Code & AI agent sandboxes: read the guide

NVIDIA OpenShell

Open source toolSandbox

The safe, private runtime for fleets of autonomous AI agents.

Homepage Docs Repository

Scanned 26 Oct 2026 · Schema v5 · features v21

Licence: Apache-2.0
Self-hosted; you pay for the infrastructure.
Free tier: Free to run
Isolation: Container
shares the host kernel
Regions: —
Not disclosed on the public surface.
Capabilities: 21 / 61
3 partial · 28 fail · 0 not assessed
Agent readiness: 10/10 · 8/9
read · drive, pre-login surface

Advertised for

Autonomous AI agents · Coding agents · Policy confined shell/command execution · Custom OCI workloads · GPU backed agent workloads

Built for

Developers, then technical

· https://raw.githubusercontent.com/NVIDIA/OpenShell/main/README.md — CLI-first sandbox creation, gRPC and SDKs.
· https://docs.nvidia.com/openshell/latest/about/architecture.md — Landlock/seccomp, supervisor, runtime drivers, declarative policy.

01 Website design Desktop public website · assessed 26 Oct 2026

Complete assessment.

The public entry is a documentation-led site: a white three-column layout, dark sans-serif headings, NVIDIA-green accents and restrained rules. Overview prose and a pale-blue release callout lead the landing; installation uses the same navigation shell with bordered command panels. This assessment covers anonymous desktop landing and documentation views only; pricing is not applicable to the free self-host-only SKU.

Corporate
Editorial minimalism

Landing page: captured
Pricing: not applicable
Docs: captured

02 Runtime Each fact carries its source; dots show confidence. Open a row for the evidence.

Isolation & lifecycle

Isolation primitive

Containershares the host kernel

Persistence

Driver-supported persistent storage: Docker/Podman named volumes, Kubernetes workspace PVCs, VM disk overlays. Stop/start preserves driver-backed files, not process memory.

Snapshots / forks

Resources & limits

Storage

Operator-managed named volumes/PVCs/VM overlays and explicit driver config; gateway metadata uses SQLite locally or PostgreSQL for HA.

Limits & quotas

Operator-configured CPU/memory, GPU counts and gRPC limits; MicroVM ignores --cpu/--memory (use vcpus/mem_mib), VM rootfs archive default 10 GiB, exec input limit 4 MiB; no vendor fleet quota.

GPU

NVIDIA CDI for Docker/Podman, Kubernetes nvidia.com/gpu scheduling, one GPU for MicroVM; actual hardware supplied by operator.

Access

Deploy interface

clisdkapidockerfilecontainer-imagetemplate

SDK languages

PythonTypeScriptGoRust

Network & secrets

Egress controls

Default-deny outer network fence; all TCP/DNS traverses a trusted supervisor with executable-aware endpoint/L7 policy checks and endpoint-bound credential injection.

Secrets

Provider credentials remain outside the workload; inject only for approved endpoints. Supports OAuth refresh, AWS SigV4, Google auth and SPIFFE JWT-SVID dynamic token grants. Plain --env secrets are readable by the sandbox.

Bring your own cloud

Placement & use

Jurisdiction

Advertised workloads

Autonomous AI agentsCoding agentsPolicy confined shell/command executionCustom OCI workloadsGPU backed agent workloads

03 Capabilities 21 pass · 3 partial · 28 fail · 61 of 61 assessed

Supported — Documented on the public surface and verified.
Partially supported — Exists with caveats: gated, thin, or enterprise-only.
Not supported — Checked and not offered.
Not assessed — Scanned before this capability was tracked. Not a lack.

Compute & storage 4/10

Persistent diskStorage that survives restart/stop (not ephemeral-only).

Supported

Object storageManaged S3-compatible / blob object store the platform PROVIDES to a user's workloads (buckets, presigned URLs). DISTINCT from persistent-disk (block/volume attached to an instance) and from a self-host dependency (source.self_host.deps[]=["object-storage"] is an infra dep to RUN an OSS tool — the opposite direction). SCOPE varies and matters: native first-class (Railway Buckets, Cloudflare R2) vs a sibling SKU / extension (Fly Tigris, Bunny Storage, DigitalOcean Spaces) — record which in the evidence quote / note. Admitted v15.

