{
 "features_resolved": {
  "snapshot_any": false,
  "snapshot_mem": false,
  "fork": false,
  "pause_resume": false,
  "persistent_disk": false,
  "whole_disk": false,
  "persist_scope": null,
  "volumes": true,
  "auto_stop_idle": true,
  "long_sessions": true,
  "custom_image": true,
  "image_snapshot": false,
  "image_oci": true,
  "vm_isolation": false,
  "docker_inside": false,
  "nested_virt": false,
  "root": null,
  "systemd": null,
  "browser": false,
  "desktop": false,
  "computer_use": false,
  "code_interpreter": false,
  "browser_control": null,
  "desktop_control": null,
  "browser_and_desktop": false,
  "anti_bot": false,
  "captcha_solving": false,
  "residential_ip": false,
  "agent_harnesses": false,
  "ssh": false,
  "public_ipv4": false,
  "inbound_https": false,
  "custom_domain": null,
  "raw_tcp_inbound": null,
  "egress_allowlist": false,
  "open_internet": true,
  "static_egress_ip": false,
  "private_network": null,
  "self_host": true,
  "byoc": null,
  "open_source": false,
  "soc2": null,
  "hipaa": null,
  "sso": null,
  "gpu": true,
  "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": false,
  "wake_on_request": false,
  "live_resize": false,
  "fork_running": false,
  "memory_snapshot_fork": false,
  "webhooks": null,
  "scheduled_wakeups": null,
  "audit_logs": null,
  "eu_data_residency": null,
  "zero_data_retention": null,
  "scoped_api_keys": null,
  "spend_limits": null,
  "mcp_server": false,
  "usage_api": false,
  "snapshot_auto": false,
  "snapshot_on_demand": false,
  "harness_api": false,
  "own_agent_api": false,
  "fast_boot": null,
  "strong_isolation": false,
  "ingress_rules": false,
  "guest_firewall": false,
  "secret_proxy": true,
  "secret_proxy_any": true,
  "volume_attach": true,
  "volume_shared": true,
  "sidecar": null,
  "fuse": null,
  "kernel_features": null,
  "burst": null
 },
 "company": {
  "id": "openreward",
  "checked": "2026-09-30",
  "company": "General Reasoning Ltd (\"GR\"), product: OpenReward",
  "ownership": {
   "value": "subsidiary",
   "parent": "General Reasoning (GR)",
   "source": "https://openreward.ai/",
   "quote": "From the team behind Papers with Code, Llama 2 and Llama 3"
  },
  "stage": {
   "value": "unknown",
   "source": "",
   "quote": "",
   "confidence": "low",
   "note": "OpenReward is a product/open-source release of General Reasoning (GR), a London-based AI research lab founded 2025 by ex-Meta researchers (Galactica, Llama 2/3 team, led by Ross Taylor). No funding round, investor, or valuation was disclosed on GR's own site (gr.inc, gr.inc/news, gr.inc/careers) or in HN discussion of GR's Feb 2025 launch. No Crunchbase/press coverage of a priced round was found within the search budget (Crunchbase/LinkedIn blocked search tools; web search engines returned no hits for '\"General Reasoning\" Ross Taylor funding')."
  },
  "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": "OpenReward (openreward.ai) is not an independent company; it is an open platform/product built by General Reasoning (GR), an AI research lab founded in 2025 in London (Shoreditch), described as run 'From the team behind Papers with Code, Llama 2 and Llama 3' and led by CEO Ross Taylor. GR's core business is training long-horizon reasoning models; OpenReward (launched 24 March 2026) is GR's open resource for hosting/training/evaluating agents on RL environments, built on the 'Open Reward Standard', with core sandbox infra plus first-party 'Firehorse' harness tooling partly open-sourced under the GeneralReasoning GitHub org. No funding, investor, valuation, accelerator, or revenue information is publicly disclosed by GR/OpenReward; nothing was found via HN Algolia, Bing, or DuckDuckGo search (search engines returned no hits or were bot-blocked) within the time budget. Given GR looks like a well-resourced research lab founded by senior ex-Meta AI researchers, it is plausible it has raised institutional funding, but this is unconfirmed — reported here as unknown rather than guessed."
