diff --git a/Makefile b/Makefile
index 27e1770..9474637 100644
--- a/Makefile
+++ b/Makefile
@@ -25,13 +25,13 @@ MRUBY_CFLAGS ?= -I./libs/mruby/build/host/include
 MRUBY_LIBS ?= -L./libs/mruby/build/host/lib -lmruby
 
 CFLAGS_DEBUG :=\
-	-std=c++20 -Wall -Wextra -fno-rtti \
+	-std=c++23 -Wall -Wextra -fno-rtti \
 	-Og -fno-inline -gsplit-dwarf \
 	-g -fno-omit-frame-pointer -ffast-math \
 	-I./include -I./libs/unicode_width
 
 CFLAGS_RELEASE :=\
-	-std=c++20 -O3 -mtune=generic -fno-rtti \
+	-std=c++23 -O3 -mtune=generic -fno-rtti \
 	-fvisibility=hidden -static -ffast-math \
 	-fomit-frame-pointer -DNDEBUG -s \
 	-I./include -I./libs/unicode_width
diff --git a/include/io/file.h b/include/io/file.h
index eb30a4e..6724f98 100644
--- a/include/io/file.h
+++ b/include/io/file.h
@@ -29,12 +29,10 @@ inline int read_file(const char *path, char **text, uint32_t *len) {
 
   if (read != *len) {
     free(*text);
-    *text = NULL;
+    *text = nullptr;
     *len = 0;
     return 0;
   }
 
-  (*text)[*len] = '\0'; // handy if treating it as a C string
-
   return 1;
 }
diff --git a/include/utils/utils.h b/include/utils/utils.h
index 6271b7d..b0c558d 100644
--- a/include/utils/utils.h
+++ b/include/utils/utils.h
@@ -1,8 +1,3 @@
 #pragma once
 
 #include "pch.h"
-
-struct Point {
-  uint32_t row;
-  uint32_t col;
-};
diff --git a/include/vase/buffer/append.h b/include/vase/buffer/append.h
index 30bb2b4..c37c233 100644
--- a/include/vase/buffer/append.h
+++ b/include/vase/buffer/append.h
@@ -1,12 +1,11 @@
 #pragma once
 
+#include "../constants.h"
 #include "buffer.h"
 #include "pch.h"
 
 struct AppendBuffer : Buffer {
-  static constexpr uint32_t CHUNK_SIZE = 1 << 16;
-
-  using TChunk = char[CHUNK_SIZE];
+  using TChunk = char[APPEND_CHUNK_SIZE];
   std::vector<TChunk *> buf;
   TChunk *t_current;
   uint32_t current_offset{0};
@@ -21,6 +20,4 @@ struct AppendBuffer : Buffer {
 
   const char *read(uint32_t pos, uint32_t *out_len);
   inline uint32_t length();
-
-  inline void new_text_chunk();
 };
diff --git a/include/vase/buffer/original.h b/include/vase/buffer/original.h
index ca164ed..cd56d32 100644
--- a/include/vase/buffer/original.h
+++ b/include/vase/buffer/original.h
@@ -7,8 +7,8 @@ struct OriginalBuffer : Buffer {
   char *buf;
   uint32_t len;
 
-  // TODO: replace with file io or even lazy loaded file io for large files/logs etc. later.
-  OriginalBuffer(char *buf, uint32_t len);
+  // Add 2 modes here, normal and lazy, lazy copies the file to a safe location and uses mmap.
+  OriginalBuffer(std::string path);
 
   ~OriginalBuffer();
 
diff --git a/include/vase/constants.h b/include/vase/constants.h
new file mode 100644
index 0000000..c4de939
--- /dev/null
+++ b/include/vase/constants.h
@@ -0,0 +1,11 @@
+#pragma once
+
+#include "pch.h"
+
+constexpr uint32_t APPEND_CHUNK_SIZE = 64 * 1024;
+constexpr uint32_t PETAL_SIZE_MAX = 32 * 1024;
+
+static_assert(
+  APPEND_CHUNK_SIZE > PETAL_SIZE_MAX,
+  "PETAL_SIZE_MAX must be smaller than APPEND_CHUNK_SIZE"
+);
diff --git a/include/vase/iterators/chunk.h b/include/vase/iterators/chunk.h
index 8d33b75..41d999f 100644
--- a/include/vase/iterators/chunk.h
+++ b/include/vase/iterators/chunk.h
@@ -3,14 +3,20 @@
 #include "../shard.h"
 #include "pch.h"
 
