) — region code `uk-london-1`. The latency from a UK user to a London-region instance is 5-15ms. To the next-closest free-tier-capable Oracle region (eu-frankfurt-1) it is 20-30ms. To US East it is 80-120ms.

For Hermes Agent, where the user-facing latency is dominated by the model API call (200-2000ms to Anthropic/OpenAI/etc.), the 10-20ms saved on the gateway hop does not transform user experience — but it does make admin SSH sessions feel snappy and reduces failure rates on tight-timeout integrations.

The London region also has the right legal-residency story for UK businesses: Oracle UK Cloud Services Limited is the contracting entity, data is held in UK data centres, and the Oracle DPA includes UK GDPR provisions. For non-regulated workloads this is irrelevant; for FCA, NHS, or law-firm work it removes a question your DPO would otherwise ask.

Setup walkthrough — from sign-up to running Hermes

Sign up to running gateway: about 45 minutes.

Step 1 — Sign up for Oracle Cloud

Go to oracle.com/cloud/free. Sign up with a real email + a credit card (used only for verification — Always Free workloads never bill). Pick UK South (London) as your home region. The home region cannot be changed later, so pick deliberately.

The sign-up flow takes 5-10 minutes plus a short delay for account activation (usually under an hour, occasionally up to 24).

Step 2 — Create the Ampere A1 instance

In the Oracle Cloud console:

1. Compute → Instances → Create Instance
2. Name: hermes-agent (or your choice)
3. Image: Ubuntu 22.04 Minimal (smaller image, faster boot, easier to reason about)
4. Shape: VM.Standard.A1.Flex (the Ampere A1 — do not pick the AMD/Intel ones, they're billable)
5. Customise the OCPU/memory: 1 OCPU, 6 GB memory (the free-tier maximum on a single instance)
6. Networking: leave defaults — Oracle creates a VCN + subnet + internet gateway automatically
7. SSH keys: paste your public key (or generate a new pair via the console — back up the private key)
8. Click Create

Provisioning takes 1-3 minutes. The instance gets a public IPv4 address — note it.

Step 3 — SSH in and update

`bash ssh ubuntu@<your-instance-public-ip> sudo apt update && sudo apt upgrade -y sudo apt install -y curl git build-essential python3 python3-pip python3-venv `

The Always Free Ampere image ships pretty bare. The above gets you ready for the Hermes install.

Step 4 — Open the network rules

Oracle's default Network Security List blocks most inbound traffic. You need:

• Port 22 (SSH) — restricted to your home IP, not 0.0.0.0/0
• Port 9119 — only if you want browser access to the dashboard (we tunnel this over SSH instead — more secure)

In the Oracle console: Networking → Virtual Cloud Networks → Default VCN → Default Security List → Add Ingress Rules.

Outbound traffic is open by default. Hermes needs to reach api.anthropic.com, api.openai.com, web.whatsapp.com, and any MCP servers you connect to — all standard HTTPS, no extra rules needed.

Step 5 — Install Hermes

Follow the main deployment guide for the install + systemd hardening + WhatsApp link steps. The install takes 5-10 minutes from a clean Ubuntu instance.

Step 6 — Test the gateway

`bash hermes status journalctl --user -u hermes-gateway.service --no-pager -n 20 `

If the gateway is active (running) and the journal shows "Gateway started successfully", you are done. Send your first test message to the agent via your configured channel (WhatsApp / Telegram / CLI).

What we have learned from 40+ days in production

Performance reality

Steady-state metrics from our production deployment:

• CPU usage: 5-15% baseline, spikes to 80% during multi-agent reasoning sessions
• RAM usage: 1.4 GB steady state, peaks of 3.5 GB during heavy load
• Disk I/O: trivial (<1 GB written per day, mostly logs)
• Network egress: 2-3 GB per month (well under the 10 TB free allowance)
• Uptime: 93.5% across 40 days (99.7% excluding the 62-hour incident in late April — see 40-day teardown)

The compute headroom means Hermes never throttles — what limits performance is model-provider latency, not local resources.

What broke

Oracle's free tier has one well-documented gotcha: Oracle reserves the right to reclaim Ampere A1 capacity if you stop the instance. They will not bill you, but if you stop the instance to save resources during a quiet period, when you start it again you may get an "out of host capacity" error. The fix is to keep the instance running 24/7 (which costs nothing because of the always-free quota).

We have not seen this in 40+ days because we never stop the instance. Worth knowing if you intended to use stop/start as a cost-control pattern.

