Vectara

5 Million Downloads and Counting! The Hughes Hallucination Evaluation Model (HHEM) has now surpassed 5 MILLION downloads on Hugging Face. This milestone underscores the growing demand for AI that is not only powerful, but verifiably trustworthy. And we’re not stopping there. Building on HHEM’s momentum: → Vectara Hallucination Corrector (VHC) is already available, taking you beyond detection to automatic correction → Vectara Agents bring the next wave of Agentic AI: enterprise-grade copilots that are grounded in RAG, governed by Guardian Agents, and designed for real-world business impact. Explore HHEM → bit.ly/3vejcTw Try VHC → bit.ly/43PXqnj Discover Vectara Agents → bit.ly/3Kiz6U1 #GenAI #TrustedAI #AIforEnterprise #LLMEvaluation #HuggingFace #RAGaaS #AIInfra #Vectara

More case studies for agentic failures added to the "awesome-agent-failures" repository Check them out: HHS Chatbot, Amazon Q and Cursor. github.com/vectara/awesom…

Lots of new LLM releases from China - we just updated @vectara's hallucination leaderboard with evaluations for @Zai_org's GLM-5, @qwen's Qwen-3.5 and @MiniMax_AI Minimax-2.5. @deepseek_ai - we're waiting for DeepSeek-v4. Excited to see all this innovation and progress in the open source LLM space. github.com/vectara/halluc…

The future of enterprise AI isn’t just systems of record. It’s systems of context. Systems like Salesforce or SAP tell you what happened. AI agents need to know why it happened. That’s where the Context Graph comes in 👇 It captures: • decision trails • exceptions & edge cases • human reasoning • organizational memory Not just data → understanding. This shift enables AI agents that reason, act safely, and operate with real-world nuance. From recording outcomes to understanding intent. That’s the next platform shift. vectara.com/blog/beyond-th… #AI #EnterpriseAI #ContextGraph #KnowledgeGraph #Vectara

Building AI agents on-prem? Our latest blog breaks down how to architect **robust on-prem AI agent orchestration**—from agent roles to data, compute, governance, and security—so you can move from demos to production with confidence. vectara.com/blog/how-to-ar…

New additions to Awesome Agent Failures: 3 fresh case studies. The most interesting one? You already know. moltbook/openClaw — a goldmine of security vulnerabilities for an AI agent. Learn from other people’s agent mistakes 👇 github.com/vectara/awesom…

📢 Calling all AI enthusiasts! Join AI by Hand ✍️ Seminar: Foundation (AI Eval): a free live Zoom seminar diving into foundational AI concepts like perplexity, cross-entropy, BERTScore, BLEU, ROUGE & more! ✨ 📍 Hosted by Prof. @ProfTomYeh | 🗓️ With guest expert @ofermend (Vectara) 🎟️ Register now 👉 luma.com/y9xvjdz9 🔍 Designed for beginners & curious minds. Don’t miss out! 🙌 #AI #MachineLearning #AIeducation #AICommunity #TechEvents

Context engineering depends on precision, and precision depends on retrieval. This is why Vectara was built on a simple, stubborn belief: retrieval is the most important layer in the AI stack. Larger context windows were supposed to solve enterprise memory. Instead, they delivered degraded performance and runaway latency. The real answer? Pair context with surgical retrieval. Read more about Contextual Engineering here: vectara.com/blog/context-e…

Three new case studies of agent failures added to Awesome-Agent-Failures repo, including the so called "LangGrinch" vulnerability that can be used by an attacker to uncover your secrets (upgrade @LangChain to avoid this). How many more hidden vulnerabilities exist in open source Agentic AI packages? github.com/vectara/awesom…

In a recent Digital IT News piece on AI predictions for the year ahead, @EvaNahari explains why AI agents will require new semantic interfaces and permission structures to retrieve accurate enterprise knowledge. Read the full article here: digitalitnews.com/prepare-for-th… #EnterpriseAI #SemanticLayer #AIGovernance #Vectara

As the year comes to an end, it’s a good moment to pause and look at where enterprise AI is actually heading. We shared our take on why 2026 looks like the moment AI grows up, and what that means for teams building real production systems. Read the full blog below: bit.ly/49fIujS

We've create this app to test your knowledge about agentic AI, just in time for the holidays. agentic-ai-challenge.vectara.com

ICYMI: we've recently released a ton of new features for developing agents with @vectara - e.g. subagents and artifacts. sub-agents: vectara.com/blog/introduci… artifacts: vectara.com/blog/beyond-te… Here is a full cookbook / tutorial (set of Jupyter notebooks) so you can see how it works: github.com/vectara/exampl…

Can you solve these 20 agentic AI problems? Download this problem set I designed 👉 byhand.ai/pset

A key design goal of Open-RAG-Eval was to allow reference-free evaluation of both retrieval (using the UMBRELA score) and generation (using AutoNuggetizer). This is extremely valuable as you scale your RAG pipeline since generating answers manually for a production system is extremely time consuming and often infeasible in practice. Yet sometimes you already have a list of questions and "golden answers" and want to use those - so we added support for that in Open-RAG-Eval through the GoldenAnswersEvaluator and now you can do both. Check out the new version 0.3.0 here: github.com/vectara/open-r… Or "pip install open-rag-eval"

When AI agents look impressive in demos but struggle in real workflows A recent article on Microsoft Copilot highlights a challenge many enterprises are running into as they scale agentic AI. Moving from pilot to production exposes gaps in reliability, accuracy, and real workflow integration that demos rarely reveal. This is exactly why enterprise AI needs to be built differently from the start. Systems have to stay grounded in real data, operate under always-on governance, and deliver consistent results inside complex, high-stakes workflows. Read the full article: bit.ly/4pEcP2B #EnterpriseAI #AgenticAI #AIAdoption #TrustedAI #AIGovernance

We’re sharing a deeper look at how the Tool Validator Guardian Agent works and why tool-call reliability is critical for agentic systems. This blog walks through the guardrails that identify incorrect or unnecessary tool calls, and how the Tool Validator validates tool usage correctness and relevance before any API call is executed. Read the full blog here 👉 bit.ly/4pAe02Y #AgentOS #GuardianAgents #AgenticRAG #TrustedAI #EnterpriseAI

We’re excited to introduce the Tool Validator, our new Guardian Agent that reviews an agent’s workflow plan before execution. It flags erroneous or irrelevant tool calls and identifies missing or unnecessary tools, prompting the agent to recalculate its plan before execution. Extending the Agent OS with another layer of always-on governance designed to keep actions safe, grounded, and aligned with enterprise policy. Full announcement: prn.to/4pVu4vW #AgentOS #AIAgents #Governance #EnterpriseAI #TrustedAI

OODA announced OODA AI, a new intelligence tool built on vetted research and supported by Vectara’s retrieval and generation capabilities to ground its insights and deliver well-cited, contextually accurate answers. Read the full announcement here: bit.ly/4iM0Tcu #RAG #TrustedAI #EnterpriseAI #Vectara #DecisionIntelligence

Here are 5 reasons to use an Agentic Operating System instead of building AI Agents with DIY tools. vectara.com/blog/5-reasons…