 },
 "id": "openreward",
 "name": "OpenReward Sandboxes",
 "url": "https://openreward.ai/pricing",
 "category": "agent-sandbox",
 "isolation": "container",
 "modes": [
  {
   "key": "on-demand",
   "label": "Pay-as-you-go, per second",
   "flags": [],
   "os": [
    "linux"
   ],
   "regions": null,
   "gpu_only": false,
   "plan_limits_exempt": false,
   "features": {
    "isolation": "container"
   },
   "plans": null,
   "vm_max_vcpu": null,
   "vm_max_ram_gib": null,
   "pricing": "resource",
   "vcpu_h": 0.0504,
   "ram_gib_h": 0.0162,
   "cpu_basis": "alloc",
   "ram_basis": "alloc",
   "active_floor": 0,
   "ram_per_vcpu": [
    1,
    4
   ],
   "ram_fixed_per_vcpu": null,
   "vcpu_options": [
    0.5,
    1,
    2,
    4
   ],
   "min_vcpu": 0.5,
   "max_vcpu": 4,
   "max_ram_gib": 16,
   "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": {
    "L4": 1.51
   },
   "note": "List: $0.0504/vCPU-h, $0.0162/GiB-h, NVIDIA L4 $1.51/GPU-h, billed by the second. Presets CPU:RAM 0.5:0.5 ... 4:16 (RAM 1-4 GiB per vCPU in 1x/2x/4x steps). 4 vCPU/8 GiB = $0.3312/h. When used through a hosted OpenReward environment, the environment server is a separate, simultaneously billed component."
  }
 ],
 "storage": {
  "included_disk_gib": null,
  "disk_gib_month": null,
  "disk_billed_when_stopped": false,
  "snapshot_gib_month": null,
  "paused_ram_billed": false
 },
 "network": {
  "egress_free_gib": null,
  "egress_gib": null,
  "ipv4_month": null,
  "ipv4_available": false,
  "internet": {
   "default": "configurable",
   "evidence": [
    {
     "quote": "By default all egress - all internet access - is allowed",
     "url": "https://docs.openreward.ai/concepts/sandboxes.md"
    }
   ]
  }
 },
 "plans": [
  {
   "name": "Pay-as-you-go",
   "fee": 0,
   "flags": [],
   "fee_is_credit": false,
   "included_usd": 0,
   "per_seat": 0,
   "waives_start_fee": false,
   "included_applies_to_modes": null,
   "concurrency": null,
   "max_session_h": null,
   "max_vcpu": 4,
   "max_ram_gib": 16,
   "sizes_allowed": null,
   "note": "Prepaid credits per organization, topped up at openreward.ai/<org>/billing. When the balance is exhausted, API calls return 402 and usage stops (a hard stop, not overage). There is no public free tier or signup credit; academic credits are on request."
  },
  {
   "name": "Enterprise",
   "fee": null,
   "flags": [
    "sales"
   ],
   "fee_is_credit": null,
   "included_usd": null,
   "per_seat": null,
   "waives_start_fee": false,
   "included_applies_to_modes": null,
   "concurrency": null,
   "max_session_h": null,
   "max_vcpu": null,
   "max_ram_gib": null,
   "sizes_allowed": null,
   "note": "On-prem, dedicated support, custom SLAs; contact sales."
