ThunderAgent
An open-source agentic workflow serving system delivering up to 3.6x throughput improvements by treating agent workflows as first-class scheduling units.
ThunderAgent serves and trains agentic AI workloads at scale using program-aware abstractions. It addresses KV cache thrashing, cross-node memory imbalance, and tool lifecycle management. Delivers 1.5-3.6x throughput for serving, 1.8-3.9x for RL rollout, and 4.2x disk memory savings. Integrates directly with Together's RL API.
Use cases
Serving coding agents and multi-step reasoning pipelines
Distributed RL rollout
Scientific discovery agent orchestration
High-throughput inference for tool-using agents
Integrations
Together AI RL API
NVIDIA B200 GPUs
distributed GPU clusters
open-source model frameworks
Expected outcomes
Up to 3.6x throughput improvement
4.2x disk memory reduction
Eliminated KV cache thrashing
Reviews
GCC buyer feedback
Reviews are written by GCC buyers and published after moderation.
No reviews yet
Buyer reviews will appear here once published.
Company
Together AI
San Francisco, USA
Founded 2022
Primary Verticals
Financial Services (BFSI)Healthcare & Life SciencesRetail & E-commerceTelecom & Media
Functions
Engineering & DevOpsData & AnalyticsIT Operations
Industries
Financial Services (BFSI)Healthcare & Life Sciences
4
Integrations
4
Use cases
Is this your company? Claim & customize your profile
This profile was created using publicly available information.