53 questions, answered.

Getting started

What is deputyOS?

deputyOS is a batteries-included Linux appliance image for running open-source personal AI assistants on hardware you own — a Raspberry Pi, a mini-PC, a cloud VPS, or a desktop VM. You flash the image, boot it, complete a wizard, and your assistant is online in about ten minutes. Unlike single-box home-server operating systems, it also manages several machines from one console.

How is deputyOS different from ChatGPT?

ChatGPT is a hosted product; deputyOS runs an assistant on hardware you control. Your conversations and files stay on your disk, there is no per-seat pricing, and it works with no internet connection at all. The trade is capability: frontier hosted models outperform anything you can run locally, so deputyOS lets you bring your own API key and call them when you want to.

How is deputyOS different from Ollama?

Ollama is a model runtime — it serves models over an API. deputyOS is the complete assistant on top of one: channels, memory, tools, scheduling, security baseline, signed updates and backup, packaged as an OS image. deputyOS can use Ollama as its model provider; they are layers rather than alternatives.

How long does setup take?

About ten to fifteen minutes on most targets, from downloading the image to a working assistant. Cloud targets are quickest at around ten minutes; a Raspberry Pi takes longer mostly because the first boot expands the filesystem.

Do I need to know Linux?

No. The first-boot wizard runs in a browser and covers profile choice, model provider credentials and remote access. There is no SSH step and no terminal required for normal setup or operation.

Which assistant profile should I pick?

OpenClaw if you are unsure — it is the generalist, supports 23 channels, and is the best first pick. Hermes if continuity across conversations matters most, since it is built around memory and recall. Khoj if searching your own notes, PDFs and email is the actual goal.

Can I change profile later?

Yes, with `deputyctl profile switch <id>`. No reflash is required — the active profile is a symlink, and switching is a single command.

Hardware

What hardware do I need?

Anything from a Raspberry Pi 4 to a mini-PC or a $5 cloud VPS. There are 15 production targets; a Raspberry Pi 5 with 8 GB is the recommended dedicated option at around $80, and a used mini-PC with 16 GB is the best value if you want to run models locally.

Can I run it on a Raspberry Pi?

Yes. The Pi 5 is the recommended board and runs the standard tier comfortably with 8 GB, or the rich tier with 16 GB. The Pi 4 works on the lean tier with a cloud model provider, but cannot run local models.

Do I need a GPU?

Only if you want to run models locally at speed. Without one, use a cloud model provider — most deployments do, and the assistant works identically. A discrete GPU with 8 GB or more of VRAM makes local inference genuinely fast, but needs passthrough configured if you are running in a VM.

How much RAM do I need?

4 GB runs the lean tier with a cloud model provider. 8 GB is comfortable for the standard tier. 16 GB or more is needed for the rich tier and to run useful local models — a 7–8B model at 4-bit quantization needs roughly 4–5 GB just for weights.

Can I try it without buying hardware?

Yes. On Windows it imports as a WSL2 distribution, and on macOS it runs as a VM under UTM or OrbStack. Both are free and take about ten minutes. The profile you configure transfers unchanged when you later move to dedicated hardware.

What does it cost to run?

A Raspberry Pi 5 draws 3–7 W, roughly £10–15 a year in electricity. A mini-PC idles at 6–15 W, so £15–35 a year. A Hetzner VPS is about €4 a month. Model API usage is separate and paid directly to your provider.

Privacy and security

Does deputyOS see my data?

No. Conversations, files and memory stay on your device. Provider API keys live at `/etc/deputyos/secrets.env` with mode 0600 and are never sent to api.deputyos.com. The optional cloud tier stores only client-side encrypted ciphertext, which we cannot read.

Does deputyOS collect telemetry?

No. There is no telemetry, opt-out or otherwise. You can verify this by watching egress with tcpdump or at your router.

Can it work completely offline?

Yes. Build with `AIRGAP=1` and the image bakes model weights via llama.cpp, points apt at a local mirror, and denies egress at nftables except RFC1918 and mDNS. It runs with `-net none`. The test is simple: boot it with the cable out and see whether it answers.

What security is enabled by default?

AppArmor in enforce mode, ufw default-deny, fail2ban, hardened sysctls, ClamAV signatures pre-loaded, and Magika hooked into the agent's file-upload path. None of it requires a checkbox — the baseline is on from first boot.

How are backups encrypted?

With age, client-side, before anything leaves the device. The storage provider only ever holds ciphertext. The corollary is important: if you lose the key, the backups are unrecoverable, so key custody is made explicit during setup.

