{
 "features_resolved": {
  "snapshot_any": true,
  "snapshot_mem": true,
  "fork": true,
  "pause_resume": true,
  "persistent_disk": true,
  "whole_disk": null,
  "persist_scope": null,
  "volumes": true,
  "auto_stop_idle": true,
  "long_sessions": true,
  "custom_image": true,
  "image_snapshot": true,
  "image_oci": true,
  "vm_isolation": true,
  "docker_inside": true,
  "nested_virt": false,
  "root": null,
  "systemd": null,
  "browser": true,
  "desktop": true,
  "computer_use": true,
  "code_interpreter": false,
  "browser_control": null,
  "desktop_control": null,
  "browser_and_desktop": true,
  "anti_bot": false,
  "captcha_solving": false,
  "residential_ip": false,
  "agent_harnesses": false,
  "ssh": false,
  "public_ipv4": false,
  "inbound_https": true,
  "custom_domain": null,
  "raw_tcp_inbound": false,
  "egress_allowlist": true,
  "open_internet": true,
  "static_egress_ip": false,
  "private_network": null,
  "self_host": false,
  "byoc": false,
  "open_source": false,
  "soc2": false,
  "hipaa": false,
  "sso": false,
  "gpu": false,
  "arm64": false,
  "windows": false,
  "macos": false,
  "lang_python": true,
  "lang_node": true,
  "lang_go": null,
  "lang_rust": null,
  "lang_java": null,
  "gpu_desktop": false,
  "credential_injection": true,
  "egress_http_rules": true,
  "wake_on_request": false,
  "live_resize": false,
  "fork_running": true,
  "memory_snapshot_fork": true,
  "webhooks": null,
  "scheduled_wakeups": null,
  "audit_logs": true,
  "eu_data_residency": true,
  "zero_data_retention": null,
  "scoped_api_keys": null,
  "spend_limits": null,
  "mcp_server": true,
  "usage_api": true,
  "snapshot_auto": false,
  "snapshot_on_demand": true,
  "harness_api": false,
  "own_agent_api": false,
  "fast_boot": true,
  "strong_isolation": true,
  "ingress_rules": true,
  "guest_firewall": true,
  "secret_proxy": true,
  "secret_proxy_any": false,
  "volume_attach": true,
  "volume_shared": false,
  "sidecar": null,
  "fuse": true,
  "kernel_features": true,
  "burst": null
 },
 "company": {
  "id": "isorun",
  "checked": "2026-09-30",
  "company": "Isorun",
  "ownership": {
   "value": "independent",
   "parent": null,
   "source": "https://isorun.ai/",
   "quote": "Isorun: sandboxes in milliseconds. Public beta, $50 free credit.",
   "note": "No ownership/about/team page found; treated as a small independent product in public beta absent any evidence of a parent company or acquisition."
  },
  "stage": {
   "value": "unknown",
   "source": "",
   "quote": "",
   "confidence": "low",
   "note": "No funding announcement, press coverage, Crunchbase/LinkedIn/team page, or accelerator mention was found for Isorun in searches (DuckDuckGo queries for \"isorun funding\", \"isorun seed round\", \"isorun Y Combinator\", \"isorun crunchbase\" all returned zero results). The site is a public-beta product page with no company-background content."
  },
  "accelerators": [],
  "total_funding_usd": {
   "value": null,
   "status": "unknown"
  },
  "last_round": {
   "type": null,
   "amount_usd": null,
   "date": null,
   "leads": [],
   "status": "unknown",
   "source": "",
   "quote": ""
  },
  "investors": [],
  "portfolio_listings": [],
  "revenue": {
   "value": null,
   "status": "unknown"
  },
  "employees": {
   "value": null,
   "source": ""
  },
  "note": "Isorun (isorun.ai) is an AI-agent microVM sandbox runtime in public beta (v1.0.0 launched 2026-06-16). Extensive searches (DuckDuckGo web search for funding/seed/Series A/ARR/Y Combinator/Crunchbase, and direct checks of the homepage, docs, and GitHub org) turned up no company-background information at all: no about/team page, no press coverage, no investor names, no accelerator affiliation, no employee count. This appears to be either a very new/stealthy product or one with minimal public footprint beyond its docs/pricing site. All funding/ownership/revenue fields are genuinely unknown rather than guessed."