What surprised us

The Ampere A1 chip is genuinely good. Single-thread performance is comparable to a mid-tier x86 cloud instance. For Hermes — which is heavily I/O bound and not CPU-bound — the difference vs an Intel/AMD instance is invisible.

Limits and when to graduate

Free Tier limits to know about:

• 1 OCPU and 6 GB RAM total across all your free Ampere instances. You can split it (two 0.5 OCPU instances) but most setups want a single instance.
• 200 GB block storage total across all free volumes — generous for Hermes
• 10 TB outbound transfer per month — irrelevant for Hermes' modest egress

When to upgrade to a paid instance:

• You're running multiple Hermes specialist harnesses and need >6 GB RAM
• You need more than 1 OCPU for parallel agent workloads
• You need a region Oracle doesn't offer free in
• Your model-provider latency is hurting and you want to be in a co-located region (often US East for Anthropic, where the Free Tier is also offered but the UK latency hit makes it irrelevant)

The cheapest upgrade path is Hetzner CX22 at ~£4/month — a 2 vCPU, 4 GB RAM x86 instance in their Falkenstein or Helsinki data centres. Higher peak performance than Free Tier Ampere, slightly worse RAM headroom. Or stay on Oracle and pay for an additional A1 instance (Oracle's paid pricing is reasonable).

Frequently asked questions

Is Oracle Cloud Always Free actually free forever?

Yes — Always Free is distinct from Oracle's 30-day trial. The Ampere A1 1 OCPU + 6 GB RAM instance, 200 GB block storage, 10 TB egress per month, and a few other resources are free with no time limit. Oracle has not changed the offer since 2021. Worth re-checking their terms periodically, but the offer has been stable.

Will Oracle bill me by surprise if I exceed the free tier?

Only if you create non-free resources. Always Free resources are clearly labelled in the console with an "Always Free" badge. If you stay within those, you cannot be billed. If you accidentally create a billable resource (e.g. a paid VM shape), Oracle's first invoice gives 30 days notice — you can delete the resource before being charged.

Can I run multiple Hermes instances on one Free Tier server?

Yes — each Hermes instance is a separate ~/.hermes/ directory + its own systemd service. Two instances at idle use about 3 GB RAM total — comfortable on the 6 GB Free Tier. We commonly run a "primary" + "test" instance on the same server for safe config-change validation.

What's the latency from London to Anthropic's API?

Roughly 80-120ms one-way to Anthropic's US-region servers (they don't yet offer a UK region for the standard API). This is the biggest latency contributor in any Hermes session — Oracle UK vs US East would change Hermes-side latency by 60-70ms but the model-side latency dominates.

For latency-sensitive workloads, deploy Hermes via Bedrock with the eu-west-2 (London) Anthropic models — single-digit ms regional latency. Comes at API-rate cost vs subscription, covered in our Hermes deployment guide.

Is the Ampere A1 ARM architecture going to cause compatibility issues?

Almost never. Hermes is Python + Node.js — both have first-class ARM64 support. Docker containers built for ARM64 work natively. The only edge case is x86-only binaries (rare for modern tooling). If you see "exec format error", you're trying to run an x86 binary on ARM — rebuild for ARM64.

Can I run Hermes Agent on Oracle Free Tier without a credit card?

You need a credit card for Oracle account verification, but they will not charge it for Always Free resources. If you genuinely cannot get a card, the cheapest paid alternative is Hetzner CX22 at ~£4/month (no card required for invoice billing in some EU countries).

What if Oracle deprecates the Always Free tier?

Always Free has been stable since 2021. If Oracle ever reduces or removes it, the migration path is straightforward: snapshot the Hermes data directory (~/.hermes/), spin up a Hetzner or DigitalOcean VPS (~£4-8/month), restore the snapshot, point your DNS at the new IP. About 30 minutes of downtime. We document the snapshot pattern in the main deployment guide.

Related reading

• ↑ How to Deploy Hermes Agent — UK Business Complete Guide — the foundational deployment pillar
• ↔ Hermes Agent Security & GDPR for UK Business — the compliance posture once you have the server
• ↔ Hermes Agent Monitoring, Uptime & Reliability in Production — the monitoring stack that goes on top of this server
• ↔ What is Hermes Agent? A UK Business Guide — the foundational pillar

What should you do next?

The Oracle setup itself is straightforward. The hard part is the systemd hardening, monitoring, and recovery patterns that make the deployment trustworthy.

See how Ampliflow uses Hermes-style agents in client work →

Or to scope your specific Oracle Cloud Hermes deployment: Book a free Hermes deployment review →