Can I verify what's in the image?

Yes. Builds are reproducible, so you can rebuild from source and compare hashes rather than trusting the publisher. Every release ships minisign signatures and SLSA provenance attestations, and an SBOM enumerating every component.

Do I have to open ports on my router?

No. Each device holds an outbound tunnel to the control plane, so it is reachable at a per-device URL with nothing forwarded and no inbound ports opened. You can also use WireGuard or Tailscale instead if you prefer.

Models and cost

Which model providers are supported?

Anthropic, OpenAI, OpenRouter and Google AI Studio, plus any OpenAI-compatible endpoint including a local Ollama instance. You bring your own key and pay the provider directly — deputyOS never proxies or meters model traffic.

Who pays for the model usage?

You do, directly to the provider. deputyOS is the messenger, not a reseller, so there is no markup and no metering in between. Running a local model instead makes this cost effectively zero.

Can I run models locally?

Yes, given enough memory. 16 GB comfortably runs 7–8B parameter models at 4-bit quantization; 32 GB reaches 13–14B. A Raspberry Pi cannot do this usefully — it has no GPU offload — so Pi deployments typically use a cloud provider.

How do I stop it running up a bill?

Cost guardrails and quiet hours are built in and worth configuring on first setup. The console shows per-device spend, so a runaway workload is visible rather than a surprise at the end of the month.

What is quantization and which should I use?

Quantization stores model weights at lower precision to reduce memory. 4-bit (Q4_K_M) is the usual sweet spot: a 7B model drops from about 14 GB to around 4 GB with quality losses most people cannot detect. A larger model quantized to 4-bit generally beats a smaller model at full precision.

Running several machines

Can deputyOS manage more than one device?

Yes, and this is the main thing that distinguishes it from single-box appliance operating systems. Linking devices to an account gives one console with live status for every machine, updates pushed to a selection, allow-listed commands run across a group, and per-device cost tracking.

Do I need an account?

No. Every feature works without one, including backup to your own B2 or R2 bucket and remote access via your own Cloudflare Tunnel or Tailscale. An account is only needed for the managed tunnel, managed encrypted backup and the fleet console.

How do updates work?

Updates land as whole signed images into an inactive A/B slot, then the device reboots into it. If the new slot fails its watchdog, the bootloader returns to the previous one automatically. Cloud targets use platform snapshots instead, since they have no A/B partitioning.

What happens if an update breaks something?

It rolls back on its own. The watchdog detects a failed boot and the bootloader returns to the previous slot, so a bad update costs you a reboot rather than an evening.

Pricing and licensing

Is deputyOS free?

Yes. It is Apache-2.0 licensed, and running deputyOS is free forever on any number of devices — all of them tracked in the console with live status and spend, no account required. You pay only for devices you want hosted, from $5 per device per month, which activates a long-lived managed tunnel, encrypted backup held by us, and migration between your hardware and the cloud.

What's in the paid tier?

Hosting a device activates its long-lived managed tunnel at its own URL, client-side encrypted backups that we store and retain, one-command restore onto replacement hardware, migration to and from cloud instances, Web Push, audit evidence and CVE tracking. Everything else — including running unlimited devices, tracking them, and backing up to your own bucket — stays free.

Is there a business or team plan?

Not yet. Email hello@deputyos.com if you need one — knowing what teams actually require is how it gets built.

Can I use it commercially?

Yes. Apache-2.0 permits commercial use, modification and redistribution, including in closed-source products, subject to the licence terms.

How many devices can I run?

As many as you like, at every tier, and all of them appear in the console with live status, version and spend. Running and tracking devices is free and unlimited. You pay per device only for the ones you want hosted, and pricing steps down as you add more.

What's the difference between a device and a hosted instance?

Hosting is about management, not location. A hosted device is one where deputyOS holds its encrypted backup, carries its tunnel and manages its agent — that device can be a Pi in your kitchen, a mini-PC at the office, or a VPS you rent. A self-managed device runs the same software and is tracked in the same console, but you handle its backup and remote access.

Do you charge for model usage?

No. You bring your own provider key and pay Anthropic, OpenAI, OpenRouter or Google directly at their rates. deputyOS never proxies, marks up, or takes a cut of model traffic, which is also why running a local model reduces that cost to effectively zero.

What if I need more than 5 GB of backup?

Backup allowance is per hosted device rather than a single account pool. If you need more than a hosted device includes, either point that device's backup at your own B2, R2 or S3-compatible bucket — free and unlimited, still encrypted client-side — or take Enterprise and have the allowance sized to you. Self-managing backup does mean giving up one-command migration, since we no longer hold the state.

