https://www.kryptonhq.com/docs/Recent content in Documentation on Krypton RuntimeHugoen-ushttps://www.kryptonhq.com/docs/roadmap/Mon, 01 Jan 0001 00:00:00 +0000https://www.kryptonhq.com/docs/roadmap/<p>A forward-looking view of where Krypton is headed. Order within each bucket suggests rough priority; nothing is a commitment and dates aren&rsquo;t promises.</p> <h2 id="recently-shipped">Recently shipped</h2> <table> <thead> <tr> <th>Item</th> <th>Notes</th> </tr> </thead> <tbody> <tr> <td>Built-in model hosting (<code>Model</code> CRD)</td> <td>Host LLMs by Hugging Face name via <a href="https://github.com/ggerganov/llama.cpp">llama.cpp</a>, aggregated behind one OpenAI-compatible endpoint. See <a href="https://www.kryptonhq.com/docs/tutorials/first-llm/">first LLM</a>.</td> </tr> </tbody> </table> <h2 id="high-priority">High priority</h2> <table> <thead> <tr> <th>Item</th> <th>Notes</th> </tr> </thead> <tbody> <tr> <td>GPU-aware scheduling</td> <td><code>spec.gpu</code> on <code>Model</code> / <code>Agent</code>; multi-GPU and MIG slicing</td> </tr> <tr> <td>More inference runtimes</td> <td>vLLM and TGI behind the same <code>Model</code> CRD</td> </tr> <tr> <td>Model autoscaling</td> <td>Concurrency-aware scaling for <code>Model</code> pods (today they&rsquo;re always-on)</td> </tr> <tr> <td>Secure sandbox runtime</td> <td><code>Sandbox</code> CRD with gVisor / Kata runtime classes</td> </tr> <tr> <td>AI-native observability</td> <td>OTEL GenAI conventions; token-usage roll-ups</td> </tr> <tr> <td>OpenTelemetry tracing</td> <td>OTLP exporter; <code>traceparent</code> propagation across hops</td> </tr> </tbody> </table> <h2 id="medium-priority">Medium priority</h2> <table> <thead> <tr> <th>Item</th> <th>Notes</th> </tr> </thead> <tbody> <tr> <td>Serverless mode (GA)</td> <td>Cold-start activator path, off by default today</td> </tr> <tr> <td>Invocation history (Postgres)</td> <td>Queryable history, surfaced in UI</td> </tr> <tr> <td>Schema-driven MCP tool forms</td> <td>Auto-generated inputs per tool from its JSON schema</td> </tr> <tr> <td>MCP resources + prompts</td> <td><code>resources/*</code> and <code>prompts/*</code> method support</td> </tr> <tr> <td>Per-agent network policies</td> <td>Egress / ingress rules generated from spec</td> </tr> <tr> <td>TCP-dial readiness check</td> <td>Closes the kube-proxy race on first cold-start</td> </tr> </tbody> </table> <h2 id="exploring">Exploring</h2> <table> <thead> <tr> <th>Item</th> <th>Notes</th> </tr> </thead> <tbody> <tr> <td>Multi-cluster federation</td> <td>Federated control plane; cross-cluster routing</td> </tr> <tr> <td>Authentication &amp; multi-tenancy</td> <td>OIDC-backed UI/API; per-tenant namespaces</td> </tr> </tbody> </table> <h2 id="out-of-scope">Out of scope</h2> <p>Things people ask about that we deliberately won&rsquo;t build:</p>