  }
 ],
 "free": {
  "monthly_credit": 0,
  "one_time_credit": 0
 },
 "features": {
  "linux": true,
  "windows": false,
  "macos": false,
  "arm64": false,
  "gpu": true,
  "gpu_types": [
   "L4"
  ],
  "snapshot": "none",
  "pause_resume": false,
  "fork": false,
  "persistent_disk": false,
  "volumes": true,
  "auto_stop_idle": true,
  "max_session_h": null,
  "custom_image": true,
  "docker_inside": false,
  "nested_virt": false,
  "root": null,
  "systemd": null,
  "browser": false,
  "desktop": false,
  "computer_use": false,
  "code_interpreter": false,
  "agent_harnesses": false,
  "ssh": false,
  "public_ipv4": false,
  "inbound_https": false,
  "custom_domain": null,
  "raw_tcp_inbound": null,
  "egress_allowlist": false,
  "static_egress_ip": false,
  "private_network": null,
  "self_host": true,
  "byoc": null,
  "open_source": false,
  "regions": null,
  "soc2": null,
  "hipaa": null,
  "sso": null,
  "lang_python": true,
  "lang_node": true,
  "lang_go": null,
  "lang_rust": null,
  "lang_java": null,
  "gpu_desktop": false,
  "credential_injection": true,
  "egress_http_rules": false,
  "wake_on_request": false,
  "live_resize": false,
  "fork_running": false,
  "memory_snapshot_fork": false,
  "webhooks": null,
  "scheduled_wakeups": null,
  "audit_logs": null,
  "eu_data_residency": null,
  "zero_data_retention": null,
  "scoped_api_keys": null,
  "spend_limits": null,
  "mcp_server": false,
  "isolation": "container",
  "image_oci": true,
  "snapshot_auto": false,
  "snapshot_on_demand": false,
  "harness_api": false,
  "secret_proxy": "any-http",
  "volume_attach": true,
  "volume_shared": true,
  "anti_bot": false,
  "captcha_solving": false,
  "egress_ip_type": "datacenter",
  "own_agent_api": false,
  "ingress_rules": false,
  "guest_firewall": false,
  "open_internet": true,
  "usage_api": false
 },
 "perf": {
  "cpu_runs_s": null,
  "cold_start_ms": null,
  "burst_tti_ms": null,
  "dax_s": null
 },
 "knobs": [],
 "caveats": [
  "Sandboxes are tied to an OpenReward environment workspace ('not a general sandbox compute solution'). Usable for RL/eval environments, not as a general agent VM.",
  "Only preset CPU:RAM sizes up to 4 vCPU / 16 GiB; no larger machines.",
  "The nvidia-l4 preset's vCPU/RAM and whether CPU/RAM are billed on top of $1.51/GPU-h are undocumented; the engine adds the GPU rate on top of resource rates.",
  "Concurrency limits, max lifetime, minimum billed duration, disk and egress pricing are undocumented (null).",
  "No pause/snapshot; the network option is all-or-nothing block_network (no allowlist).",
  "CPU/RAM rates are numerically identical to E2B's list rates; the backend is not disclosed.",
  "Academic credits are available on request (not modelled).",
  "SSH: not offered in the docs (sandbox concepts/openreward pages, llms.txt index list no SSH); access is through the environment session SDK. Disk size and disk price are not published (pricing page lists CPU, memory and L4 GPU only), so extra disk stays unpriced (2026-09-28).",
  "Hosted OpenReward environments bill an environment server in addition to the sandbox. Published examples use 1 vCPU/4 GiB = $0.0000320/s ($0.1152/h) at the public CPU+RAM rates. For example, JobTasks with a 2 vCPU/2 GiB sandbox totals $0.2484 per session-hour.",
  "Machine sizes: docs (concepts/sandboxes) list 0.5:0.5 to 4:16 plus nvidia-l4. The official Python SDK added 8:8, 16:16, 8:16, 8:32 and 16:64 in v0.1.159 (2026-09-24), and backend availability of these sizes isn't documented. A separate docs page (sandboxes/openreward) claims sizes 'From 0.1:0.1', but no 0.1 size exists in the SDK or the size list.",
  "Custom sandbox images must be public on Docker Hub and built for linux/amd64 (no private registries, no arm64).",
  "Hosted-environment server sizes vary by environment: 1 vCPU/4 GB is the default (CLI default --cpu-memory 1:4), and 2:4 and 4:8 have also been seen. Each is billed per second of each session at the same CPU/RAM rates. Examples: DataScienceComps environment $0.0000920/s plus sandbox $0.0000640/s = $0.5616 per session-hour; KernelBench environment $0.0000460/s plus a task-dependent sandbox."