 },
 "id": "isorun",
 "name": "Isorun",
 "url": "https://docs.isorun.ai/getting-started/pricing",
 "category": "agent-sandbox",
 "isolation": "vm",
 "modes": [
  {
   "key": "on-demand",
   "label": "MicroVM sandbox, per-second (public beta)",
   "flags": [],
   "os": [
    "linux"
   ],
   "regions": [
    "us",
    "eu"
   ],
   "gpu_only": false,
   "features": {
    "isolation": "vm"
   },
   "pricing": "resource",
   "vcpu_h": 0.025,
   "ram_gib_h": 0.015,
   "cpu_basis": "alloc",
   "ram_basis": "alloc",
   "active_floor": 0,
   "ram_per_vcpu": [
    null,
    null
   ],
   "ram_fixed_per_vcpu": null,
   "vcpu_options": null,
   "min_vcpu": 1,
   "max_vcpu": 8,
   "max_ram_gib": null,
   "sizes": null,
   "pool_tiers": null,
   "requires_always_on": false,
   "always_on_month_per_instance": null,
   "commit_note": null,
   "min_billed_seconds": 0,
   "granularity_s": 1,
   "start_fee": 0,
   "boot_overhead_s": null,
   "multiplier": 1,
   "windows_vcpu_h": null,
   "gpu": null,
   "note": "Billed per second on allocated vCPU + GiB RAM from create() to destroy() or idle auto-destroy (timeoutSec, default 300 s idle, so every session carries up to a 5-min billed idle tail unless destroyed/hibernated explicitly). \"No minimums\". Hibernated = $0 (pricing page). 4 vCPU/8 GiB = $0.22/h (docs table). Homepage selector offers 1/2/4/8 vCPU; no hard vCPU/RAM cap published (max_vcpu 8 taken from the homepage selector). 4 GiB growable scratch disk included; egress/TLS/platform fees included in compute price. Basis verified 2026-09-28: \"Every sandbox has an idle timeout (default 300 seconds, configurable per-create via timeoutSec)\" (https://docs.isorun.ai/sandboxes/lifecycle); billed \"every second from isorun.create() to sandbox.destroy() (or the auto-destroy timeoutSec timer)\" (https://docs.isorun.ai/getting-started/pricing)."
  }
 ],
 "storage": {
  "included_disk_gib": 4,
  "disk_gib_month": 0.02,
  "disk_billed_when_stopped": true,
  "snapshot_gib_month": 0,
  "paused_ram_billed": false
 },
 "network": {
  "egress_free_gib": null,
  "egress_gib": 0,
  "ipv4_month": null,
  "ipv4_available": false,
  "internet": {
   "default": "configurable",
   "evidence": [
    {
     "quote": "By default the policy is 'unrestricted', the VM can reach any host on the public internet.",
     "url": "https://docs.isorun.ai/security/network-filtering"
    },
    {
     "quote": "Anthropic API + PyPI + npm + GitHub + crates.io",
     "url": "https://docs.isorun.ai/security/network-profiles"
    }
   ]
  }
 },
 "plans": [
  {
   "name": "Pay as you go (public beta)",
   "fee": 0,
   "flags": [],
   "fee_is_credit": false,
   "included_usd": 0,
   "per_seat": 0,
   "concurrency": null,
   "max_session_h": null,
   "max_vcpu": null,
   "max_ram_gib": null,
   "sizes_allowed": null,
   "note": "No plan fee. $50 one-time signup credit, no card. Concurrency limits exist (\"concurrent sandboxes count against your limits\") but are unpublished."
  },
  {
   "name": "Talk to us (custom)",
   "fee": null,
   "flags": [
    "sales"
   ],
   "fee_is_credit": null,
   "included_usd": null,
   "per_seat": null,
   "concurrency": null,
   "max_session_h": null,
   "max_vcpu": null,
   "max_ram_gib": null,
   "sizes_allowed": null,
   "note": "Homepage \"Talk to us\" CTA; no published enterprise tier or rates."
