Introducing Hindsight Memlake: serving 100M Memories at <100ms p99 on object storage
A new object-storage-native engine for Hindsight’s agent memory, built for fast retrieval, high-throughput ingest, and lower cost at scale.
24 stories
A new object-storage-native engine for Hindsight’s agent memory, built for fast retrieval, high-throughput ingest, and lower cost at scale.
Claude Code already knows your repo. ChatGPT already knows your preferences. With Hindsight, a Grok Bot can read those banks instead of making you explain it all again.
Tag filters on entities and the knowledge base, per-candidate reranker token caps, faster bank imports, and a round of retain, consolidation, reflect and Oracle reliability fixes
Paperclip gives agents an org chart. Hindsight's bankGranularity decides what each agent is allowed to remember about the others.
Hindsight used to ship inside Hermes. It now installs from the plugin catalog. Your settings and memory are untouched, but one thing genuinely changed: where updates come from.
Hindsight 0.10.2 lets each observation scope consolidate under its own mission. The company-wide view can record trends while a user's own scope keeps the detail, from the same facts.
Bank aliases for zero-downtime id changes, per-scope consolidation strategies, a refresh budget per mental model, multimodal dry-run extraction, streamed exports, CUDA for ONNX embeddings, and a long list of reliability fixes
We connected Meta Muse to a live Hindsight account and asked what it knew. It runs in Meta's cloud with no plugin host, so there is nothing to install: a prompt, an OAuth sign-in, and it reads the banks your other AI tools already write to.
Hindsight just crossed 40,000 GitHub stars in eleven months. The stats behind it, the 45 days it took to double, and the one number that tells us more than stars do.
Hermes moved its memory providers out of core and into the plugin catalog. Hindsight went first. Here is what changed, what you need to do about it, and where updates come from now.
Six things we learned adding Jev, a model that returns typed decisions instead of text, as a reranker in Hindsight 0.10.1. Including two designs that failed and the numbers that killed them.
When you migrate a memory history, its duplicate conclusions come with it. The fix is to import the source material and let your bank rebuild the beliefs itself.
Clone a memory bank in one call, export and import it with an explicit scope, or rename it in place. What travels, what gets rebuilt, and where it will bite you.
Hindsight Cloud now runs 0.10.0. Images and files are first-class memory, you can preview the exact prompt an operation sends, and recall is harder to miss with.
Reranking with TypeSafe Jev, move a bank anywhere with clone, export and rename, filter memories and documents by time, a separate LLM for mental-model refreshes, and many other stability fixes
Six releases in eight weeks, from 0.8.5 to 0.10.0. The biggest things Hindsight learned to do this summer, grouped by what they let you build.
You configured a mission, a disposition and three strategies. What did the model actually receive? Hindsight 0.10.0 will show you, without spending a token.
Storing an image is easy. Making a fact cite the screenshot it was read from, without inventing evidence, is the hard part. How attachments work in Hindsight 0.10.0.
Showing 18 of 24 stories
Load more storiesGet the top AI news and tools straight to your inbox every week.
Weekly · AI news
Join 15,000+ AI enthusiasts