 ],
 "sources": [
  "https://openreward.ai/pricing",
  "https://docs.openreward.ai/concepts/sandboxes.md",
  "https://docs.openreward.ai/sandboxes/openreward.md",
  "https://docs.openreward.ai/environments/gpu-environments.md"
 ],
 "fleet": {
  "default_max_instances": null,
  "default_max_vcpu": null,
  "default_max_gpus": null,
  "raise": null,
  "raise_time_days": null,
  "verification": "Prepaid organization credits; further verification unknown.",
  "tiers": null,
  "notes": "Rendered PAYG/Enterprise pricing establishes prepaid usage and custom/on-prem option, not numeric concurrent-environment quota. No verified default count, GPU allowance, verification checklist or approval time. Hosted environment servers default to concurrency 500 sessions per pod and max-scale 10 pods (owner-configurable via `orwd update-link`); beyond that callers get 429 \"Environment at maximum capacity\". No per-account sandbox concurrency or creation-rate limit is published.",
  "sources": [
   "https://openreward.ai/pricing"
  ],
  "researched": "2026-09-28"
 },
 "audit_2026_09_29": {
  "applied": 4,
  "source": "checked against the provider's live pricing and docs pages",
  "reviewed": 4
 },
 "feature_evidence": {
  "snapshot_auto": {
   "value": false,
   "source": "https://docs.openreward.ai/concepts/sandboxes.md",
   "quote": "which will terminate the sandbox as soon as the session ends.",
   "note": "Lifecycle is start() -> run/upload/download -> stop() (deletion). No pause, sleep or state save documented; SDK 0.1.160 sandbox client exposes only run, check_run, upload, download, start, stop.",
   "checked": "2026-09-29"
  },
  "snapshot_on_demand": {
   "value": false,
   "source": "https://docs.openreward.ai/concepts/sandboxes.md",
   "quote": "Available operations:\n\n* `run()`: Execute command and return output\n* `check_run()`: Execute command, raise on error if it returns a non-zero exit code\n* `upload()`: Upload files to sandbox\n* `download()`: Download files from sandbox",
   "note": "No snapshot/checkpoint/fork API in docs or SDK 0.1.160. Only custom (public Docker Hub) images as templates.",
   "checked": "2026-09-29"
  },
  "harness_api": {
   "value": false,
   "source": "https://docs.openreward.ai/harnesses/quickstart.md",
   "quote": "This launches Claude Code as a subprocess against the terminal-bench-2-verified environment, running 5 tasks from the test split.",
   "note": "Close but inverted: first-party Firehorse CLI (pip install firehorse-cli) runs claude-code, codex, gemini, resum, react harnesses on the CALLER's machine and proxies their tools (via MCP / session-scoped 'harness toolsets' claude-code, codex, gemini-cli, OpenClaw, Hermes) into the sandbox; writes trajectory.jsonl. The harness does not run inside the sandbox, and there is no API to start/stream/resume an in-sandbox agent.",
   "checked": "2026-09-29"
  },
  "egress_allowlist": {
   "value": false,
   "source": "https://docs.openreward.ai/concepts/sandboxes.md",
   "quote": "By default all egress - all internet access - is allowed, but there may be use cases where this is problematic: ... To enable network isolation, set `block_network=True`",
   "note": "Only an all-or-nothing switch: block_network cuts all outbound traffic (SDK: 'disables outbound network access (Kubernetes egress policies required)'). No per-domain or per-IP allow/deny list is documented (concepts/sandboxes, sandboxes/openreward), so not an egress allowlist; open_internet stays true. The secret proxy's per-secret allowed hosts restrict where secrets are injected, not where traffic may go.",
   "checked": "2026-10-14"
  },
  "egress_http_rules": {
   "value": false,
   "source": "https://docs.openreward.ai/concepts/sandboxes.md",
   "quote": "To enable network isolation, set `block_network=True`",
   "note": "Only network control is the block_network boolean; no method/path/header rules documented.",
   "checked": "2026-09-29"
  },
  "secret_proxy": {
   "value": "any-http",
   "source": "https://docs.openreward.ai/environments/keeping-secrets-secret.md",
   "quote": "Rather than exposing raw secret values inside your environment or sandbox, OpenReward replaces them with placeholders before they reach your server. The actual values are injected on egress - when outbound HTTP requests leave the environment or sandbox, OpenReward swaps the placeholders back in for allowed destination hosts only.",