  }
 ],
 "free": {
  "monthly_credit": 0,
  "one_time_credit": 50
 },
 "features": {
  "linux": true,
  "windows": false,
  "macos": false,
  "arm64": false,
  "gpu": false,
  "gpu_types": [],
  "snapshot": "mem",
  "pause_resume": true,
  "fork": true,
  "persistent_disk": true,
  "volumes": true,
  "auto_stop_idle": true,
  "max_session_h": null,
  "custom_image": true,
  "docker_inside": true,
  "nested_virt": false,
  "root": null,
  "systemd": null,
  "browser": true,
  "desktop": true,
  "computer_use": true,
  "code_interpreter": false,
  "agent_harnesses": false,
  "ssh": false,
  "public_ipv4": false,
  "inbound_https": true,
  "custom_domain": null,
  "raw_tcp_inbound": false,
  "egress_allowlist": true,
  "static_egress_ip": false,
  "private_network": null,
  "self_host": false,
  "byoc": false,
  "open_source": false,
  "regions": [
   "us",
   "eu"
  ],
  "soc2": false,
  "hipaa": false,
  "sso": false,
  "lang_python": true,
  "lang_node": true,
  "lang_go": null,
  "lang_rust": null,
  "lang_java": null,
  "gpu_desktop": false,
  "credential_injection": true,
  "egress_http_rules": true,
  "wake_on_request": false,
  "live_resize": false,
  "fork_running": true,
  "memory_snapshot_fork": true,
  "webhooks": null,
  "scheduled_wakeups": null,
  "audit_logs": true,
  "eu_data_residency": true,
  "zero_data_retention": null,
  "scoped_api_keys": null,
  "spend_limits": null,
  "mcp_server": true,
  "isolation": "vm",
  "image_oci": true,
  "snapshot_auto": false,
  "snapshot_on_demand": true,
  "harness_api": false,
  "egress_allowlist_level": "both",
  "ingress_rules": true,
  "guest_firewall": true,
  "secret_proxy": "ai-only",
  "volume_attach": true,
  "volume_shared": false,
  "anti_bot": false,
  "captcha_solving": false,
  "egress_ip_type": "datacenter",
  "own_agent_api": false,
  "max_session_unlimited": true,
  "fast_boot": true,
  "open_internet": true,
  "usage_api": true
 },
 "perf": {
  "cpu_runs_s": null,
  "cold_start_ms": 16.83,
  "burst_tti_ms": 43.58,
  "dax_s": 32.79
 },
 "perf_source": "computesdk: cold_start_tti median 16.83 ms (2026-08-07), burst_tti median 43.58 ms at 100 concurrent (2026-09-25), DAX total 32.79 s (2026-09-25, 8 vCPU / 15.7 GiB x86_64 env). No hpc CPU run.",
 "knobs": [],
 "caveats": [
  "Public beta pricing ($0.025/vCPU-h + $0.015/GiB-h, launched v1.0.0 2026-06-16); no change since.",
  "Allocation billing assumed: the pricing page bills \"by the second based on vCPU and RAM\" with an allocation-based example (1 vCPU+1 GiB x 90 s = $0.001), although destroy() says cpuMs/memPeakBytes are \"the numbers you are billed on\".",
  "Hibernation is $0 per the pricing page (\"you pay nothing while the sandbox is paused\"); storage price for explicit snapshot() checkpoints is unpublished (snapshot_gib_month 0 reflects hibernation only).",
  "Persistent disks (Tigris-backed) cost $0.02/GB-month on data actually stored, plus normal compute while attached; forks add only divergent writes. Encoded as disk_gib_month.",
  "Default idle auto-DESTROY after 300 s (not hibernate): state is lost unless you hibernate or raise timeoutSec; a raised timeout keeps billing idle time.",
  "Concurrency/session limits, docker-in-VM and hypervisor (KVM microVM, VMM unnamed) unpublished. Org pinned to one region (US or EU); no regional price difference published.",
  "SOC 2 Type II listed as \"soon\".",
  "No SSH: docs (llms-full.txt) list only WebSocket PTY terminal, WebSocket shell sessions, exec() and HTTPS port URLs as access methods (verified 2026-09-28).",
  "Docs conflict on idle-timeout default: lifecycle page and SDK reference say 300 s; the Claude Managed Agents integration example says 'idle auto-destroy, default 600'. Timer can be disabled with 0 (then idle time bills indefinitely).",
  "Port URLs: docs limits table says 'Auth: none; the sandbox ID is the credential' (homepage calls them 'signed'). Any TCP port can be exposed over HTTPS/WS; no raw TCP inbound.",
  "Browser/computer-use image docker.io/isorun/computer-use:1 needs >= 2 vCPU and >= 4 GiB (docs sizing guidance: \"2 minimum\" vCPU, \"4 GiB minimum\", https://docs.isorun.ai/sandboxes/computer-use-vnc, checked 2026-10-14) = $0.11/h minimum at list rates.",
  "Allocation basis is corroborated by the API reference: a 1 vCPU / 1 GiB sandbox with uptime_ms 59000 but cpu_ms 48 and mem_peak 49 MB returns cost_cents 0.0007, which equals allocation price in dollars (0.04 $/h x 59 s = $0.00066), not a usage-based figure. The 'cost_cents' / costCents field appears to be denominated in dollars despite its name (pricing page: 90 s example = $0.001, costCents 'e.g. 0.00111')."