Can I buy extra devices without going to Enterprise?

Yes — hosting is priced per device with volume bands, so you add devices as you need them rather than jumping tiers. $5 each for the first three, $4 each for the next seven, $3 each after that. Past 25 hosted devices it becomes an Enterprise conversation, because what you need at that scale changes shape. Self-managed devices remain unlimited and free regardless.

Can I move a device between my hardware and the cloud?

Yes, for hosted devices. Because deputyOS holds a signed image and an encrypted state backup, a hosted device can be reconstituted elsewhere on demand — migrate it from a Pi to a cloud instance, or bring it back home, and it resumes with its memory intact. Cloud instances are a destination a hosted device can move to rather than a separate product.

What exactly does 'hosted' mean?

That deputyOS holds the device's state: client-side encrypted backups, a managed tunnel at its own URL, and managed agent services. It says nothing about whose hardware it runs on. A hosted Raspberry Pi in your house and a hosted cloud VPS are the same thing commercially.

What does paying actually activate?

The managed tunnel, primarily. Free accounts get 15-minute tunnel sessions, which is enough to try remote access or reach a device occasionally; hosting a device makes its tunnel long-lived at its own URL. It also activates encrypted backup held by us, one-command restore, and the ability to migrate that device between your hardware and a cloud instance.

Why per device rather than a flat fee?

Because each hosted device is something we actively run: its tunnel stays connected, its backups are stored and retained, and its state is kept ready to move. That cost scales with the number of devices, so the price does too — and it steps down in bands, so a fleet is not priced like the first machine.

Is there a bandwidth limit on the tunnel?

Yes, published rather than buried: 1 TiB per hosted device per calendar month, counted in both directions. That is deliberately far above interactive use — a full day of terminal work is a few megabytes, and you would need to move about 35 GB every day to reach it. It exists because tunnel traffic is the one hosted cost that does not scale with a flat per-device fee.

What happens if a device exceeds fair use?

We get in touch. Nothing is cut off automatically. In practice crossing 1 TiB means bulk file movement is being relayed through us, which is better done directly between machines — that is usually the fix rather than a bigger plan.

Is the backup allowance shared across my devices?

No, it is per device. Each hosted device gets its own allowance and its own retention, so paying for several devices gives you several allowances rather than one pool split between them.

Enterprise and compliance

Can we deploy deputyOS across an organisation?

Yes. Devices link to one fleet console with live status, staged updates, allow-listed remote commands and per-device cost attribution, and each holds an outbound tunnel so no site needs inbound firewall changes. Enterprise terms — unlimited devices, air-gapped support, procurement evidence and invoicing — are arranged directly rather than listed.

Is deputyOS ready for enterprise use?

Partly, and it depends what you need. The technical posture is strong: signed reproducible builds, SBOM, CVE tracking, audit evidence and air-gapped operation. What does not exist yet is a third-party security audit, a formal SLA, and SSO or SCIM. Organisations deploying today do so understanding those gaps.

Do you provide an SBOM?

Yes, in CycloneDX format, generated as part of the build rather than assembled afterwards. It covers the entire image rather than only the application layer, which is what makes a new CVE answerable in seconds instead of days.

How do you handle vulnerability disclosure?

Email security@deputyos.com. We aim to acknowledge within two working days and to ship a fix or documented mitigation within 90 days, coordinating disclosure with the reporter. Please do not open a public issue for a security report.

Has deputyOS been independently audited?

Not yet. The trust chain has had no third-party review, and commissioning one is part of procurement readiness rather than something already done. We would rather say that plainly than imply otherwise.

Can it run in a classified or air-gapped environment?

Yes. Built with AIRGAP=1, the image carries model weights and signature databases inside and denies egress at nftables except RFC1918 and mDNS. Updates move as whole signed images on physical media. The test is to boot it with the cable out and confirm it still works.

What compliance evidence can we export?

Audit batches are captured as compressed evidence and queryable with SQL, alongside CVE posture enriched from OSV, NVD, CISA KEV and EPSS, and signed build provenance. What a given regime requires varies, so tell us which one you report against.

Do you offer support contracts or an SLA?

Not formally yet. Enterprise arrangements include a named contact who knows your deployment, but a contractual SLA is not something we offer today. If that blocks approval, tell us on what timeline you need it.

Still stuck?

The documentation goes deeper, the glossary defines the vocabulary, and GitHub Discussions is where questions get answered.