Not supported

Snapshots / forkingSnapshot or copy-on-write fork of a running instance (filesystem-only counts). Whether the snapshot also restores memory + running processes is recorded separately by memory-snapshot (v19).

Not supported

Live-state (memory) snapshotsA snapshot / checkpoint / pause captures a running instance's MEMORY and PROCESS state and restores it, so running processes continue where they left off instead of a freshly booted instance with the same files. Counts whether user-triggered (snapshot, checkpoint, pause-to-storage) or used to restore pre-initialised state (memory snapshots for fast cold starts). Distinct from snapshots-forks (any snapshot or fork, including filesystem-only; a product passing that key may still fail this one) and from scale-to-zero (an idle-cost property; suspension may or may not keep memory). Actor multiplexing (see scale-to-zero) records its live-state half here — only when the suspended actor's MEMORY/process state is what comes back, not a freshly booted worker with the actor's files.

Not supported

Scale to zeroIdle instances suspend at no / negligible cost. ACTOR MULTIPLEXING (many suspended stateful actors sharing a small pool of pre-warmed workers, woken on request — Agent Substrate, AX, kagent; Rivet's hibernating actors) has NO key of its own (proposal REJECTED v20) — score the idle-cost half HERE and the live-state half under memory-snapshot; the oversubscription ratio / density is a performance characteristic the registry does not key.

Not supported

GPU computeGPU-backed instances available.

Supported

Multi-regionCompute placeable in more than one geographic region.

Not applicable

Sandbox templates / custom imagesBuild reusable sandbox/VM images from an OCI image or Dockerfile (a base you fork instances from).

Supported

Cloud-init / ignition configCustomise a VM / microVM / instance at boot via cloud-init (or ignition) user-data — the same boot-metadata interface as AWS/Azure/GCP. Inject packages, users, SSH keys, files, and run-commands declaratively at launch. Distinct from sandbox-templates (a PREBUILT image you fork instances from) — this is PER-LAUNCH declarative customisation, not a baked image.

Not supported

Desktop / computer-use sandboxA virtual desktop (GUI) for LLM computer-use — browser + desktop automation, often via noVNC/VNC.

Not supported

Selectable isolation runtimeThe user chooses the isolation runtime per workload / environment / sandbox from more than one option (e.g. microVM vs gVisor vs plain container, native namespaces vs gVisor, gVisor vs VM). Records the CHOICE; the isolation_primitive facet still records the best-evidenced primary. A platform that switches runtime automatically without a user choice (e.g. gVisor forced for GPU) does not pass on that alone.

Supported

Networking & delivery 2/8

Automatic HTTPSTLS endpoints provisioned automatically per workload.

Not supported

Built-in reverse proxyPlatform routes external traffic to workload services.

Supported

Custom domainsBring-your-own domain via CNAME/DNS.

Not supported

Public preview URLsShareable per-deploy / per-branch URLs.

Partially supported

Direct public IPv6Per-instance stable public IPv6 for raw TCP/UDP (beyond a managed HTTPS proxy) — e.g. mosh / raw sockets.

Not supported

Egress controlsConfigurable OUTBOUND network policy — egress firewall / allowlist / egress proxy on traffic LEAVING a workload. Scoped to OUTBOUND only (v15): private-network REACHABILITY between a user's own services is private-networking, not this. Static-egress-IP belongs here (an outbound concept).

Supported

Private networkingA platform-internal private network between a user's OWN services — service-to-service connectivity, VPC, service mesh, and/or service discovery, not exposed to the public internet. DISTINCT from egress-controls (OUTBOUND firewall/allowlist), reverse-proxy (PUBLIC inbound ingress), and direct-ipv6 (per-instance PUBLIC addressing). Admitted v15.

Not supported

Email send / receiveSend and/or receive email from the workload.

Not supported

Edge KV storeA globally-replicated key-value store read from WITHIN an edge runtime at the PoP (low-latency, eventually-consistent) — Cloudflare Workers KV, Deno KV, Fastly KV Store, Akamai EdgeKV, Vercel Edge Config. Distinct from first-party-db (a managed DATABASE service you provision and connect to) and from http-query-api (HTTP access to an external store) — edge-kv is colocated with the edge compute and read at the edge.

