docs: refresh README first fold, reconcile stale benchmark numbers, unify the one-liner

README first fold:
- banner 800px -> 520px
- badges condensed from two paragraphs into one row (Trendshift kept, inline)
- lead with Introduction instead of benchmarks
- new How it works section explaining the add/search loop
- benchmarks moved below the intro and retitled from the date-stamped
  New Memory Algorithm (April 2026)
- dropped Research Highlights, which restated the benchmark table verbatim

Benchmarks: the README was already correct at 92.5 / 94.4, matching
mem0.ai/research and docs/core-concepts/memory-evaluation.mdx. The stale
copies were in the migration guides, which quote the numbers as a live
reason to upgrade. Updated both to 92.5 / 94.4 (+21 / +27).
docs/changelog/highlights.mdx keeps 91.6 / 93.4 inside its dated
2026-04-14 entry, which records what was announced at the time.

One-liner: standardized on 'the memory layer for AI agents', already the
canonical form in cli-spec.json, cli/python/pyproject.toml and the CLI
specification. Swept the remaining variants.

docs/docs.json

@@ -1,7 +1,7 @@
 {
   "$schema": "https://mintlify.com/docs.json",
   "name": "Mem0",
-  "description": "Mem0 is a self-improving memory layer for LLM applications, enabling personalized AI experiences that save costs and delight users.",
+  "description": "Mem0 is the memory layer for AI agents, giving them persistent, personalized context across sessions.",
   "theme": "aspen",
   "colors": {
     "primary": "#8F74E0",

docs/integrations/raycast.mdx

@@ -3,7 +3,7 @@ title: "Raycast Extension"
 description: "Mem0 Raycast extension for intelligent memory management"
 ---
 
-Mem0 is a self-improving memory layer for LLM applications, enabling personalized AI experiences that save costs and delight users. This extension lets you store and retrieve text snippets using Mem0's intelligent memory system. Find Mem0 in [Raycast Store](https://www.raycast.com/dev_khant/mem0) for using it.
+Mem0 is the memory layer for AI agents, giving them persistent, personalized context across sessions. This extension lets you store and retrieve text snippets using Mem0's intelligent memory system. Find Mem0 in [Raycast Store](https://www.raycast.com/dev_khant/mem0) for using it.
 
 ## Getting Started
 

docs/integrations/respan.mdx

@@ -7,7 +7,7 @@ Build AI applications with persistent memory and comprehensive LLM observability
 
 ## Overview
 
-Mem0 is a self-improving memory layer for LLM applications, enabling personalized AI experiences that save costs and delight users. Respan (formerly Keywords AI) provides complete LLM observability.
+Mem0 is the memory layer for AI agents, giving them persistent, personalized context across sessions. Respan (formerly Keywords AI) provides complete LLM observability.
 
 Combining Mem0 with Respan allows you to:
 1. Add persistent memory to your AI applications

docs/migration/oss-v2-to-v3.mdx

@@ -20,7 +20,7 @@ The new Mem0 release redesigns both extraction and retrieval, and cleans up the
 - **SDK cleanup**: Deprecated parameters removed, naming conventions standardized
 - **API surface aligned with Platform**: Entity IDs now follow the same convention across OSS and Platform: top-level kwargs for `add()` / `delete_all()`, inside `filters` for `search()` / `get_all()`
 
-These changes produce a **+20 point improvement on LoCoMo** (71.4 → 91.6) and **+26 point improvement on LongMemEval** (67.8 → 93.4), while cutting extraction latency roughly in half.
+These changes produce a **+21 point improvement on LoCoMo** (71.4 → 92.5) and **+27 point improvement on LongMemEval** (67.8 → 94.4), while cutting extraction latency roughly in half.
 
 ## Breaking Changes
 

docs/migration/platform-v2-to-v3.mdx

@@ -11,7 +11,7 @@ iconType: "solid"
 
 ## Overview
 
-The new Mem0 memory algorithm is a ground-up redesign of how memories are extracted, stored, and retrieved. It scores **91.6 on LoCoMo** and **93.4 on LongMemEval**: a +20 and +26 point improvement over the previous algorithm: while cutting extraction latency roughly in half.
+The new Mem0 memory algorithm is a ground-up redesign of how memories are extracted, stored, and retrieved. It scores **92.5 on LoCoMo** and **94.4 on LongMemEval**, a +21 and +27 point improvement over the previous algorithm, while cutting extraction latency roughly in half.
 
 | What Changed | Before | After |
 |---|---|---|

docs/platform/overview.mdx

@@ -4,7 +4,7 @@ description: "Managed memory layer for AI agents, production-ready in minutes"
 icon: "cloud"
 ---
 
-Mem0 Platform is the fully managed memory layer for your AI apps and agents. Your users stop repeating themselves and your agents keep context across sessions, with no vector store, reranker, or infrastructure to run.
+Mem0 Platform is the fully managed memory layer for AI agents. Your users stop repeating themselves and your agents keep context across sessions, with no vector store, reranker, or infrastructure to run.
 
 ## Why teams pick the Platform