diff --git a/packages/opencode/src/altimate/workspace/memory-api.ts b/packages/opencode/src/altimate/workspace/memory-api.ts
index 1729f613c..5ce0fce40 100644
--- a/packages/opencode/src/altimate/workspace/memory-api.ts
+++ b/packages/opencode/src/altimate/workspace/memory-api.ts
@@ -52,6 +52,10 @@ export interface MirrorMetadata {
   visibility: "private"
   block_created: string
   block_updated: string
+  /** Heading shown for the record in the workspace UI. The create's extractor
+   * writes its own, but the repair ``update`` replaces the metadata dict
+   * wholesale, so a synced block has no heading unless we supply one. */
+  title?: string
   /** Absent for global blocks — that is what makes them span workspaces. */
   datamate_id?: string
   datamate_name?: string
diff --git a/packages/opencode/src/altimate/workspace/memory-sync.ts b/packages/opencode/src/altimate/workspace/memory-sync.ts
index 8e7b9c0b3..d27f24d5f 100644
--- a/packages/opencode/src/altimate/workspace/memory-sync.ts
+++ b/packages/opencode/src/altimate/workspace/memory-sync.ts
@@ -303,6 +303,41 @@ export function decodeTags(raw: unknown): string[] {
     .filter(Boolean)
 }
 
+/** Longest heading we will mirror. A block may be up to MEMORY_MAX_BLOCK_SIZE,
+ * and its first line can be most of that. */
+const TITLE_MAX = 120
+
+/** Heading for a synced block.
+ *
+ * A create runs the extractor, which writes its own `title`, but the repair
+ * `update` replaces the metadata dict wholesale — so without this the record
+ * reaches the workspace with no heading at all. Blocks conventionally
+ * open with a markdown heading; the block id is a readable fallback for those
+ * that do not.
+ */
+export function blockTitle(block: MemoryBlock): string {
+  // A training block opens with its metadata comment; the heading is after it.
+  for (const line of stripTrainingMeta(block.content).split("\n")) {
+    // CommonMark: up to three leading spaces, and a closing run of hashes that
+    // is decoration rather than title text. The closing run must be preceded by
+    // whitespace, so a heading like `# C#` keeps its hash. (stripTrainingMeta
+    // trims, so deeper indentation on the first line is gone before we look —
+    // a usable title beats falling back to the id over leading whitespace.)
+    const heading = line.match(/^ {0,3}#{1,6}\s+(\S.*?)(?:\s+#+)?\s*$/)
+    if (heading) {
+      const text = heading[1]
+      // Count code points, not UTF-16 units: slicing mid-surrogate leaves a
+      // lone half that renders as a replacement character in the workspace.
+      const points = Array.from(text)
+      if (points.length <= TITLE_MAX) return text
+      return `${points.slice(0, TITLE_MAX - 1).join("")}\u2026`
+    }
+    // Content that opens with body text has no heading to borrow.
+    if (line.trim()) break
+  }
+  return block.id
+}
+
 export function buildMetadata(block: MemoryBlock, binding: CachedBinding | null): MirrorMetadata {
   const meta: MirrorMetadata = {
     source: MIRROR_SOURCE,
@@ -311,6 +346,7 @@ export function buildMetadata(block: MemoryBlock, binding: CachedBinding | null)
     visibility: "private",
     block_created: block.created,
     block_updated: block.updated,
+    title: blockTitle(block),
   }
   // JSON, not a comma join: a tag containing a comma split into two on read.
   if (block.tags.length > 0) meta.block_tags = JSON.stringify(block.tags)
diff --git a/packages/opencode/test/altimate/workspace/memory-sync.test.ts b/packages/opencode/test/altimate/workspace/memory-sync.test.ts
index a81510ae1..be9ee7afe 100644
--- a/packages/opencode/test/altimate/workspace/memory-sync.test.ts
+++ b/packages/opencode/test/altimate/workspace/memory-sync.test.ts
@@ -40,6 +40,7 @@ const {
   backfill,
   belongsHere,
   buildMetadata,
+  blockTitle,
   hydrate,
   isEnabled,
   memoryEnabledCached,
@@ -332,6 +333,14 @@ describe("buildMetadata", () => {
     expect(buildMetadata(block(), null).visibility).toBe("private")
   })
 
