# UnixShells

The place where AI agents live — a real Linux machine for Claude Code

Source: https://wheredoirunthis.com/products/unixshells · JSON: https://wheredoirunthis.com/api/products/unixshells · Read on 2026-06-06

## At a glance

- Type: Persistent environment
- Kind: Hosted
- Isolation: Virtual machine
- Free tier: No
- Starting price: $12/mo
- GPU: No
- Regions: Not disclosed
- Coverage: 6 of 48 assessed capabilities supported · 13 not assessed
- Agent readiness: read 4 of 11 signals, drive 1 of 11 signals

## Links

- Homepage: https://unixshells.com/
- Pricing: https://unixshells.com/#pricing
- Repository: https://github.com/unixshells

## Pricing

- Model: Flat-rate monthly per VM, no metered billing
- Unit: per managed VM per month
- Free tier: No

| Tier | Price | Includes | Notes |
| --- | --- | --- | --- |
| Relay | Free (with any account) | SSH to your OWN machines through any NAT; 9 relay servers / 3 continents; E2E encrypted. Not a managed VM. |  [source](https://unixshells.com/relay.html) |
| Shell+ | $12/mo ($120/yr) | Managed KVM VM: 4 GB RAM, 2 vCPU, 25 GB disk; 200+ tools pre-installed; relay access included. Always-on flat rate. |  [source](https://unixshells.com/) |

- Overages: none (flat-rate, no metered billing)
- Card required: Not documented
- Source: https://unixshells.com/

## Runtime

- **GPU & accelerators**: No — "No GPU offering; small fixed VM shapes (4GB/2vCPU/25GB)." (source: site) (confidence: high)
- **BYOC (Bring your own cloud)**: na — "Managed VMs on UnixShells infra; no BYOC." (source: hosted service) (confidence: high)
- **Secrets management**: No — "No platform secrets store; you manage your own secrets inside the VM." (source: site) (confidence: medium)
- **Storage options**: 10-25 GB local disk per shell (Shell+ = 25 GB); home dir persists, overlay resets — "Shell+ = 25 GB disk; persistent home dir." [source](https://unixshells.com/) (confidence: high)
- **Supported workloads**: ai-coding-agents, persistent-dev-env, always-on-services, docker, web-apps-via-preview-url — "Built for AI coding agents (Claude Code/Codex/OpenCode/OpenClaw/Aider pre-installed); real Docker; systemd services; web apps via preview URL." [source](https://unixshells.com/) (confidence: high)
- **Persistence**: persistent home dir; overlay resets on restart; sessions persist indefinitely via latch — "'The overlay resets on restart, but your home directory persists.' latch keeps terminal sessions alive across disconnects/reboots/network changes." [source](https://unixshells.com/) (confidence: high)
- **Data jurisdiction**: not disclosed for shells; relay = 9 servers / 3 continents — "Relay network spans 9 servers on 3 continents; the managed-shell region(s) are not disclosed." [source](https://unixshells.com/) (confidence: low)
- **Resource limits & quotas**: Shell+ = 4 GB RAM / 2 vCPU / 25 GB disk, flat $12/mo (no metered limits); Relay free — "Flat-rate, no metered billing; fixed VM shape." [source](https://unixshells.com/) (confidence: high)
- **SDK languages**: Not documented — "No platform SDK. Open-source components (latch, mosh-go/dart, mobile) are tools/clients, not a programmatic platform SDK." (source: site) (confidence: high)
- **Network & egress controls**: user-controlled (full in-VM iptables/ip), not a platform policy — "Full Linux network stack: iptables, ip link/addr, NAT, TUN/TAP, WireGuard/Tailscale. Egress is the user's own VM firewall, not a platform-enforced egress policy." [source](https://unixshells.com/) (confidence: medium)
- **Snapshots & forks**: No — "Memory-snapshot is a BOOT mechanism (200+ tools loaded into a snapshot so boot is sub-second) — not a user-facing snapshot/fork primitive of the running VM." [source](https://unixshells.com/) (confidence: medium)
- **Deployment methods**: ssh, web-terminal, mobile-app — "Access via standard SSH, Mosh-over-WebTransport web terminal, and iOS/Android app. No git-deploy / no programmatic create API." [source](https://unixshells.com/) (confidence: high)
- **Isolation**: Virtual machine — "'Each shell is a real KVM virtual machine with its own kernel.' Built on their own MIT-licensed minimal VMM (`unixshells/clone`). 'Isolation: KVM VM with own kernel' (vs containers/gVisor on cloud IDEs)." [source](https://unixshells.com/) (confidence: high)

