
Chris Laffra
1.3K posts











🤝 We teamed up with @Sourcegraph, the creators of Cody AI, to see how long context improves their AI coding assistance. Cody's performance with Gemini 1.5 Flash's 1M context window saw improvements across key benchmarks. Explore the results → goo.gle/3AmlylP

















"Is It Time To Version Observability? (Signs Point To Yes)" by @mipsytipsy. Great write up as I was happy to attend the Charity's talk about this topic at @GOTOamst in May this year. charity.wtf/2024/08/07/is-…



Today we’re excited to introduce @llama_index workflows - a new event-driven way of building multi-agent applications. Model each agent as a component that subscribes to events and emits events; you can build complex orchestration in a way that’s readable, Pythonic, and lets you take advantage of batching, async, and streaming. Graph/Pipeline-based approach had limitations: We didn’t get here overnight. In fact we launched our Query Pipeline abstraction earlier this year which was an attempt at LLM orchestration through DAGs. But as we tried to make it better we noticed a bunch of issues: ⚠️ The orchestration got baked into the edges and made the code unreadable/cumbersome to write ⚠️ There were too many edge cases once we added loops, hard to reason about ⚠️ It was hard to debug when things went wrong. Towards Event-Driven: In an event-driven architecture, the component subscribes to events and you are responsible for writing Python code to handle these events! This gives you a fully async framework. What you can do with this: You can orchestrate anything from advanced RAG (query-rewriting, reranking, CRAG) to multi-agent systems What’s Next: Workflows is still in beta, but we’re working it hard to make it the default way of orchestrating within LlamaIndex. Another next step is to have a native integration with llama-agents - translate your multi-agents into services! Stay tuned. Blog post: llamaindex.ai/blog/introduci… Core Module Guide: docs.llamaindex.ai/en/latest/modu… RAG Guide: docs.llamaindex.ai/en/latest/exam… Agent Guide: docs.llamaindex.ai/en/latest/exam…