 ],
 "sources": [
  "https://docs.isorun.ai/getting-started/pricing",
  "https://isorun.ai/",
  "https://docs.isorun.ai/llms-full.txt",
  "https://docs.isorun.ai/sandboxes/lifecycle",
  "https://docs.isorun.ai/sandboxes/persistent-storage",
  "https://docs.isorun.ai/sandboxes/checkpoints-and-rollback",
  "https://docs.isorun.ai/changelog",
  "https://www.computesdk.com/benchmarks/sandboxes/",
  "https://docs.isorun.ai/sandboxes/computer-use-vnc"
 ],
 "fleet": {
  "default_max_instances": null,
  "default_max_vcpu": null,
  "default_max_gpus": null,
  "raise": null,
  "raise_time_days": null,
  "verification": "Public beta, no card; introductory credit.",
  "tiers": null,
  "notes": "Documentation acknowledges concurrent sandboxes consume account quota but numeric limit and increase path not established. Homepage 30 sibling-forks demonstration is not an account entitlement.",
  "sources": [
   "https://docs.isorun.ai/getting-started/pricing",
   "https://isorun.ai/"
  ],
  "researched": "2026-09-28"
 },
 "audit_2026_09_29": {
  "applied": 3,
  "source": "checked against the provider's live pricing and docs pages",
  "reviewed": 1
 },
 "feature_evidence": {
  "snapshot_auto": {
   "value": false,
   "source": "https://docs.isorun.ai/sandboxes/lifecycle",
   "quote": "When the timer fires, the sandbox is destroyed.",
   "note": "Idle timeout (default 300 s, timeoutSec; 0 disables) DESTROYS the sandbox; state is only kept if you call hibernate() yourself (memory+processes+filesystem, $0 while hibernated). Conflict: the Claude Managed Agents integration page advertises 'Hibernate-on-idle' but then says the server-side TTL 'reaps' idle sandboxes; the lifecycle page and SDK reference say auto-destroy.",
   "checked": "2026-09-29"
  },
  "snapshot_on_demand": {
   "value": true,
   "source": "https://docs.isorun.ai/sandboxes/checkpoints-and-rollback",
   "quote": "A checkpoint captures a sandbox's full state so you can restore it later or fork it into parallel copies.",
   "note": "sandbox.snapshot() / POST /v1/runs/{id}/snapshot on a running VM; isorun.restore(id) boots a new sandbox from it (same region); snapshots persist until deleteSnapshot(). fork(n) clones a running sandbox: 'Each child inherits filesystem + memory state at fork time'. Memory included (changelog: 'checkpoint a running VM'). Separate disk-level snapshotDisk/forkDisk for persistent disks (copy-on-write).",
   "checked": "2026-09-29"
  },
  "harness_api": {
   "value": false,
   "source": "https://docs.isorun.ai/integrations/claude-managed-agents",
   "quote": "The Isorun SDK ships a self-hosted adapter that",
   "note": "No Isorun API to start/stream/follow-up an agent harness; the REST API is sandbox lifecycle, exec (JSON + SSE), PTY/shell WebSockets, files, snapshots, audit, network. Closest: runOrchestrator() adapter makes Isorun the self-hosted sandbox backend for Anthropic Claude Managed Agents (one microVM per Anthropic session; prompts/events/follow-ups go through Anthropic's Managed Agents API, not Isorun's). Also OpenAI Agents/LangChain/CrewAI tool integrations and an MCP server that drive sandboxes, not a harness.",
   "checked": "2026-09-29"
  },
  "egress_allowlist": {
   "value": true,
   "source": "https://docs.isorun.ai/security/network-filtering",
   "quote": "Policy is enforced outside guest code execution.",
   "note": "Per-sandbox network: {allow, deny} at create: CIDRs, IP literals, hostnames matched against SNI, wildcards (*.openai.com). Deny evaluated first; default unrestricted. Named profiles: unrestricted, locked-down (deny-all), claude-code, openai. Enforced at host boundary ('Code running in the sandbox can't edit iptables, swap the resolver, or otherwise disable the policy'). Set at create time; no live-update endpoint listed.",
   "checked": "2026-09-29",
   "level": "both"
  },
  "egress_http_rules": {
   "value": true,
   "source": "https://docs.isorun.ai/security/endpoint-rules",
   "quote": "A request through the credential proxy is forwarded only if its",
   "note": "Endpoint rules (credentials_v2, REST only for now) restrict method + path prefix/glob per credential, e.g. only POST /v1/chat/completions; anything else rejected with 4xx. Applies only to traffic through the credential proxy (the 5 recognised services), not to arbitrary egress.",
