A 15-person ML team is replacing LiteLLM after a prod outage and looking at TrueFoundry, Kong, and Portkey
Valuable_Working7557 · reddit · 2026-07-21
A 15-person ML team is looking for LiteLLM alternatives after a production outage caused by a bad upgrade.
- Their current pain points are unreliable dashboards, routing failures in prod, and too much maintenance overhead.
- The main candidates discussed are TrueFoundry, Kong, and Portkey.
- What they care about most:
- per-team cost attribution,
- observability without adding another tool,
- and avoiding a gateway that becomes a new ops project.
- The post asks for 3+ months of real production experience, not vendor marketing.
More from Infra
- DeepSpeed shows AutoEP/AutoTP/AutoSP parallelism gains, hosts office hours — PyTorch · 2026-10-27
- Cloudflare Durable Objects validated as top architecture for AI agent WebSocket servers — tristanbob · 2026-10-27
- Gary Marcus: With Q3 data added, Anthropic's compute-efficiency growth shows deceleration — GaryMarcus · 2026-10-27
- SpaceX locks in $40B GPU financing deal with Apollo, the largest ever committed — TechCrunch AI · 2026-10-27
- Inference is just Kubernetes with a multi-trillion-dollar market, and the communities overlap — matt_slotnick · 2026-10-27
- vLLM eyes Rust migration as inference engines race to rewrite the stack — AccBalanced · 2026-10-27