File size: 9,527 Bytes
9baba57 639191c 9baba57 639191c 9baba57 639191c 9baba57 639191c 9baba57 639191c 9baba57 639191c 9baba57 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 | <!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>OpenDsec — self-hosted sandboxes for AI agents and RL</title>
<meta name="description" content="OpenDsec is an open-source, self-hosted sandbox platform for reinforcement-learning rollouts and code-executing AI agents. An independent implementation of the DSec paper (arXiv:2609.22978).">
<style>
:root {
--bg: #ffffff;
--fg: #12151a;
--muted: #5c6470;
--line: #e3e6ea;
--accent: #1f5eff;
--code-bg: #f6f7f9;
}
@media (prefers-color-scheme: dark) {
:root {
--bg: #0f1115;
--fg: #e8eaee;
--muted: #9aa3af;
--line: #262b33;
--accent: #6f97ff;
--code-bg: #171a20;
}
}
* { box-sizing: border-box; }
body {
margin: 0;
background: var(--bg);
color: var(--fg);
font: 16px/1.65 -apple-system, BlinkMacSystemFont, "Segoe UI", Inter, Roboto, "Helvetica Neue", Arial, sans-serif;
-webkit-font-smoothing: antialiased;
}
.wrap { max-width: 780px; margin: 0 auto; padding: 64px 24px 96px; }
header { border-bottom: 1px solid var(--line); padding-bottom: 32px; margin-bottom: 40px; }
h1 { font-size: 2.1rem; line-height: 1.2; margin: 0 0 12px; letter-spacing: -0.02em; }
.lede { font-size: 1.08rem; color: var(--muted); margin: 0; }
h2 { font-size: 1.22rem; margin: 44px 0 14px; letter-spacing: -0.01em; }
p { margin: 0 0 14px; }
a { color: var(--accent); text-decoration: none; }
a:hover { text-decoration: underline; }
.note {
border-left: 3px solid var(--line);
padding: 2px 0 2px 16px;
color: var(--muted);
font-size: 0.94rem;
margin: 24px 0;
}
ul { margin: 0 0 14px; padding-left: 20px; }
li { margin-bottom: 8px; }
code {
font: 0.88em/1.5 ui-monospace, SFMono-Regular, Menlo, Consolas, monospace;
background: var(--code-bg);
padding: 1px 5px;
border-radius: 4px;
}
pre {
background: var(--code-bg);
border: 1px solid var(--line);
border-radius: 8px;
padding: 16px 18px;
overflow-x: auto;
font: 0.85rem/1.6 ui-monospace, SFMono-Regular, Menlo, Consolas, monospace;
}
pre code { background: none; padding: 0; }
table { width: 100%; border-collapse: collapse; margin: 18px 0; font-size: 0.94rem; }
th, td { text-align: left; padding: 9px 12px; border-bottom: 1px solid var(--line); }
th { font-weight: 600; color: var(--muted); font-size: 0.82rem; text-transform: uppercase; letter-spacing: 0.04em; }
td:last-child, th:last-child { text-align: right; font-variant-numeric: tabular-nums; }
.links { display: flex; flex-wrap: wrap; gap: 10px; margin-top: 18px; }
.links a {
border: 1px solid var(--line);
border-radius: 6px;
padding: 8px 14px;
font-size: 0.92rem;
}
footer { margin-top: 64px; padding-top: 24px; border-top: 1px solid var(--line); color: var(--muted); font-size: 0.86rem; }
</style>
</head>
<body>
<div class="wrap">
<header>
<h1>OpenDsec</h1>
<p class="lede">Self-hosted sandboxes for AI agents and reinforcement-learning rollouts: Firecracker microVMs, containers and full VMs behind one API, with memory-preserving fork and an E2B-compatible gateway.</p>
<div class="links">
<a href="https://github.com/Kernflux/opendsec">Source code</a>
<a href="https://kernflux.github.io/">Website and docs</a>
<a href="https://kernflux.github.io/en/docs/quickstart">Quickstart</a>
<a href="https://arxiv.org/abs/2609.22978">The DSec paper</a>
</div>
</header>
<div class="note">
OpenDsec is an independent implementation of the architecture described in <em>DeepSeek Elastic Compute (DSec): A Sandbox Infrastructure for Effective Agentic Training at Scale</em> (arXiv:2609.22978). It is not DeepSeek code, not an official release, and not affiliated with or endorsed by that paper's authors.