Not applicable

Developer access 4/9

Public APIDocumented programmatic API.

Supported

OpenAPI specMachine-readable OpenAPI / Swagger spec published.

Not supported

CLIOfficial command-line tool.

Supported

SDKOfficial client library/libraries.

Supported

SSH accessDirect SSH into workloads.

Supported

Filesystem APIRemote file operations over the API (list/read/write/watch/copy/upload) without SSH.

Not supported

Git deployDeploy from a Git repo / push-to-deploy.

Not supported

Private container registryHosted private image registry.

Not supported

E2B-compatible APIDrop-in compatibility with the E2B sandbox API/SDKs — existing E2B client code runs against this product by changing the endpoint/domain + key, with no business-code changes. E2B itself passes as the reference implementation. Distinct from public-api / sdk (the product has its own API — this key says it also speaks another vendor's) and from self-host-oss (a self-hosted E2B-compatible server is scored on both).

Not supported

HTTP data APIQuery / read / write the stored DATA over plain HTTP/REST (or GraphQL) WITHOUT the native protocol driver (Postgres wire / Redis RESP) — agent- and edge-friendly data access via fetch(). The DATA plane; distinct from public-api (a control-plane / management API). Auto-generated-from-schema REST/GraphQL (PostgREST, pg_graphql) is a flavour of this.

Not applicable

AI & agents 4/12

Built-in coding agentShips a first-party AI coding agent in the product.

Not supported

Agent skillsInstallable / declarable agent skills or tools.

Supported

AI app builderPrompt-to-app / vibe-coding generation surface.

Not supported

MCP serverExposes a Model Context Protocol server for agents.

Partially supported

MCP clientActs as an MCP client / consumes MCP tools.

Not supported

MCP-server hostingRun/host MCP servers as a workload on the platform (distinct from exposing a first-party MCP control server — that is mcp-server).

Not supported

LLM gatewayRoutes LLM API calls through the platform (multi-provider).

Supported

BYOK for LLMsBring-your-own LLM provider keys.

Supported

Model inference API (first-party)A FIRST-PARTY hosted LLM/model inference API — the provider serves its OWN catalogue of (usually open-weight) models behind a (typically OpenAI-compatible) endpoint, billed per token. Distinct from llm-gateway (a multi-PROVIDER proxy/router for YOUR calls to external providers like OpenAI/Anthropic) and from byok-llm (bring your own provider keys) — here the platform IS the inference provider, serving its own models.

Not supported

Autonomous QA agentA first-party AUTONOMOUS AI agent that QA-TESTS a running app / implementation — exercising it like a real user (clicking through flows, checking behaviour) and reporting bugs/defects (often back to a coding agent, or before a human review). Distinct from built-in-agent (a code-AUTHORING agent — this one VERIFIES the result, it does not write the feature; a product can ship both), from ai-guided-code-review (organises a diff for a HUMAN to review — not an autonomous tester), and from observability (runtime metrics/logs, not active testing).

Not supported

Credential-injection proxyThe platform holds real secrets OUTSIDE the workload and injects them on the wire (an egress / broker proxy swaps a placeholder or phantom token for the real credential on matching outbound requests), so code running in the sandbox — e.g. an untrusted agent — never sees or can exfiltrate the real key. Distinct from secrets-management (storage + injection INTO the workload as env vars/files — the workload holds the secret) and from egress-controls (outbound allow/deny POLICY; a product can have an allowlist without injecting credentials, and vice versa).

Supported

Preinstalled coding agentThe platform installs and launches a THIRD-PARTY coding agent (a harnesses.yml entry — Claude Code, Codex, OpenCode, Pi ...) inside a workspace/sandbox it provisions, with the agent's auth wired in (your own account or API key), so you start an agent session rather than an empty box. Distinct from built-in-agent (a FIRST-PARTY agent the vendor ships) and from harnesses[] (which records that a harness CAN target the product, not that the product runs it for you).

Not supported

Data 0/1

First-party databasePlatform-provided managed database.

