Productionizing LangChain Agents with Oso
summary
Three-stage path from prototype to QA to production with concrete tools: integrates LangChain 1.0 middleware with OSO to enforce wrap tool call and wrap model call checks. Demo shows a SQL query returning a password hash flagged as untrusted content plus private data and the system blocks posting the comment, returns a clear denial reason, and quarantines misbehaving agents from the OSO monitor. Uses deterministic rules for prompt injection and an LLM-as-judge for intent-based tool filtering; the approach works in live demos.
AI-written summary · about
chapters (6)
- 0:00Introduction to agent productionization
- 1:13Prototype to QA challenges
- 10:48Demo of session-based guardrails
- 15:47QA to production intent alignment
- 22:25Production governance for agents
- 24:29Q&A on zero-day exploits
related talks
01h 59mLLM applications with LangChain
29mPrivacy-Friendly Applications with Ollama, Vector Functions, and LangChainJS
01h 50mChatGPT & LLMs: A Practical Starter's Guide to Language Chains
51mNext-Level Features of LangChain4j for Production-Ready AI Applications
39mLLM Agents Gone Wild (And How to Tame Them with Quarkus)
36mProduction Ready RAG
49mPushing LLMs over the Edge: Exploring the Limits of the Possible
25mFrom LangChain to LangSmith and LangGraph: the Future of AI