RodCor · RodCor · Sep 12, 2026 · Sep 12, 2026
diff --git a/content/docs/memory-benchmark/beam.mdx b/content/docs/memory-benchmark/beam.mdx
@@ -71,8 +71,9 @@ of the 35 conversations (300 probes), uniform 96k budget.
 | abstention | 9/30 (30%) |
 | **overall** | **198/300 (66.0%)** |
 
-- **66.0% edges mem0's self-reported 62% at the matched bucket.** Caveats in
-  [How Kimetsu compares](/docs/memory-benchmark/comparison).
+- **Kimetsu scores 66.0%; mem0 reports 64.1% in the 1M bucket.** The readers
+  and conversation samples differ. Sources and test differences are recorded
+  in [How Kimetsu compares](/docs/memory-benchmark/comparison).
 - **Global and temporal abilities degrade with scale, as expected.** When the
   conversation is ~10x the budget, whole-arc tasks lose ground that local-fact
   tasks keep.

diff --git a/content/docs/memory-benchmark/comparison.mdx b/content/docs/memory-benchmark/comparison.mdx
@@ -22,14 +22,15 @@ bill, or the cloud.**
 | LongMemEval (`_s`) | **83.0%** (200-q slice) · ~80.9% weighted | 94.4% (their reader + harness) | not reported |
 | LoCoMo (1,540 q) | **89.4%** | 92.5% | not reported |
 | BEAM 100K | **73.3%** (400 probes) | n/a | 79% |
-| BEAM **1M** | **66.0%** (300 probes) | 62% | not reported |
+| BEAM **1M** | **66.0%** (300 probes) | 64.1% (700 probes) | not reported |
 | BEAM 10M | future work | 48.6% | 67% |
 
 Caveats, because the table is not apples-to-apples:
 
-- **Only the 1M row is a matched bucket.** Our 66.0% edges mem0's 62%, but with
-  a different reader and 15 of 35 conversations: read it as at least on par,
-  not a decisive win.
+- **The 1M row uses the same token bucket, with different setups.** Kimetsu's
+  66.0% covers 15 of 35 conversations; mem0 reports 64.1% over 700 probes.
+  The readers and samples differ, so the score difference does not establish
+  a head-to-head win.
 - **Cognee leads at 100K/10M.** Our 73.3% matches the prior public state of the
   art on 100K (the 0.735 Cognee cites as the number it beat), model-free.
   Cognee needs an LLM key on both the write and read paths.
@@ -46,6 +47,9 @@ Cognee's [BEAM figures](https://github.com/topoteretes/cognee), the
 [LongMemEval](https://arxiv.org/abs/2410.10813) and
 [BEAM](https://github.com/mohammadtavakoli78/BEAM) papers.
 
+Mem0's BEAM 1M figure was checked on September 12, 2026. Its current reported
+64.1% replaces the 62% previously quoted here.
+
 ## What we do not yet claim
 
 - Multi-hop retrieval of obliquely relevant memories is v2.6 work; the

diff --git a/content/docs/memory-benchmark/index.mdx b/content/docs/memory-benchmark/index.mdx
@@ -21,7 +21,7 @@ on v2.8.0.
 |-----------|--------|
 | LoCoMo | **89.4%** (1,540 questions, LLM-judged accuracy) |
 | BEAM 100K | **73.3%**, matching the prior public state of the art, model-free |
-| BEAM 1M | **66.0%**, ahead of mem0's self-reported 62% |
+| BEAM 1M | **66.0%** (300 probes, 15 of 35 conversations) |
 | LongMemEval (`_s`) | **83.0%** (200-question stratified slice) |
 | BrainBench quality index | **80.0%** (142 scenarios, reader-free) |
 | retrieval quality | recall@4 **0.949**, MRR **0.914** at ~138 ms |

diff --git a/src/app/(home)/_components/about-contact.tsx b/src/app/(home)/_components/about-contact.tsx
@@ -12,32 +12,25 @@ export function AboutContact() {
       >
         <div className="grid gap-8 md:grid-cols-[0.7fr_1.3fr]">
           <div>
-            <p className="km-eyebrow">About</p>
             <h2 id="about-title" className="km-heading">
-              Built for the next session.
+              About Kimetsu
             </h2>
           </div>
           <div className="max-w-2xl">
             <p className="text-lg leading-8 text-fd-muted-foreground">
-              Kimetsu is an open-source project created by{" "}
+              Created and maintained by{" "}
               <span className="font-medium text-fd-foreground">
                 Rodrigo Córdoba
               </span>
-              . It gives coding agents a local place to keep project decisions,
-              conventions, and fixes, so useful knowledge carries across
-              sessions.
-            </p>
-            <p className="mt-4 leading-7 text-fd-muted-foreground">
-              Built in Rust and backed by SQLite, Kimetsu works alongside the
-              tools you already use. The source, documentation, and benchmarks
-              are public, so you can inspect how it works and make it your own.
+              . Written in Rust and released under the MIT and Apache-2.0
+              licenses.
             </p>
             <div className="mt-6 flex flex-wrap gap-x-6 gap-y-3">
               <a href={links.github} className="km-text-link">
-                Explore the source →
+                Source code →
               </a>
-              <Link href="/docs" className="km-text-link">
-                Read the documentation →
+              <Link href="/projects" className="km-text-link">
+                Related projects →
               </Link>
             </div>
           </div>
@@ -50,13 +43,12 @@ export function AboutContact() {
       >
         <div className="grid items-center gap-8 rounded-2xl border border-fd-border bg-fd-card/40 p-6 sm:p-10 lg:grid-cols-[1fr_auto]">
           <div>
-            <p className="km-eyebrow">Contact</p>
             <h2 id="contact-title" className="km-heading">
-              Let’s talk about Kimetsu.
+              Contact
             </h2>
             <p className="mt-4 max-w-xl leading-7 text-fd-muted-foreground">
-              Have a question, an idea, or something you’d like to build
-              together? Get in touch with Rodrigo on LinkedIn.
+              For questions, feedback, or collaboration, contact Rodrigo on
+              LinkedIn.
             </p>
           </div>
           <a