Not supported

Database branchingCopy-on-write branch of a DATABASE (schema + data) as a dev-lifecycle primitive — branch your data like code, usually instant via a metadata operation. Distinct from snapshots-forks (forks a running compute INSTANCE / VM state) and from preview-environments (a full per-PR environment that may USE a DB branch). For a data product a CoW branch maps HERE, not to snapshots-forks.

Not applicable

Vector searchFirst-class storage, indexing, and similarity search of vector embeddings — a native vector type/index or a managed vector DB. The data-layer primitive for RAG / semantic retrieval.

Not applicable

Ephemeral instant DB (no-signup)Provision a working, usable instance via a single command / HTTP call with NO account, signup, or card — typically time-boxed (expires unless claimed). A zero-onboarding path an agent can hit directly.

Not applicable

Platform & operations 1/5

Scheduled tasksCron / scheduled job primitive.

Not supported

Durable workflowsDurable execution / long-running workflow primitive.

Not supported

Preview environmentsA full EPHEMERAL environment spun up per PR / branch — compute + addons + (often) a DB branch — torn down on merge. More than preview-urls (which is just a shareable URL for one deploy).

Not supported

Built-in observabilityFirst-party metrics / logs / tracing / APM / profiling included in the platform (no third-party stack required to see what a workload is doing).

Supported

Managed CI/CD runnersPlatform-hosted, drop-in CI/CD runner VMs (GitHub Actions / GitLab), typically faster and cheaper than the CI vendor's own hosted runners, billed per runner-minute; isolated + ephemeral per job. Distinct from git-deploy (push-to-deploy of an APP) and scheduled-tasks (cron) — this is managed CI COMPUTE you point your existing pipelines at.

Not applicable

Project template marketplaceA marketplace / gallery of prebuilt PROJECT STARTER templates (apps or websites) a user clones or launches a new project from and then customises — the "start from a template" surface of a builder / PaaS. Distinct from sandbox-templates (a COMPUTE base IMAGE you fork runtime INSTANCES from — Dockerfile / OCI), from app-builder (the prompt-to-app GENERATION surface — a product can have both), and from one-click OSS-app catalogues / helm-addon-marketplace (deploy a FINISHED app as-is, not a starter you build on).

Not supported

Organisation & security 6/7

Built-in authPlatform-provided authentication / IAM for workloads.

Supported

SSO / SAMLEnterprise single sign-on.

Supported

Secrets managementFirst-class secrets storage / injection.

Supported

TeamsMulti-user team accounts.

Supported

SharingShare a running workload/resource via link (optionally gated).

Partially supported

BYOCRuns inside the customer's own cloud account.

Not applicable

Deploy into customers' VPCsA SaaS vendor uses the platform to deploy and operate ITS software inside ITS CUSTOMERS' cloud accounts / VPCs (nested or "customer BYOC"), managed from the vendor's control plane. Distinct from byoc (your own workloads in your own cloud account).

Not applicable

Workload identityThe platform gives each workload its own verifiable identity (OIDC token, SPIFFE SVID, cloud IAM federation) that it exchanges for short-lived credentials, so no long-lived static secret is stored or injected. Distinct from secrets-management (stores / injects static secrets) and credential-injection-proxy (swaps real secrets in on the wire).

Supported

Self-host / OSS coreOpen-source core that can be self-hosted.

Supported

Proposed, not verified

Capabilities the scan saw but the registry does not track yet. A key is admitted only once two distinct products exhibit it. These are not scored.

Formally verified policy changesformal-policy-verification
“The policy prover is OpenShell's policy verification engine. It uses an SMT solver to check whether policies satisfy specified properties, such as staying within an allowed access boundary. Its guarantees apply to the policy features and behavior represented by its model.” docs.nvidia.com
Supervisor Middlewaresupervisor-traffic-middleware
“Inspect, block, or change sandbox requests and responses with supervisor middleware.” docs.nvidia.com

04 Agent readiness Pre-login surface · counts, not a grade

Can an agent read it?