</div>
<h2>What it is</h2>
<p>OpenDsec runs many isolated sandboxes on your own Linux machines, from a single host to a cluster, for reinforcement-learning rollouts and agents that execute code. It ships a Python SDK, a CLI, and an API compatible with E2B's SDKs, under AGPL-3.0.</p>
<ul>
<li><strong>Good fit:</strong> teams whose sandboxes must stay in their own data centre or an offline network; RL training that starts thousands of environments at once; workflows that need to fork a running environment with its memory; keeping existing E2B SDK code but moving it onto your own infrastructure.</li>
<li><strong>Poor fit:</strong> anyone who wants a managed service with no machines to operate (there is no hosted edition), or non-x86_64 Linux hosts.</li>
</ul>
<h2>Architecture</h2>
<ul>
<li>A <strong>stateless apiserver</strong> handles authentication, quota and placement, choosing nodes from the state a watcher aggregates.</li>
<li><strong>IAM</strong> keeps identities, projects and quotas in a Raft-replicated store.</li>
<li>Each compute node runs an <strong>edge</strong> that makes the final admission decision against its real resources.</li>
<li><strong>Aether</strong> runs inside every sandbox for the runtime channel; <strong>Chronus</strong> keeps persistent shell sessions. Exec, files, HTTP, processes and PTY share one runtime API.</li>
<li>Components talk over gRPC with mTLS, and the sandbox id encodes its edge, so no per-sandbox routing table is needed. Any component can restart without disturbing running sandboxes.</li>
</ul>
<h2>Lifecycle</h2>
<p>A microVM can be paused into a memory snapshot, restored, or forked while running. Copies on the same node share memory pages; a fork on another node pulls the snapshot from the source. A sandbox can also be packed back into a new image from its changes.</p>
<pre><code>import asyncio
from opendsec import DSecClient, DSecRunArgs
async def main():
async with DSecClient() as client:
sb = await client.run_sandbox(DSecRunArgs(
backend="microvm", image="root/py:3.12",
cpu_cores_limit=1, memory_limit_mb=1024,
network_rules={"pypi": True})) # outbound denied unless allowed
try:
print((await sb.run_shell("python3 -c 'print(6*7)'")).stdout)
forks = await sb.fork(count=4) # four copies, memory included
for f in forks:
await f.stop()
finally:
await sb.stop()
asyncio.run(main())</code></pre>
<h2>Measured on our hardware</h2>
<p>These are measurements of OpenDsec on its own machines, not figures from the paper. Method and conditions are in <code>docs/SPEC.md</code>.</p>
<table>
<thead>
<tr><th>Measurement</th><th>Value</th></tr>
</thead>
<tbody>
<tr><td>microVM cold start</td><td>~0.1 s</td></tr>
<tr><td>Restore from memory snapshot</td><td>~55 ms</td></tr>
<tr><td>Resume after pause</td><td>~50 ms</td></tr>
<tr><td>Source VM stall during fork</td><td>~10 ms</td></tr>
<tr><td>Burst of microVMs, one 384-thread node</td><td>5,120</td></tr>
<tr><td>Burst of containers, two nodes</td><td>8,192, no failures</td></tr>
</tbody>
</table>
<h2>Where it differs from the paper</h2>
<ul>
<li>The network policy is enforced with nftables and a DNS proxy rather than eBPF.</li>
<li>Images live on any shared filesystem (NFS, CephFS, JuiceFS) instead of 3FS.</li>
<li>Firecracker is the primary backend rather than one option among several.</li>
</ul>
<p>Where the paper leaves a decision open, <code>docs/SPEC.md</code> marks each point as either taken from the paper, with its section, or the project's own decision.</p>
<h2>Known limits</h2>
<ul>
<li>The image store's availability is whatever the chosen shared filesystem provides.</li>
<li>A host reboot loses that node's sandboxes, paused microVMs included.</li>
<li>QEMU PCI passthrough has not been tested on real devices.</li>
<li>The E2B gateway does not yet cover directory watching, signed file URLs, unauthenticated browser access to sandbox ports, or template building.</li>
</ul>
<h2>Try it</h2>
<p>One Linux x86_64 machine with systemd and cgroup v2 is enough. MicroVMs need <code>/dev/kvm</code>; without it the script sets up containers only.</p>
<pre><code>git clone https://github.com/Kernflux/opendsec
cd opendsec
deploy/quickstart.sh # or --no-microvm for containers only</code></pre>
<p>Existing E2B programs need only three environment variables pointed at the gateway:</p>
<pre><code>export E2B_API_URL=https://sandbox.internal:7400
export E2B_SANDBOX_URL=https://sandbox.internal:7400
export E2B_API_KEY=dsk.…</code></pre>
<h2>Links</h2>
<ul>
<li><a href="https://github.com/Kernflux/opendsec">Source and changelog</a></li>
<li><a href="https://kernflux.github.io/">Website</a></li>
<li><a href="https://kernflux.github.io/en/docs/quickstart">Quickstart</a></li>
<li><a href="https://kernflux.github.io/en/docs/e2b">Moving from E2B</a></li>
<li><a href="https://huggingface.co/papers/2609.22978">The DSec paper on Hugging Face</a></li>
</ul>
<h2>Licensing</h2>
<p>The server and tools are AGPL-3.0. The client parts (the Python SDK, the Go client, and the API definitions) are Apache-2.0, so programs using OpenDsec are not bound by the AGPL.</p>
<footer>
"DeepSeek" and "DSec" are names of their respective owners. OpenDsec uses them only to identify the paper it implements.
</footer>
</div>
</body>
</html>
|