diff --git a/src/app/(home)/_components/benchmark-results.tsx b/src/app/(home)/_components/benchmark-results.tsx
@@ -0,0 +1,102 @@
+import Link from "next/link";
+
+// Published runs: content/docs/memory-benchmark/{beam,index,longmemeval}.mdx.
+// Keep sample sizes beside the scores; these are separate benchmark protocols.
+export function BenchmarkResults() {
+  return (
+    <section
+      id="benchmarks"
+      aria-labelledby="benchmarks-title"
+      className="km-section km-benchmarks scroll-mt-24"
+    >
+      <div className="mb-6 flex flex-wrap items-baseline justify-between gap-3">
+        <h2 id="benchmarks-title" className="km-heading">
+          Benchmark results
+        </h2>
+        <Link href="/docs/memory-benchmark" className="km-text-link">
+          All results and methodology →
+        </Link>
+      </div>
+      <div className="grid overflow-hidden rounded-2xl border border-fd-border lg:grid-cols-3">
+        <article className="flex flex-col bg-fd-primary/5 p-6 sm:p-8">
+          <h3 className="font-mono text-sm font-semibold">BEAM 100K</h3>
+          <p className="km-benchmark-score text-fd-primary">73.3%</p>
+          <p className="text-sm text-fd-muted-foreground">
+            Accuracy · 400 probes · 20 conversations
+          </p>
+          <div
+            className="mt-6 space-y-3"
+            role="img"
+            aria-label="BEAM 100K accuracy on the same 400 probes: Kimetsu with graph retrieval, 73.3 percent; flat retrieval baseline, 62.3 percent. Bars use a zero to 100 percent scale."
+          >
+            <div aria-hidden>
+              <p className="mb-2 text-xs font-medium">
+                Kimetsu · graph retrieval
+              </p>
+              <div className="h-2 rounded-full bg-fd-primary/10">
+                <div className="h-full w-[73.3%] rounded-full bg-fd-primary" />
+              </div>
+            </div>
+            <div aria-hidden>
+              <div className="mb-2 flex justify-between gap-3 text-xs text-fd-muted-foreground">
+                <span>Flat retrieval baseline</span>
+                <span className="font-mono tabular-nums">62.3%</span>
+              </div>
+              <div className="h-2 rounded-full bg-fd-primary/10">
+                <div className="h-full w-[62.3%] rounded-full bg-fd-muted-foreground" />
+              </div>
+            </div>
+          </div>
+          <p className="mt-4 text-xs leading-5 text-fd-muted-foreground">
+            Both runs use the same test set.
+          </p>
+          <Link
+            href="/docs/memory-benchmark/beam"
+            className="km-text-link mt-auto pt-6"
+          >
+            BEAM test settings →
+          </Link>
+        </article>
+        <article className="flex flex-col border-t border-fd-border p-6 sm:p-8 lg:border-t-0 lg:border-l">
+          <h3 className="font-mono text-sm font-semibold">LoCoMo</h3>
+          <p className="km-benchmark-score">89.4%</p>
+          <p className="text-sm text-fd-muted-foreground">
+            Accuracy · 1,540 questions
+          </p>
+          <p className="mt-6 text-sm leading-6 text-fd-muted-foreground">
+            Questions about facts, events, and relationships across long
+            conversations.
+          </p>
+          <Link
+            href="/docs/memory-benchmark/comparison"
+            className="km-text-link mt-auto pt-6"
+          >
+            LoCoMo results and comparison →
+          </Link>
+        </article>
+        <article className="flex flex-col border-t border-fd-border p-6 sm:p-8 lg:border-t-0 lg:border-l">
+          <h3 className="font-mono text-sm font-semibold">LongMemEval S</h3>
+          <p className="km-benchmark-score">83.0%</p>
+          <p className="text-sm text-fd-muted-foreground">
+            Accuracy · 200-question stratified slice
+          </p>
+          <p className="mt-6 text-sm leading-6 text-fd-muted-foreground">
+            Sampled from the 500-question set, covering recall, changed facts,
+            and reasoning across sessions.
+          </p>
+          <Link
+            href="/docs/memory-benchmark/longmemeval"
+            className="km-text-link mt-auto pt-6"
+          >
+            LongMemEval test settings →
+          </Link>
+        </article>
+      </div>
+      <p className="mt-4 max-w-4xl text-xs leading-5 text-fd-muted-foreground">
+        Historical results from the linked runs. An LLM answers and grades the
+        questions; the memory pipeline uses local search, embeddings, and
+        reranking, with no LLM calls.
+      </p>
+    </section>
+  );
+}
diff --git a/src/app/(home)/_components/explore-section.tsx b/src/app/(home)/_components/explore-section.tsx
@@ -1,21 +1,17 @@
-import { Database, MessageSquare, Search } from "lucide-react";
 import Link from "next/link";
 