## Capabilities

### AI & agents

| Capability | Status | Evidence |
| --- | --- | --- |
| Agent skills | Not supported | "No installable agent-skills surface." [source](https://unixshells.com/) |
| AI app builder | Not supported | "No prompt-to-app builder." [source](https://unixshells.com/) |
| Autonomous QA agent | Not assessed |  |
| Built-in coding agent | Not supported | "Pre-installs THIRD-PARTY agents (Claude Code/Codex/Aider) — no first-party coding agent." [source](https://unixshells.com/) |
| BYOK for LLMs | Not applicable | "You bring your own agent keys inside your VM; not a platform BYOK feature." [source](https://unixshells.com/) |
| Credential-injection proxy | Not assessed |  |
| LLM gateway | Not supported | "No LLM gateway." [source](https://unixshells.com/) |
| MCP client | Not supported | "No MCP client product." [source](https://unixshells.com/) |
| MCP-server hosting | Not supported | "Could run an MCP server in the VM, but not advertised as a feature." [source](https://unixshells.com/) |
| MCP server | Not supported | "No first-party MCP server." [source](https://unixshells.com/) |
| Model inference API (first-party) | Not assessed |  |
| Preinstalled coding agent | Not assessed |  |

### Networking & delivery

| Capability | Status | Evidence |
| --- | --- | --- |
| Automatic HTTPS | Partially supported | "Preview-URL subdomains per shell; custom domains via 2 DNS records; certbot available in-VM (Let's Encrypt is DIY for custom domains)." [source](https://unixshells.com/) |
| Custom domains | Supported | "'Bring your own domain with two DNS records — no tunnels needed.'" [source](https://unixshells.com/) |
| Direct public IPv6 | Supported | "'Every shell has a public IPv6 address'; IPv4 + IPv6; full TUN/TAP." [source](https://unixshells.com/) |
| Edge KV store | Not applicable | "Not an edge runtime." [source](https://unixshells.com/) |
| Egress controls | Partially supported | "Full in-VM iptables/ip/NAT/TUN-TAP — user-configured firewall, not a platform-enforced egress policy." [source](https://unixshells.com/) |
| Email send / receive | Not supported | "No email product (you could run your own in the VM)." [source](https://unixshells.com/) |
| Public preview URLs | Supported | "'Every shell gets a unique subdomain ... share the URL.' dev preview URL per shell." [source](https://unixshells.com/) |
| Private networking | Not assessed |  |
| Built-in reverse proxy | Supported | "'Every shell gets a unique subdomain. Start a web server on any port, share the URL.'" [source](https://unixshells.com/) |

### Organisation & security

| Capability | Status | Evidence |
| --- | --- | --- |
| Built-in auth | Not applicable | "OPAQUE auth is the platform login; not auth-as-a-service for your workloads (it's your VM)." [source](https://unixshells.com/) |
| BYOC | Not applicable | "Managed VMs on UnixShells infra; no BYOC." [source](https://unixshells.com/) |
| Deploy into customers' VPCs | Not assessed |  |
| Secrets management | Not supported | "No platform secrets store; manage your own in the VM." [source](https://unixshells.com/) |
| Self-host / OSS core | Partially supported | "VMM (`clone`, MIT, ★327) + latch + mosh + mobile are MIT-open and reusable, but there is NO deployable self-hostable UnixShells platform — 'the shells and relay are just infrastructure you pay for'." [source](https://github.com/unixshells/clone) |
| Sharing | Partially supported | "Preview URLs share a web app by link; no first-class resource-sharing/collaboration." [source](https://unixshells.com/) |
| SSO / SAML | Not supported | "No SSO/SAML." [source](https://unixshells.com/) |
| Teams | Not supported | "Single-user shells; no team/org management." [source](https://unixshells.com/) |
| Workload identity | Not assessed |  |

