Agentic RAG and Context Engineering for Agents

We’ve been building in this space too. The big pain we kept hitting is repeated context, messy memory, and bloated prompts.

RAG alone isn’t enough for agent workflows. You need: - memory that evolves (e.g. “my daughter is 10 → now 12”) - entity linking (“my daughter” = Emma = user_42) - context deduping + summarization under token budgets

We ended up building Flumes, an API that turns unstructured user history into structured facts + human-readable summaries. Think memory + graph + packing, all wrapped in one.

If curious: https://www.flumes.ai Would love to chat more if others are solving this differently.