 const steps = [
   {
-    icon: Database,
-    title: "Keep the useful lesson",
-    body: "Save a project convention, a decision, or the command that finally worked. Memories live in your project’s SQLite database.",
+    title: "Store",
+    body: "Memories live in a SQLite database in your project. No hosted memory account is required.",
   },
   {
-    icon: Search,
-    title: "Recall it in context",
-    body: "Search by words or meaning. Kimetsu selects relevant memories within a delivery budget and brings them into the next task.",
+    title: "Retrieve",
+    body: "Start with keyword search. Add local embeddings and reranking for semantic search. Set a token budget for the context sent to your agent.",
   },
   {
-    icon: MessageSquare,
-    title: "Learn from the outcome",
-    body: "Citations and feedback help track which memories contributed. Corrections and lifecycle rules keep old advice from outliving its usefulness.",
+    title: "Update",
+    body: "Track changed facts and suppress outdated results. Export and import memories to move them between machines.",
   },
 ];
 
@@ -27,48 +23,29 @@ export function ExploreSection() {
       className="km-section scroll-mt-24"
     >
       <div className="mb-10 flex flex-wrap items-end justify-between gap-5">
-        <div className="max-w-2xl">
-          <p className="km-eyebrow">Across sessions, across agents</p>
-          <h2 id="workflow-title" className="km-heading">
-            The useful part of yesterday, ready today.
-          </h2>
-        </div>
+        <h2 id="workflow-title" className="km-heading">
+          How it works
+        </h2>
         <Link href="/docs/how-kimetsu-works" className="km-text-link">
-          Inside the memory loop →
+          Technical documentation →
         </Link>
       </div>
       <ol className="grid gap-6 md:grid-cols-3">
         {steps.map((step, index) => (
           <li key={step.title} className="border-t border-fd-border pt-6">
-            <div className="mb-5 flex items-center justify-between">
-              <step.icon className="size-6 text-fd-primary" aria-hidden />
-              <span className="font-mono text-sm text-fd-muted-foreground">
-                0{index + 1}
-              </span>
-            </div>
+            <span
+              className="mb-4 block font-mono text-sm text-fd-primary"
+              aria-hidden
+            >
+              0{index + 1}
+            </span>
             <h3 className="mb-3 text-lg font-semibold">{step.title}</h3>
             <p className="text-sm leading-6 text-fd-muted-foreground">
               {step.body}
             </p>
           </li>
         ))}
       </ol>
-      <div className="mt-10 flex flex-col justify-between gap-4 rounded-xl border border-fd-border bg-fd-card p-5 sm:flex-row sm:items-center">
-        <p className="max-w-2xl text-sm leading-6 text-fd-muted-foreground">
-          <strong className="text-fd-foreground">
-            Choose your compute budget.
-          </strong>{" "}
-          Start with lexical search. Add local embeddings and reranking when
-          your workload benefits from them. Delivered context still uses your
-          agent’s tokens.
-        </p>
-        <Link
-          href="/docs/how-kimetsu-works/retrieval-models"
-          className="km-text-link shrink-0"
-        >
-          Compare retrieval models →
-        </Link>
-      </div>
     </section>
   );
 }
diff --git a/src/app/(home)/_components/quick-start.tsx b/src/app/(home)/_components/quick-start.tsx
@@ -35,13 +35,12 @@ export function QuickStart() {
     >
       <div className="grid gap-8 lg:grid-cols-[0.85fr_1.15fr] lg:items-center">
         <div>
-          <p className="km-eyebrow">Start with your agent</p>
           <h2 id="setup-title" className="km-heading">
-            Two commands. A memory that stays.
+            Install Kimetsu
           </h2>
           <p className="mt-4 max-w-md text-fd-muted-foreground">
-            Install the binary, then run setup from your project directory.
-            Choose the agent you already use.
+            Select your agent and run these commands from your project
+            directory. Available for Linux, macOS, and Windows.
           </p>
           <Link href="/docs/install" className="km-text-link mt-4 inline-block">
             All installation options →
@@ -96,7 +95,7 @@ export function QuickStart() {
                 ? feedback.ok
                   ? "Commands copied to clipboard."
                   : "Copy is unavailable. Select and copy the commands above."
-                : "Setup wires this project to your selected agent."}
+                : ""}
             </p>
           </div>
           <noscript>