   "note": "Session `secrets` dict; well-known keys (OpenAI, Anthropic, Gemini, Tavily, HF, Kaggle, E2B, Modal, Daytona...) have default hosts, arbitrary keys via (value, [hosts]) tuple: 'Custom hosts for a secret without defaults'. HTTP only per docs; no DB/SSH/git coverage documented. Scope is per ORS session (environment + sandbox).",
   "checked": "2026-09-29"
  },
  "volume_attach": {
   "value": true,
   "source": "https://docs.openreward.ai/storage/buckets.md",
   "quote": "Configure storage access via `bucket_config` in `SandboxSettings`",
   "note": "Only the environment's own cloud bucket, mounted read-only ('Buckets are always mounted read-only'); no writable block/network volumes. SDK ephemeral disk_request/disk_limit fields are commented out.",
   "checked": "2026-09-29"
  },
  "volume_shared": {
   "value": true,
   "source": "https://docs.openreward.ai/storage/buckets.md",
   "quote": "Key Concept: The same storage is accessible in both environments and sandboxes.",
   "note": "Read-only: `read_only: Literal[True]` 'Buckets are always mounted read-only'. Every sandbox of the environment can mount the same bucket (optionally a subdirectory via only_dir) concurrently.",
   "checked": "2026-09-29"
  },
  "docker_inside": {
   "value": false,
   "source": "https://docs.openreward.ai/concepts/sandboxes.md",
   "quote": "A sandbox is an isolated container for executing arbitrary code.",
   "note": "not offered: outside the product's scope. Isolation is `container` (shared kernel); no privileged mode or docker-in-docker is documented anywhere in the sandbox/docker-images docs.",
   "checked": "2026-09-30"
  },
  "nested_virt": {
   "value": false,
   "source": "https://docs.openreward.ai/concepts/sandboxes.md",
   "quote": "A sandbox is an isolated container for executing arbitrary code.",
   "note": "not offered: outside the product's scope. Container isolation (shared kernel) with no /dev/kvm or privileged-mode option documented.",
   "checked": "2026-09-30"
  },
  "browser": {
   "value": false,
   "source": "https://docs.openreward.ai/concepts/sandboxes.md",
   "quote": "Sandboxes are used in conjunction with environments when an agent needs access to a computer. For example, a software engineering task requires interacting with a filesystem, writing and executing files, and more.",
   "note": "not offered: outside the product's scope. No browser/Playwright/Chromium/CDP feature is mentioned anywhere in the docs; sandboxes are plain code-execution containers for RL/eval environments.",
   "checked": "2026-09-30"
  },
  "anti_bot": {
   "value": false,
   "source": "https://openreward.ai/pricing",
   "quote": "Pay-as-you-go environments. Simple, transparent pricing. You only pay for the resources you use, billed by the second.",
   "note": "not offered: outside the product's scope. This is an RL/eval sandbox and environment-hosting product with no browser or web-automation stealth feature.",
   "checked": "2026-09-30"
  },
  "captcha_solving": {
   "value": false,
   "source": "https://openreward.ai/pricing",
   "quote": "Pay-as-you-go environments. Simple, transparent pricing. You only pay for the resources you use, billed by the second.",
   "note": "not offered: outside the product's scope. No CAPTCHA-solving feature is documented; product is a code sandbox / RL environment host, not a browser-automation service.",
   "checked": "2026-09-30"
  },
  "egress_ip_type": {
   "value": "datacenter",
   "source": "https://docs.openreward.ai/concepts/sandboxes.md",
   "quote": "By default all egress - all internet access - is allowed",
   "note": "derived: sandboxes are cloud-hosted containers with plain internet egress; no residential/mobile proxy or static-IP offering is documented, so outbound traffic is assumed to use ordinary datacenter IPs.",
   "checked": "2026-09-30"
  },
  "auto_stop_idle": {
   "value": true,
   "source": "https://docs.openreward.ai/concepts/environments.md",
   "quote": "Time-limited: Sessions expire after inactivity or when the session is deleted.",