parse-ability · docs, metadata, machine-readable surfaces

10/10

7 full · 3 partial · 100% of applicable checks

llms.txt
Supported
“https://docs.nvidia.com/openshell/latest/llms.txt — substantive index with installation, lifecycle, policies, SDKs and architecture links.”
llms-full.txt
Partially supported
“https://docs.nvidia.com/openshell/latest/llms.txt — requested llms-full.txt redirects to llms.txt index, not complete page bodies.”
robots.txt AI policy
Supported
“https://docs.nvidia.com/robots.txt — explicitly names GPTBot/Claude/Perplexity/Google-Extended and allows Markdown and llms.txt; no blanket AI Disallow.”
JSON-LD
Partially supported
“https://docs.nvidia.com/openshell/about/overview — BreadcrumbList and TechArticle, not a SoftwareApplication/Product with structured offers.”
Sitemap
Supported
“https://docs.nvidia.com/openshell/sitemap.xml — nontrivial product sitemap discovered via robots.txt -> fern-sitemap-index.xml.”
Meta tags
Supported
“https://docs.nvidia.com/openshell/about/overview — descriptive title, description, OG title/description/image/URL.”
Markdown negotiation
Supported
“https://docs.nvidia.com/openshell/about/overview — Accept: text/markdown returns text/markdown; linked .md pages are readable.”
Content efficiency
Partially supported
“https://docs.nvidia.com/openshell/about/overview — visible text excluding script/style/noscript: 4.74% of HTML bytes (1–5% bracket); artifacts/site-signals.json records the current byte counts.”
Docs discoverable
Supported
“https://raw.githubusercontent.com/NVIDIA/OpenShell/main/README.md — first-class Documentation badge and deep guide links; landing itself is the current guide.”
Machine-readable pricing
Not applicable
“No first-party maintained OpenShell hosted tier/pricing. Apache-2.0 self-host software; own infrastructure costs are not tiers.”
No anti-bot wall
Supported
“https://docs.nvidia.com/openshell/about/overview and linked .md corpus fetched without a public-content challenge; Brev JS launchable rendered successfully after consent.”

Can an agent drive it?

act-ability · APIs, SDKs, MCP, credentials, webhooks

8/9

6 full · 2 partial · 89% of applicable checks

Public API
Supported
“https://docs.nvidia.com/openshell/latest/how-it-works/gateways/authentication.md — public openshell.v1.OpenShell gRPC API advertised by reflection; four SDKs are documented in the README. Endpoint is self-hosted.”
OpenAPI spec
Not applicable
“Current API is gRPC/protobuf, not a documented REST/Swagger service; common OpenAPI/Swagger paths 404 and no spec found in source index.”
MCP server
Supported
“https://docs.nvidia.com/openshell/latest/llms.txt — documented MCP endpoint https://docs.nvidia.com/openshell/_mcp/server; documentation-only, not sandbox operations.”
SDK packages
Supported
“https://raw.githubusercontent.com/NVIDIA/OpenShell/main/README.md — Python, TypeScript, Go and Rust SDK install references.”
Auth documented
Supported
“https://docs.nvidia.com/openshell/latest/how-it-works/gateways/authentication.md — mTLS, OIDC, client credentials, roles/scopes and credential layouts documented.”
Self-serve credentials
Not applicable
“No hosted signup/key issuance service; operator owns self-hosted gateway credentials and IdP.”
Webhooks
Not supported
“https://docs.nvidia.com/openshell/latest/llms.txt — no documented signed webhook integration found in the current corpus; gateway gRPC interceptors are not webhooks.”
CLI tool
Supported
“https://docs.nvidia.com/openshell/latest/about/installation.md — official openshell CLI installer/packages documented.”
Error model
Partially supported
“https://docs.nvidia.com/openshell/latest/sdk/api-errors.md — detailed gRPC status/protobuf errors (BadRequest, ErrorInfo, RetryInfo), not the rubric's full-credit JSON error model.”
Rate limits documented
Partially supported
“https://docs.nvidia.com/openshell/latest/how-it-works/gateways/configuration.md — grpc_rate_limit_requests/window_seconds settings; no hosted HTTP rate-limit response-header contract.”
Agent config files
Supported
“https://raw.githubusercontent.com/NVIDIA/OpenShell/main/AGENTS.md and https://raw.githubusercontent.com/NVIDIA/OpenShell/main/README.md — AGENTS.md plus installable NVIDIA/OpenShell agent skills.”

