Memory Categorization Webhooks — Get notified via webhook when memories are categorized, enabling real-time downstream workflows

Temporal Search Pre-filtering — Time-aware filtering in memory search so queries like "what happened last week" work out of the box

Graph Status in Events API — Graph processing status now exposed directly in the event API response

V4 memory improvements rolled out to 100% of users

LLM abstraction refactored to a factory/strategy pattern for cleaner multi-provider support

Improved reranking recall in semantic search for more relevant results

RRF relevance scores capped at 0.9 to set clearer user expectations on match quality

Revamped settings page layout with new UI tokens

New invite flow for organization and project member management

Standardized tooltip styling with sidebar tooltips

Token budget enforcement for oversized memory inputs

Fixed accidental delete-all when search filters don't match any memories

Fixed GitHub login failing when the user email isn't present

Fixed summary trigger using boundary crossing detection instead of an exact modulo match

Custom instructions are now correctly passed to the merge step across all memory versions

Resolved login flow issues and invite flow bugs

Fixed pagination, drawer backgrounds, and various dark theme inconsistencies

Memory Engine Changes — Completely revamped memory extraction and merge prompts, introducing Reciprocal Rank Fusion (RRF) hybrid search that combines semantic and keyword-based retrieval for significantly better recall

Billing & Usage Dashboard — Fully rebuilt billing experience with Stripe integration, usage breakdowns by org owner, and a modernized subscription management flow

Async Bulk Deletion — Large-scale memory deletion now runs asynchronously via SQS and auto-scaling EKS workers, making bulk operations reliable at any scale

Graph Memory Status Tracking — You can now track the processing status of graph memory operations in real time

Webhook & Export Lambda Deployment — New deployment pipelines for the webhook and memory export services

Redis-Cached Search Embeddings — Search embedding generation is now cached in Redis, making repeated and similar queries significantly faster

Memory Exports Page Overhaul — Redesigned the exports experience with a modern export builder interface

Rich contextual memory creation prompts for higher-quality extracted memories

Custom instructions now flow through all stages of memory extraction and merge

Simplified entity API queries — faster responses when date filters aren't used

State management migrated from React Context to Redux for better performance

Unified design token system with new typography, color palette, and input components

Redesigned organization and project creation flow

Revamped onboarding experience with API key page, quick-copy, and category selection

Graph maintenance workflow and cleanup automation

Dashboard cache TTL reduced from 3 hours to 1 minute for near-real-time data

Fixed events getting stuck in "running" state indefinitely

Resolved connection pool timeouts during embedding resolution

Prevented implicit attribute inference in memory prompts, causing hallucinated metadata

Addressed multiple frontend theme and color consistency issues across the platform

Fixed org and project creation flow edge cases