Roadmap
What we are building next.
A public, customer-influenced view of the Mandarum roadmap. Vote and comment from inside the product.
NowQ2 2026
NVIDIA Dynamo GA
Disaggregated prefill/decode serving on H100/H200/Blackwell.
NeMo Guardrails v2
Multilingual content safety with configurable severity levels.
NIM marketplace v1
Browse and deploy optimized NIM microservices with revenue share.
NextQ3 2026
Jetson edge deployment
Serve NIM models on NVIDIA Jetson for ultra-low-latency edge inference.
Federated fine-tuning
Customer-private NeMo fine-tunes without raw-data egress.
Causal eval framework
Beyond pass/fail — measure why an inference run succeeded or failed.
LaterQ4 2026 +
Mandarum Studio
Full IDE for partners and customers to build NIM-powered services on the platform.
Cross-vendor inference protocol
Open standard for inter-model transactions and chained inference.
Blackwell / GB200 NVL72 support
Native support for next-gen NVIDIA rack-scale GPU systems.