   "note": "Sandbox lifetime is tied to the environment session (`teardown()` calls `sandbox.stop()`); when the session expires from inactivity the sandbox is torn down with it. Not a dedicated per-sandbox idle timer distinct from session expiry.",
   "checked": "2026-09-30"
  },
  "live_resize": {
   "value": false,
   "source": "https://docs.openreward.ai/concepts/sandboxes.md",
   "quote": "machine_size=\"1:2\",                       # Required: CPU:Memory",
   "note": "`machine_size` is set once in `SandboxSettings` at creation; no API to change CPU/RAM of a running sandbox is documented.",
   "checked": "2026-09-30"
  },
  "own_agent_api": {
   "value": false,
   "source": "https://docs.openreward.ai/harnesses/quickstart.md",
   "quote": "This launches Claude Code as a subprocess against the terminal-bench-2-verified environment, running 5 tasks from the test split.",
   "note": "not offered: outside the product's scope. OpenReward's `firehorse` runner drives third-party coding-agent CLIs (Claude Code, Codex, Gemini, etc.) against sandboxes; there is no first-party OpenReward-owned browser/computer-use/research agent product.",
   "checked": "2026-09-30"
  },
  "mcp_server": {
   "value": false,
   "source": "https://docs.openreward.ai/harnesses/quickstart.md",
   "quote": "**Subprocess + MCP agents** (claude-code, codex, gemini) launch the respective CLI as a child process and proxy environment tools via [MCP](https://modelcontextprotocol.io).",
   "note": "OpenReward proxies its own sandbox-backed tools into third-party MCP-based coding CLIs; no standalone MCP server product for controlling OpenReward sandboxes (e.g. from Claude Desktop) was found in the docs.",
   "checked": "2026-09-30"
  },
  "lang_python": {
   "value": true,
   "source": "https://docs.openreward.ai/concepts/sandboxes.md",
   "quote": "pip install openreward",
   "note": "Official SDK is Python (`openreward` package on PyPI); default pre-made sandbox images (`generalreasoning/python-ds:3.12-tools`, `python:3.11-slim`, etc.) are Python-based.",
   "checked": "2026-09-30"
  },
  "lang_node": {
   "value": true,
   "source": "https://docs.openreward.ai/environments/docker-images.md",
   "quote": "| `node:20`          | Node.js environments         |",
   "note": "Node.js is listed as a common out-of-the-box Docker Hub image option for sandboxes; no first-party Node SDK, but a Node runtime image is offered directly.",
   "checked": "2026-09-30"
  },
  "public_ipv4": {
   "value": false,
   "source": "https://docs.openreward.ai/concepts/sandboxes.md",
   "quote": "OpenReward Sandboxes are tied to an environment workspace on OpenReward which provisions compute. This means it is not a general sandbox compute solution.",
   "note": "Sandboxes are internal execution containers reached only via the OpenReward API (run/upload/download); no public IPv4 per sandbox is documented.",
   "checked": "2026-09-30"
  },
  "inbound_https": {
   "value": false,
   "source": "https://docs.openreward.ai/concepts/sandboxes.md",
   "quote": "Available operations: run(), check_run(), upload(), download()",
   "note": "No feature to expose an HTTPS URL to a port on the sandbox is documented; access is only via the SDK's run/upload/download calls.",
   "checked": "2026-09-30"
  },
  "ingress_rules": {
   "value": false,
   "source": "https://docs.openreward.ai/concepts/sandboxes.md",
   "quote": "OpenReward Sandboxes are tied to an environment workspace on OpenReward which provisions compute.",
   "note": "not offered: outside the product's scope. No inbound-port exposure exists at all, so there is nothing to apply ingress/allowlist rules to.",
   "checked": "2026-09-30"
  },
  "guest_firewall": {
   "value": false,
   "source": "https://openreward.ai/pricing",
   "quote": "\"isolation\": \"container\"",
   "note": "derived: isolation is `container` (shared kernel), not a full VM/microVM with root, so in-guest nftables/iptables control is not assumed.",
   "checked": "2026-09-30"
  },
  "static_egress_ip": {
   "value": false,
   "source": "https://docs.openreward.ai/concepts/sandboxes.md",
   "quote": "By default all egress - all internet access - is allowed, but there may be use cases where this is problematic ... To enable network isolation, set `block_network=True`",
