Weni ⚡️

Enterprise-grade AI agents for retail & commerce — born in LATAM, now powering large-scale customer support and conversational commerce across WhatsApp and beyond.

Weni is a VTEX business unit focused on production-ready, cost-efficient, multi-agent AI systems that deliver measurable outcomes (e.g., ~75% automation and ~20% support cost reduction for enterprise retailers in Brazil).

What we build

• Customer Support Agents (CS)
  Multi-agent orchestration for L1/L2 support, policy grounding, escalations, and KPI tracking.

• Agent Platform & Tooling
  Patterns and templates for Amazon Bedrock Agents (inline & managed), including tool calling, memory/session design, guardrails, and post-processing.

• Eval & Reliability
  Agent evaluation harnesses, cost/latency telemetry, prompt-caching strategies, and hallucination detection/regeneration pipelines.

On this Hugging Face org you’ll find

• Models: lightweight components for retail intents, policy routing, and post-processing.
• Datasets: anonymized, domain-specific corpora for traning and eval.

Licenses vary per repo — see the card of each asset for details.

Important note on models

Models are not supported anymore — they’re mostly ablation experiments (a bit of a mess of ablations).
Use at your own risk; they’re kept for reference only.