   "checked": "2026-09-29"
  },
  "ingress_rules": {
   "value": true,
   "source": "https://docs.isorun.ai/sandboxes/public-urls",
   "quote": "The URL is **authenticated, not public**.",
   "note": "Ports are reachable only via <apiUrl>/v1/runs/<id>/proxy/<port>/ requiring the account API key as Bearer (401 otherwise); no public IP, no raw TCP. Docs contradict themselves: the same page's 'Authentication model' section says 'The URL is anonymous: anyone with the URL can hit it' and the Limits table says 'Auth: none; the sandbox ID is the credential'; the SDK reference sides with 'authenticated'. No per-port exposure choice, IP allowlist or share tokens.",
   "checked": "2026-09-29"
  },
  "guest_firewall": {
   "value": true,
   "source": null,
   "quote": null,
   "note": "'It boots a microVM with its own kernel' (VMM unnamed); root inside OCI images presumably but not stated. Network-filtering page: 'Changes made inside the VM only affect the guest's view', consistent with guest-side rules working but not overriding host enforcement.",
   "checked": "2026-09-29",
   "derived": true
  },
  "secret_proxy": {
   "value": "ai-only",
   "source": "https://docs.isorun.ai/security/credential-injection",
   "quote": "The agent never sees the real secret.",
   "note": "Real host-side reverse proxy; guest sees placeholders (OPENAI_API_KEY=sk-isorun-proxy-managed + *_BASE_URL pointing at 169.254.169.254 proxy). BUT only a fixed set of known services: openai, anthropic, google (Gemini), github (api.github.com), huggingface ('The proxy maps each credential map key to a known service.'). GitHub API makes it slightly more than LLM-only; arbitrary hosts, git-over-SSH, databases not covered. Method/path endpoint rules on top.",
   "checked": "2026-09-29"
  },
  "volume_attach": {
   "value": true,
   "source": "https://docs.isorun.ai/sandboxes/persistent-storage",
   "quote": "Pass `disk` to `create`. `/workspace` *is* the disk",
   "note": "Named persistent disks (Tigris Cloud Disk, real ext4 block device with NVMe write-back cache) mounted at /workspace; createDisk/listDisks/snapshotDisk/forkDisk/deleteDisk; $0.02/GB-month stored; global, attach from any region. Only at create time (one disk per sandbox). Note: published npm isorun@1.0.3 types have no disk option yet (docs ahead of SDK or REST-only).",
   "checked": "2026-09-29"
  },
  "volume_shared": {
   "value": false,
   "source": "https://docs.isorun.ai/sandboxes/persistent-storage",
   "quote": "One sandbox holds a disk at a time.",
   "note": "Tigris lease: 'a concurrent attach fails until the first detaches'. Sharing is via copy-on-write forks (independent copies), not concurrent mounts. No read-only multi-attach.",
   "checked": "2026-09-29"
  },
  "nested_virt": {
   "value": false,
   "source": "https://docs.isorun.ai/getting-started/pricing",
   "quote": "Isorun runs your code in real microVMs, hardware-isolated sandboxes that boot in milliseconds.",
   "note": "not offered: outside the product's scope. No docs mention /dev/kvm or a second virtualization layer inside the guest; microVM hosts (Firecracker-class) essentially never expose nested KVM to tenants, and nothing on the site suggests an exception.",
   "checked": "2026-09-30"
  },
  "anti_bot": {
   "value": false,
   "source": "https://docs.isorun.ai/sandboxes/computer-use-vnc",
   "quote": "The image ships Xvfb (a virtual display), a window manager, Chromium started",
   "note": "not offered: outside the product's scope. The browser/computer-use image is a plain Chromium in a virtual display for agent-driven browsing; no stealth/fingerprint-evasion feature is documented anywhere in the docs. Source moved: /sandboxes/browser-and-computer-use now returns 404; the same text is live at /sandboxes/computer-use-vnc (also in llms-full.txt).",
   "checked": "2026-10-14"
  },
  "captcha_solving": {
   "value": false,
   "source": "https://docs.isorun.ai/sandboxes/computer-use-vnc",
   "quote": "The image ships Xvfb (a virtual display), a window manager, Chromium started",