### Developer access

| Capability | Status | Evidence |
| --- | --- | --- |
| CLI | Partially supported | "latch (MIT) is an official CLI multiplexer with built-in SSH/web/mosh; no shell-management CLI/API." [source](https://github.com/unixshells/latch) |
| E2B-compatible API | Not assessed |  |
| Filesystem API | Not applicable | "SSH/SFTP + FUSE in-VM; no remote filesystem API." [source](https://unixshells.com/) |
| Git deploy | Not supported | "No push-to-deploy (it's a raw VM you SSH into)." [source](https://unixshells.com/) |
| HTTP data API | Not applicable | "Not a data-layer product." [source](https://unixshells.com/) |
| OpenAPI spec | Not supported | "No public API → no OpenAPI spec." [source](https://unixshells.com/) |
| Private container registry | Not supported | "No hosted container registry (Docker runs in-VM)." [source](https://unixshells.com/) |
| Public API | Not supported | "No API for programmatic shell creation/management (SSH/web/mobile only)." [source](https://unixshells.com/) |
| SDK | Not supported | "No platform SDK (OSS repos are standalone tools/clients)." [source](https://unixshells.com/) |
| SSH access | Supported | "'Standard SSH, any client'; Ed25519 keys — SSH is the core access path." [source](https://unixshells.com/) |

### Compute & storage

| Capability | Status | Evidence |
| --- | --- | --- |
| Cloud-init / ignition config | Not applicable | "Managed snapshot-boot VM; no user cloud-init/user-data path." [source](https://unixshells.com/) |
| Desktop / computer-use sandbox | Not applicable | "Terminal/SSH product; no GUI desktop sandbox." [source](https://unixshells.com/) |
| GPU compute | Not supported | "No GPU; fixed 4GB/2vCPU shape." [source](https://unixshells.com/) |
| Live-state (memory) snapshots | Not assessed |  |
| Multi-region | Partially supported | "Relay = 9 servers / 3 continents; managed-shell region placement not disclosed (no user region choice surfaced)." [source](https://unixshells.com/) |
| Object storage | Not assessed |  |
| Persistent disk | Supported | "'your home directory persists'; Shell+ 25 GB disk (overlay resets, home persists)." [source](https://unixshells.com/) |
| Sandbox templates / custom images | Not applicable | "200+ tools are a baked memory snapshot, not user-defined fork-from-template images." [source](https://unixshells.com/) |
| Scale to zero | Not supported | "Always-on flat-rate VMs ('sessions that never die') — no scale-to-zero." [source](https://unixshells.com/) |
| Selectable isolation runtime | Not assessed |  |
| Snapshots / forking | Not supported | "Memory snapshot is a boot mechanism (pre-installed tools), not a user snapshot/fork primitive." [source](https://unixshells.com/) |

### Data

| Capability | Status | Evidence |
| --- | --- | --- |
| Database branching | Not applicable | "Not a data-layer product." [source](https://unixshells.com/) |
| Ephemeral instant DB (no-signup) | Not applicable | "Not a data-layer product." [source](https://unixshells.com/) |
| First-party database | Not supported | "No managed DB (run your own in the VM)." [source](https://unixshells.com/) |
| Vector search | Not applicable | "Not a data-layer product." [source](https://unixshells.com/) |

### Platform & operations

| Capability | Status | Evidence |
| --- | --- | --- |
| Managed CI/CD runners | Not assessed |  |
| Built-in observability | Not supported | "No metrics/observability dashboard." [source](https://unixshells.com/) |
| Preview environments | Not applicable | "No per-PR ephemeral environments (persistent single VM)." [source](https://unixshells.com/) |
| Scheduled tasks | Not applicable | "Full systemd + cron in the VM; no platform scheduled-task primitive needed." [source](https://unixshells.com/) |
| Project template marketplace | Not assessed |  |
| Durable workflows | Not applicable | "Not a workflow engine." [source](https://unixshells.com/) |