   "note": "Only a network on/off (block_network) toggle is documented; no fixed/static outbound IP feature is mentioned.",
   "checked": "2026-09-30"
  },
  "open_source": {
   "value": false,
   "source": "https://github.com/GeneralReasoning",
   "quote": "Ready-to-use harnesses for OpenReward environments",
   "note": "The GeneralReasoning GitHub org publishes open-source client tooling (firehorse harness runner, example environment repos) but the core hosted platform (environment server hosting + OpenReward Sandboxes infrastructure) is closed-source SaaS; no OSI license covers the core product.",
   "checked": "2026-09-30"
  },
  "persistent_disk": {
   "value": false,
   "source": "https://docs.openreward.ai/concepts/sandboxes.md",
   "quote": "await sandbox.stop() ... which will terminate the sandbox as soon as the session ends.",
   "note": "No contradiction: pause_resume/snapshot/fork are all also false (only start()/run()/upload()/download()/stop() exist; stop deletes the container). The card's 'volumes: true' refers to environment cloud-storage buckets mounted into the sandbox 'always mounted read-only' (buckets.md) — a read-only external mount, not a writable persistent disk that survives a stop/pause. So persistent_disk=false is correct as-is.",
   "checked": "2026-09-30"
  },
  "snapshot": {
   "value": "none",
   "source": "https://docs.openreward.ai/concepts/sandboxes.md",
   "quote": "Available operations:\n\n* `run()`: Execute command and return output\n* `check_run()`: Execute command, raise on error if it returns a non-zero exit code\n* `upload()`: Upload files to sandbox\n* `download()`: Download files from sandbox",
   "note": "Full lifecycle documented is create -> start() -> run/upload/download -> stop() (delete). No snapshot/checkpoint API anywhere in the docs or SDK.",
   "checked": "2026-09-30"
  },
  "pause_resume": {
   "value": false,
   "source": "https://docs.openreward.ai/concepts/sandboxes.md",
   "quote": "To delete an OpenReward Sandbox, we use the `stop` method ... which will terminate the sandbox as soon as the session ends.",
   "note": "Only start() and stop() lifecycle calls are documented; stop() terminates (deletes) the sandbox rather than pausing it. No pause/resume method exists.",
   "checked": "2026-09-30"
  },
  "fork": {
   "value": false,
   "source": "https://docs.openreward.ai/concepts/sandboxes.md",
   "quote": "Available operations:\n\n* `run()`\n* `check_run()`\n* `upload()`\n* `download()`",
   "note": "No clone/fork operation in the documented sandbox API (start/run/check_run/upload/download/stop only); the only way to get a 'copy' is starting a new sandbox from the same base image.",
   "checked": "2026-09-30"
  },
  "fork_running": {
   "value": false,
   "source": "https://docs.openreward.ai/concepts/sandboxes.md",
   "quote": "Available operations:\n\n* `run()`\n* `check_run()`\n* `upload()`\n* `download()`",
   "note": "Since no fork capability exists at all (see `fork`), forking a still-running sandbox is a fortiori unsupported.",
   "checked": "2026-09-30"
  },
  "memory_snapshot_fork": {
   "value": false,
   "source": "https://docs.openreward.ai/concepts/sandboxes.md",
   "quote": "Available operations:\n\n* `run()`\n* `check_run()`\n* `upload()`\n* `download()`",
   "note": "No memory-snapshot or checkpoint/restore mechanism is documented anywhere; lifecycle is only start/run/upload/download/stop.",
   "checked": "2026-09-30"
  },
  "ssh": {
   "value": false,
   "source": "https://docs.openreward.ai/concepts/sandboxes.md",
   "quote": "Available operations:\n\n* `run()`: Execute command and return output\n* `check_run()`\n* `upload()`: Upload files to sandbox\n* `download()`: Download files from sandbox",
   "note": "Access is exclusively through the SDK's run/upload/download calls proxied by the OpenReward API; the full docs index (llms.txt) lists no SSH/terminal-access page, and no ssh key or connection-string feature exists anywhere in the docs.",
   "checked": "2026-09-30"
  },
  "desktop": {
   "value": false,
   "source": "https://docs.openreward.ai/concepts/sandboxes.md",