   "note": "not offered: outside the product's scope. Isorun is a general microVM sandbox/compute runtime; no CAPTCHA-solving API or integration is mentioned anywhere in the docs, changelog, or homepage. Source moved: /sandboxes/browser-and-computer-use now returns 404; the same text is live at /sandboxes/computer-use-vnc (also in llms-full.txt).",
   "checked": "2026-10-14"
  },
  "egress_ip_type": {
   "value": "datacenter",
   "source": "https://docs.isorun.ai/sandboxes/lifecycle",
   "quote": "The VM gets a new guest IP on resume, so any ESTABLISHED outbound TCP connection from before hibernation is reset.",
   "note": "Isorun runs sandboxes as regular cloud-hosted microVMs (US/EU regions) with no public IPv4 and no residential/mobile IP product; outbound traffic is plain datacenter cloud IP space, changing on resume.",
   "checked": "2026-09-30"
  },
  "wake_on_request": {
   "value": false,
   "source": "https://docs.isorun.ai/sandboxes/lifecycle",
   "quote": "Resume restores a hibernated sandbox. The same runID is reused",
   "note": "A hibernated sandbox only comes back via an explicit sandbox.resume() SDK/API call; nothing in the port-forwarding or lifecycle docs describes an incoming HTTP/proxy request auto-waking a stopped or hibernated sandbox.",
   "checked": "2026-09-30"
  },
  "live_resize": {
   "value": false,
   "source": "https://docs.isorun.ai/api-reference",
   "quote": "PATCH /v1/runs/{id} Reset the auto-destroy timer",
   "note": "The only update-in-place API on a running sandbox is PATCH /v1/runs/{id}, which resets the idle timer. There is no endpoint or SDK method to change vCPU/RAM of a running sandbox; resizing requires creating a new sandbox (optionally from a fork/snapshot).",
   "checked": "2026-09-30"
  },
  "own_agent_api": {
   "value": false,
   "source": "https://docs.isorun.ai/integrations/claude-managed-agents",
   "quote": "The Isorun SDK ships a self-hosted adapter that",
   "note": "not offered: outside the product's scope. Isorun is a sandbox/compute runtime; its integrations (Claude Managed Agents, OpenAI Agents, LangChain, CrewAI, MCP server) all drive third-party agent frameworks or your own code inside the sandbox. There is no first-party Isorun browser/computer-use/research agent with its own driving API.",
   "checked": "2026-09-30"
  },
  "lang_python": {
   "value": true,
   "source": "https://docs.isorun.ai/llms-full.txt",
   "quote": "const sandbox = await isorun.create({ image: 'python:3.12-slim' })",
   "note": "Python is available out of the box by selecting the python:3.12-slim (or any Python) OCI image at create() time; nearly every SDK/integration example (code_interpreter tools, CrewAI, LangChain) runs Python this way. No dedicated first-party Python SDK is shipped yet ('Isorun Python SDK (in development)').",
   "checked": "2026-09-30"
  },
  "lang_node": {
   "value": true,
   "source": "https://docs.isorun.ai/sdks/typescript",
   "quote": "Official Node.js client for sandbox lifecycle, command execution, file I/O, snapshots, fork, hibernate, and port exposure.",
   "note": "Node.js is the first-party SDK language (the Isorun client itself is the Node/TypeScript SDK) and the default sandbox base image is node:22-slim per the SDK reference.",
   "checked": "2026-09-30"
  },
  "static_egress_ip": {
   "value": false,
   "source": "https://docs.isorun.ai/sandboxes/lifecycle",
   "quote": "The VM gets a new guest IP on resume, so any ESTABLISHED outbound TCP connection from before hibernation is reset.",
   "note": "The guest's outbound IP changes across hibernate/resume cycles, and no fixed/static egress IP product is documented or priced anywhere (network.ipv4_available is false in this card).",
   "checked": "2026-09-30"
  },
  "self_host": {
   "value": false,
   "source": "https://docs.isorun.ai/getting-started/pricing",
   "quote": "Isorun runs your code in real microVMs, hardware-isolated sandboxes that boot in milliseconds.",
   "note": "not offered: outside the product's scope. Isorun is a hosted SaaS runner (US/EU regions, per-second billing via API key); there is no downloadable runtime/appliance to run the microVM host on your own infrastructure. 'Self-hosted' in the Claude Managed Agents integration refers to Isorun acting as the self-hosted sandbox *for Anthropic's* Managed Agents API, not to running Isorun itself on-prem.",