05 Coding harnesses Which agents and IDEs are documented to run here

Advertised integrations 4

OpenCode opencode
Supported
“OpenCode detects the OpenRouter credential and starts with the model passed to `-m`.”
Source docs.nvidia.com
Claude Code Anthropic
Supported
“openshell sandbox create --from registry.example.com/your-org/claude-agent:latest -- claude”
Source docs.nvidia.com
Codex CLI OpenAI
Supported
“This example builds a sandbox image with Codex, runs its app server inside an OpenShell sandbox, exposes the WebSocket service through the local gateway, and connects a Codex client running on the host.”
Source raw.githubusercontent.com
Pi earendil-works
Supported
“Build a Pi coding-agent image, configure OpenRouter access, and run Pi inside an OpenShell sandbox.”
Source docs.nvidia.com

Plausible compatibility 11

Derived, not stored: the product passes SSH access, so any CLI coding agent is installable inside it. This is plausible compatibility, not a supported integration.

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

06 Compliance & regions Standards held, with ordinal rigor; never a score

Certifications held 0

No attestations found on the public surface. Empty means nothing is advertised, after checking, not that nothing was checked.

Regions —

No region list captured.

Legal

Privacy notice Terms of service

07 Open source Apache-2.0 · fetched by machine, not quoted from marketing

Licence & repository

Licence: Apache-2.0OSI
Primary language: Rust
Stars: 19,954
Forks: 1,972
Created: 24 Feb 2026
Archived: No

Activity

Last commit: 26 Oct 2026
Latest release: v0.1.5 · 20 Oct 2026
Issues open / closed: 377 / 1345 · 78% closed

Latest stable release v0.1.5 published 2026-10-20; v0.1.4 published 2026-10-13 and v0.1.3 on 2026-10-09. Rolling dev tag observed separately (createdAt 2026-10-21, publishedAt 2026-03-18); do not confuse a mutable development tag with the newest stable release.

Fetched via gh on 26 Oct 2026.

Self-hosting

Methods: dockercomposekuberneteshelmbinarypackage-manager
You operate: dockerpodmankubernetesagent-sandboxpostgres

Free Apache-2.0 software plus the operator's own infrastructure and model-provider costs; no maintained first-party hosted tier established. Dependencies are alternatives/conditional: Docker or Podman local containers, or opt-in MicroVM with KVM/Apple Hypervisor; Kubernetes requires Agent Sandbox controller and a NetworkPolicy-enforcing CNI. PostgreSQL is for shared/HA gateways, not required by every local install. Operator maintains workload images, providers, TLS and storage; default Ubuntu has no agent installed. Windows WSL2 is experimental. Retired community-image launchable must not be used for new deployments.

Documentation signals 7/8

SupportedQuickstart
SupportedAPI reference
SupportedSelf-host guide
SupportedChangelog
Not supportedOpenAPI
Supportedllms.txt
SupportedExamples
SupportedArchitecture doc

08 Company & presence Facts only, not scored

Company

Headquarters: Not disclosed
Founded: —
Team size: Not disclosed
Hiring: NVIDIA corporate careers page accepts applications; OpenShell-specific hiring not confirmed.

Activity

Last release: v0.1.5 (2026-10-20T17:28:32Z) github.com
Last blog post: —

Presence 5

Links found on the public surface scanned; not an exhaustive search.

09 Sources & method How this entry was produced

Public surface only

Everything here was read from docs.nvidia.com's unauthenticated pages, docs, pricing and repository on 26 Oct 2026. Nothing behind a login was used.

Statuses, not scores

Capabilities are pass, partial or fail against a written rubric (features v21). “Not assessed” means the scan predates the key.

Confidence is separate

Each fact carries its own confidence. Low-confidence facts are shown, never hidden, so you can weigh them.

Schema: scan v5
Features registry: v21
Keys tracked today: 61
Assessed on this entry: 61

Something wrong on this page?

Nearby in the atlas

All Sandbox products →