## Agent readiness

### Read: 4 of 11 signals present

4/11 signals (2 full, 2 partial)

- Llms txt: fail — "GET /llms.txt 404"
- Llms full txt: fail — "No /llms-full.txt"
- Robots AI policy: fail — "robots.txt BLOCKS major AI UAs (ClaudeBot, CCBot, Bytespider, Amazonbot, ...) Disallow: / + Content-Signal: ai-train=no (Cloudflare's AI-crawler-blocking default)"
- JSON LD: fail — "No application/ld+json (static HTML site)"
- Sitemap: fail — "GET /sitemap.xml 404"
- Meta quality: partial — "Descriptive <title> + meta description present; OG tags not confirmed"
- Markdown negotiation: fail — "Static HTML; no .md / markdown content negotiation"
- Content efficiency: pass — "Static HTML 24.8 KB -> 6.8 KB readable (~27.5%) -> >5%"
- Docs discoverable: fail — "No developer docs section (only per-agent marketing pages)"
- Pricing machine readable: partial — "Pricing on homepage as cards (Relay free, Shell+ $12/mo with specs) — readable but simple marketing prose, not a structured table"
- Anti bot: pass — "Behind Cloudflare but no challenge on public content (homepage 200); only robots blocks training crawlers"

### Drive: 1 of 11 signals present

1/11 signals (0 full, 1 partial)

- Public API: fail — "No public API for programmatic shell creation/management"
- Openapi spec: fail — "No API → no spec"
- MCP server: fail — "No MCP server"
- SDK packages: fail — "No platform SDK (OSS repos are standalone tools)"
- Auth docs: fail — "OPAQUE web login only; no API-auth docs"
- Agent self serve creds: fail — "Signup -> a VM; no programmatic key issuance"
- Webhooks: fail — "No webhooks"
- CLI tool: partial — "latch (MIT) is an official CLI multiplexer; not a shell-management CLI"
- Error model: fail — "No API → no error model"
- Rate limit docs: fail — "No API rate-limit docs"
- Agent config: fail — "No agents.md / well-known / skills surface"

## Coding harnesses

- Claude Code (Anthropic) — "'Run Claude Code on a dedicated VM' — pre-installed, dedicated landing page." [source](https://unixshells.com/claude-code.html)
- Codex CLI (OpenAI) — "'Give OpenAI Codex ... their own sandbox' — pre-installed, dedicated page." [source](https://unixshells.com/codex.html)
- OpenCode (opencode) — "OpenCode pre-installed, dedicated landing page." [source](https://unixshells.com/opencode.html)
- OpenClaw (Peter Steinberger) — "'Run multiple agents in parallel' — OpenClaw pre-installed, dedicated page." [source](https://unixshells.com/openclaw.html)
- Aider (Aider) — "Aider named among the ~6 pre-installed agents ('Claude Code, OpenClaw, Docker, every major language runtime ... Aider')." [source](https://unixshells.com/)

Plausible compatibility (derived, not stored):

- Amp
- Antigravity
- Devin
- Droid
- Gemini CLI
- Goose
- Hermes Agent
- Kilo Code
- Kiro CLI
- Pi
- Qwen Code

## Compliance & regions

- Regions note: Managed-shell region placement not disclosed (no user region choice surfaced). The free Relay network spans 9 servers across 3 continents (NAT traversal), but that is not workload placement.

## Company & presence

- Team size: small / likely solo (bootstrapped)
- GitHub: https://github.com/unixshells
- Mobile app (iOS + Android): https://unixshells.com/

## Nearby in the atlas

- [Catnip](/products/catnip.md)
- [Docker Offload](/products/docker-offload.md)
- [exe.dev](/products/exe.dev.md)
- [machine](/products/machine.md)
- [machine0](/products/machine0.md)
- [Multipass](/products/multipass.md)

## Method

Facts come from the product's own public pages. Each claim carries its evidence: a quote and the page it came from. We count capabilities rather than grade products, and "Not assessed" means we have not checked, never that the product lacks it.