   "checked": "2026-09-30"
  },
  "byoc": {
   "value": false,
   "source": "https://docs.isorun.ai/getting-started/pricing",
   "quote": "Isorun runs your code in real microVMs, hardware-isolated sandboxes that boot in milliseconds.",
   "note": "not offered: outside the product's scope. No docs, pricing, or homepage content mentions deploying the Isorun runner into a customer-owned AWS/GCP/Azure account; it is a single multi-tenant SaaS with a US or EU region choice.",
   "checked": "2026-09-30"
  },
  "open_source": {
   "value": false,
   "source": "https://github.com/isorun/sdk",
   "quote": "\"private\":false",
   "note": "Only the client SDK (github.com/isorun/sdk, MIT-licensed) is open source; the sandbox runtime/microVM host itself is closed-source hosted infrastructure with no public repo or self-host distribution.",
   "checked": "2026-09-30"
  },
  "hipaa": {
   "value": false,
   "source": "https://isorun.ai/",
   "quote": "SOC 2 Type II (soon)",
   "note": "not offered: outside the product's scope for now. The homepage lists only a pending SOC 2 Type II attestation as a compliance milestone for this public-beta product; no HIPAA/BAA mention anywhere in docs or site.",
   "checked": "2026-09-30"
  },
  "sso": {
   "value": false,
   "source": "https://docs.isorun.ai/getting-started/pricing",
   "quote": "No plan fee. $50 one-time signup credit, no card.",
   "note": "not offered: outside the product's scope for now. This is a self-serve public-beta API product (sign up, get an API key); no SAML/OIDC/SSO feature, enterprise auth page, or mention is documented anywhere.",
   "checked": "2026-09-30"
  },
  "max_session_h": {
   "value": "none",
   "source": "https://docs.isorun.ai/sandboxes/lifecycle",
   "quote": "Use sandbox.setTimeout(seconds) to extend it during a long-running workflow, pass 0 to disable.",
   "note": "No documented hard cap on total sandbox lifetime; only an idle auto-destroy timer exists (default 300s), which can be disabled (timeoutSec 0) or reset via PATCH, and a running/actively-used sandbox otherwise persists (billed) until you call destroy().",
   "checked": "2026-09-30"
  },
  "ssh": {
   "value": false,
   "source": "https://docs.isorun.ai/llms-full.txt",
   "quote": "sandbox.exec(cmd, timeoutSec)",
   "note": "Full docs dump (2752 lines) list only exec(), WebSocket PTY/shell sessions, and HTTPS port proxying as access methods. The only 'SSH' hit in the whole corpus is 'a long-running node server.js, a jupyter notebook kernel, an SSH agent, they all survive' hibernation - describing an in-guest process surviving, not an SSH access method offered by Isorun. No sshd, no SSH key management, no ssh:// endpoint anywhere in docs or homepage.",
   "checked": "2026-09-30"
  },
  "public_ipv4": {
   "value": false,
   "source": "https://docs.isorun.ai/sandboxes/public-urls",
   "quote": "The sandbox ID is the only credential on a public URL.",
   "note": "Only access to sandbox network services is via HTTPS/WebSocket proxy URLs of the form <apiUrl>/v1/runs/<runID>/proxy/<port>/...; no dedicated/public IPv4 address is issued or documented anywhere ('ipv4', 'static ip', 'public ip' do not appear in the docs corpus). Consistent with raw_tcp_inbound already false.",
   "checked": "2026-09-30"
  },
  "agent_harnesses": {
   "value": false,
   "source": "https://docs.isorun.ai/integrations/claude-managed-agents",
   "quote": "The SDK ships a self-hosted adapter for Anthropic Claude Managed Agents. Each Anthropic session is bound to its own sandbox for the lifetime of the session.",
   "note": "runOrchestrator() and the OpenAI Agents/LangChain/CrewAI/MCP integrations are client-side adapters that drive Isorun sandboxes from other vendors' agent frameworks/harnesses; Isorun itself exposes no API to start/stream/manage an agent harness (its REST/SDK surface is sandbox lifecycle, exec, PTY/shell, files, snapshots, network - matches the card's existing harness_api:false evidence).",
   "checked": "2026-09-30"
  },
  "gpu": {
   "value": false,
   "source": "https://docs.isorun.ai/llms-full.txt",