   "quote": "A sandbox is an isolated container for executing arbitrary code. Sandboxes provide secure, temporary compute resources for running untrusted agent code in an ORS environment.",
   "note": "Product is a headless code-execution container for RL/eval environments (run/upload/download/stop API only); no VNC/RDP/GUI desktop feature exists in the docs, so this is outside the product's scope.",
   "checked": "2026-09-30"
  },
  "computer_use": {
   "value": false,
   "source": "https://docs.openreward.ai/concepts/sandboxes.md",
   "quote": "A sandbox is an isolated container for executing arbitrary code.",
   "note": "No screenshot+click computer-use API is documented; the sandbox exposes only run()/check_run()/upload()/download(), consistent with a headless code sandbox rather than a GUI-driving product.",
   "checked": "2026-09-30"
  },
  "wake_on_request": {
   "value": false,
   "source": "https://docs.openreward.ai/concepts/sandboxes.md",
   "quote": "await sandbox.start()  # Creates the sandbox ... await sandbox.stop() ... which will terminate the sandbox as soon as the session ends.",
   "note": "Sandboxes are explicitly created with start() and torn down (deleted) with stop(); there is no idle/sleep state and thus nothing to 'wake' on an incoming request.",
   "checked": "2026-09-30"
  },
  "agent_harnesses": {
   "value": false,
   "source": "https://docs.openreward.ai/harnesses/quickstart.md",
   "quote": "This launches Claude Code as a subprocess against the terminal-bench-2-verified environment, running 5 tasks from the test split.",
   "note": "Coding-agent harnesses (claude-code, codex, gemini, resum, react) run via the first-party 'firehorse-cli' as a subprocess on the CALLER's own machine (proxying tools into the sandbox over MCP/toolsets), not preinstalled inside the sandbox container. The pre-made sandbox images (docker-images.md) are `generalreasoning/python-ds:3.12-tools` and `generalreasoning/knowledge-worker` — neither ships a coding-agent CLI.",
   "checked": "2026-09-30"
  },
  "windows": {
   "value": false,
   "source": "https://docs.openreward.ai/environments/docker-images.md",
   "quote": "OpenReward's sandbox infrastructure runs on `linux/amd64`.",
   "note": "Sandbox infra is explicitly Linux/amd64 container-based; no Windows OS option exists anywhere in the docs.",
   "checked": "2026-09-30"
  },
  "macos": {
   "value": false,
   "source": "https://docs.openreward.ai/environments/docker-images.md",
   "quote": "OpenReward's sandbox infrastructure runs on `linux/amd64`. If you're building custom images on an ARM machine (e.g. Apple Silicon), you'll need to target this platform explicitly — more on this below.",
   "note": "Sandboxes are Linux/amd64 Docker containers only; macOS is not offered as a sandbox OS (Apple Silicon is only mentioned as a client build machine that must cross-compile to linux/amd64).",
   "checked": "2026-09-30"
  },
  "arm64": {
   "value": false,
   "source": "https://docs.openreward.ai/environments/docker-images.md",
   "quote": "OpenReward's sandbox infrastructure runs on `linux/amd64`. If you're building custom images on an ARM machine (e.g. Apple Silicon), you'll need to target this platform explicitly — more on this below.",
   "note": "Docs explicitly state the sandbox backend runs linux/amd64 only; custom images built on ARM machines must be cross-built for amd64, confirming no arm64 sandbox architecture is offered.",
   "checked": "2026-09-30"
  },
  "usage_api": {
   "value": false,
   "source": "https://docs.openreward.ai/concepts/sandboxes.md",
   "quote": "Available operations: run(), check_run(), upload(), download(), start(), stop(). Prepaid credits per organization, topped up at openreward.ai/<org>/billing.",
   "note": "granularity: account-only; cost: yes; Billing is managed at the organization level via a web dashboard; no REST API, SDK method, or CLI command to query billable usage of a specific sandbox is documented. Sandbox API only exposes lifecycle and execution methods.",
   "confidence": "high",
   "checked": "2026-10-01"
  }
 },
 "locations": [],
 "locations_source": "https://docs.openreward.ai/concepts/sandboxes.md",
 "locations_quote": "OpenReward Sandboxes are tied to an environment workspace on OpenReward which provisions compute.",
 "locations_checked": "2026-10-01"
}