+  test("a heading is always supplied, so the record is not title-less", () => {
+    // The create's extractor writes a title, but the repair update() replaces
+    // the metadata dict wholesale — without one here the record reaches the
+    // workspace with no heading.
+    const meta = buildMetadata(block({ content: "# Staging Model Convention\n\n- prefix stg_" }), null)
+    expect(meta.title).toBe("Staging Model Convention")
+  })
+
   test("created and updated timestamps are carried", () => {
     const meta = buildMetadata(block({ created: NOW, updated: NOW }), null)
     expect(meta.block_created).toBe(NOW)
@@ -361,6 +370,66 @@ function gatedFetch() {
   }
 }
 
+describe("blockTitle", () => {
+  test("uses the block's leading markdown heading", () => {
+    expect(blockTitle(block({ content: "# Warehouse access\n\nbody" }))).toBe("Warehouse access")
+  })
+
+  test("accepts any heading level and trims surrounding space", () => {
+    expect(blockTitle(block({ content: "###   Deploy steps   \nbody" }))).toBe("Deploy steps")
+  })
+
+  test("skips blank lines before the heading", () => {
+    expect(blockTitle(block({ content: "\n\n## Naming rules\nbody" }))).toBe("Naming rules")
+  })
+
+  test("falls back to the block id when the content opens with body text", () => {
+    // Borrowing a body line would produce a heading the user never wrote.
+    const b = block({ id: "warehouse/snowflake", content: "Snowflake account is acme-prod.\n\n# Later heading" })
+    expect(blockTitle(b)).toBe("warehouse/snowflake")
+  })
+
+  test("falls back to the block id for a hash with no heading text", () => {
+    expect(blockTitle(block({ id: "a/b", content: "#hashtag not a heading" }))).toBe("a/b")
+  })
+
+  test("reads past a training block's metadata comment to its heading", () => {
+    // Training blocks always open with this comment, so scanning raw content
+    // sees it as body text and falls back to the id.
+    const content = "<!-- training\nkind: rule\napplied: 3\n-->\n# Naming rules\n\nbody"
+    expect(blockTitle(block({ id: "t/1", content }))).toBe("Naming rules")
+  })
+
+  test("allows the three leading spaces CommonMark permits", () => {
+    expect(blockTitle(block({ content: "   # Indented heading\nbody" }))).toBe("Indented heading")
+  })
+
+  test("drops a closing run of hashes", () => {
+    expect(blockTitle(block({ content: "## Release notes ##\nbody" }))).toBe("Release notes")
+  })
+
+  test("keeps a hash that is part of the heading text", () => {
+    // The closing run must be preceded by whitespace, so this is not one.
+    expect(blockTitle(block({ content: "# Style guide for C#\nbody" }))).toBe("Style guide for C#")
+  })
+
+  test("truncating never splits an emoji into a lone surrogate", () => {
+    // slice() counts UTF-16 units; cutting mid-pair renders as U+FFFD.
+    const content = `# ${"x".repeat(118)}\u{1F600}${"y".repeat(10)}`
+    const title = blockTitle(block({ content }))
+    expect(Array.from(title).length).toBe(120)
+    expect(title).toContain("\u{1F600}")
+    expect(title.isWellFormed()).toBe(true)
+  })
+
+  test("truncates a heading longer than the cap", () => {
+    const long = "x".repeat(200)
+    const title = blockTitle(block({ content: `# ${long}` }))
+    expect(title.length).toBe(120)
+    expect(title.endsWith("\u2026")).toBe(true)
+  })
+})
+
 // ── write path ──────────────────────────────────────────────────────────────
 describe("mirrorBlock", () => {
   test("flushPendingMirrors waits for a mirror a short-lived process would abandon (#1332)", async () => {