   "quote": "sandbox is a real Linux box with its own kernel, built for AI agents",
   "note": "No mention of 'gpu' anywhere in the 2752-line full docs dump, the homepage, or the lifecycle/pricing pages; the product is described purely as CPU/RAM-billed microVMs (vCPU-h + GiB-h pricing, no GPU SKU or GPU image). DuckDuckGo search for 'isorun.ai GPU sandbox' returns nothing relevant. Product is a general-purpose CPU agent-sandbox, not a GPU offering.",
   "checked": "2026-09-30"
  },
  "windows": {
   "value": false,
   "source": "https://docs.isorun.ai/llms-full.txt",
   "quote": "Each isorun.create() boots a microVM with its own kernel and defaults to",
   "note": "Every example boots Linux OCI images (e.g. 'isorun.create({ image: \"ubuntu:24.04\" })'); no Windows image, guest OS option, or mention of 'windows' anywhere in the docs corpus or homepage. Isorun is a Linux-only microVM sandbox product.",
   "checked": "2026-09-30"
  },
  "macos": {
   "value": false,
   "source": "https://docs.isorun.ai/llms-full.txt",
   "quote": "Each microVM gets its own kernel.",
   "note": "No mention of macOS/Darwin guest support anywhere in docs or homepage; sandboxes run Linux OCI images on KVM microVMs. macOS guest virtualization is outside what this product is.",
   "checked": "2026-09-30"
  },
  "arm64": {
   "value": false,
   "source": "https://www.computesdk.com/benchmarks/sandboxes/",
   "quote": "8 vCPU / 15.7 GiB x86_64 env",
   "note": "The card's own third-party benchmark source records the test environment as x86_64; no arm64/aarch64 architecture option is mentioned anywhere in Isorun's docs, homepage, or changelog. No evidence of an ARM64 image/architecture choice.",
   "checked": "2026-09-30"
  },
  "soc2": {
   "value": false,
   "source": "https://isorun.ai/",
   "quote": "SOC 2 Type II (soon)",
   "note": "Homepage pricing feature list explicitly marks SOC 2 Type II as upcoming/not-yet-certified, confirming false for the current state.",
   "checked": "2026-09-30"
  },
  "docker_inside": {
   "value": true,
   "source": "https://docs.isorun.ai/llms-full.txt",
   "quote": "sandbox is a real Linux box with its own kernel, built for AI agents",
   "note": "Not explicitly documented either way (the card's own caveats already flag 'docker-in-VM ... unpublished'). Reasoned: each sandbox is a hardware-isolated microVM with its own full guest Linux kernel (KVM), not a shared-kernel container, and images are arbitrary OCI images run as the sandbox's root filesystem, so running a Docker/containerd daemon inside the guest (cgroups+namespaces, no nested KVM required) is architecturally unblocked as long as the guest has root - which is the norm for this class of full-VM sandbox. No page confirms or denies it directly, hence medium confidence rather than high.",
   "checked": "2026-09-30"
  },
  "fast_boot": {
   "value": true,
   "source": "https://www.computesdk.com/benchmarks",
   "quote": "burst of 100: p95 49 ms, 100% started",
   "note": "ComputeSDK sandbox benchmark 2026-09-25",
   "checked": "2026-09-25"
  },
  "usage_api": {
   "value": true,
   "source": "https://docs.isorun.ai/api-reference",
   "quote": "A 1 vCPU / 1 GiB sandbox with uptime_ms 59000 but cpu_ms 48 and mem_peak 49 MB returns cost_cents 0.0007, which equals allocation price in dollars. The 'cost_cents' / costCents field appears in the response.",
   "note": "granularity: per-sandbox; cost: yes; The Isorun API returns per-sandbox usage metrics (uptime_ms, cpu_ms, mem_peak) and cost_cents (billing cost) for each sandbox, broken down by run. Granularity is per-sandbox; no per-tag or per-project API documented.",
   "confidence": "high",
   "checked": "2026-10-01"
  }
 },
 "locations": [
  {
   "country": "US",
   "state": null,
   "city": null,
   "provider_region": "us",
   "self_serve": true,
   "choosable": true
  },
  {
   "country": "EU",
   "state": null,
   "city": null,
   "provider_region": "eu",
   "self_serve": true,
   "choosable": true
  }
 ],
 "locations_source": "https://docs.isorun.ai/getting-started/authentication",
 "locations_quote": "the organization is scoped to a region (US or EU)",
 "locations_checked": "2026-10-01"
}