+enum struct Direction : uint8_t {
+  Forward,
+  Backward
+};
+
 struct ChunkIterator {
+  Direction dir;
   Shard *root = nullptr;
   std::vector<Shard *> stack;
   Petal *petal = nullptr;
   uint32_t petal_offset = 0;
-  uint32_t global_offset;
+  uint32_t global_offset = 0;
 
-  ChunkIterator(Shard *r);
+  ChunkIterator(Shard *r, Direction dir);
 
   ~ChunkIterator();
 
diff --git a/include/vase/iterators/line.h b/include/vase/iterators/line.h
index ab49310..e2a68ed 100644
--- a/include/vase/iterators/line.h
+++ b/include/vase/iterators/line.h
@@ -6,14 +6,12 @@
 
 struct LineIterator {
   ChunkIterator it;
+  Direction dir;
 
-  const char *cursor = nullptr;
-  bool eof = false;
-  uint32_t remaining = 0;
-  uint32_t line_offset = 0;
-
-  LineIterator(Shard *r, uint32_t start_line);
+  const char *chunk;
+  uint32_t len;
 
+  LineIterator(Shard *r, uint32_t start_line, Direction dir);
   ~LineIterator() = default;
 
   bool next(std::string &line);
diff --git a/include/vase/shard.h b/include/vase/shard.h
index 3f12427..8b2c467 100644
--- a/include/vase/shard.h
+++ b/include/vase/shard.h
@@ -2,6 +2,7 @@
 
 #include "buffer/buffer.h"
 #include "buffer/original.h"
+#include "constants.h"
 #include "pch.h"
 #include "utils/utils.h"
 
diff --git a/include/vase/vase.h b/include/vase/vase.h
index 41fe041..d366863 100644
--- a/include/vase/vase.h
+++ b/include/vase/vase.h
@@ -2,14 +2,29 @@
 
 #include "buffer/append.h"
 #include "buffer/original.h"
+#include "constants.h"
 #include "iterators/line.h"
 #include "pch.h"
 #include "search.h"
 #include "shard.h"
 #include "utils/utils.h"
 
+struct Point {
+  uint32_t row;
+  uint32_t col;
+};
+
+struct Range {
+  Point start;
+  Point end;
+};
+
 struct Vase {
-  Vase(char *data, uint32_t length);
+  OriginalBuffer original;
+  AppendBuffer append;
+  Shard *root;
+
+  Vase(std::string path);
 
   ~Vase();
 
@@ -20,26 +35,33 @@ struct Vase {
 
   LineIterator iterate(uint32_t line);
 
-  void input(Point *point, char key);
+  void insert(Point *point, char key);
   void insert(Point *point, const char *data, uint32_t len);
   void erase(Point *point, int64_t amount);
-  bool jump(Point *point, int64_t amount);
-  bool clamp(Point *point);
+  void erase(Range range);
+  void replace(Range range, const char *data, uint32_t len);
+
+  void move_clusters(Point *point, int64_t amount);
+  void move_lines(Point *point, int64_t amount);
+  void clamp(Point *point);
+
   void regex_search_replace(
     std::string_view pattern,
     Point start, Point end,
     std::string_view replace, std::string_view options
   );
 
-private:
-  OriginalBuffer original;
-  AppendBuffer append;
-  Shard *root;
+  bool undo();
+  bool redo();
 
-  /*std::vector<Shard *> undo; // TODO: later
-  uint16_t top; // of the undo stack.
-  uint16_t max; // for redo when no edits have been done after some undo.*/
+  void snapshot();
+  void prune_history(uint32_t n);
+
+private:
+  std::vector<Shard *> history;
+  uint32_t history_top;
 
   uint32_t offset_of(Point point);
   static void flatten(Shard *s, std::string &out);
+  void _insert(Point *point, const char *data, uint32_t len);
 };
diff --git a/src/main.cc b/src/main.cc
index 8559376..d9d85a3 100644
--- a/src/main.cc
+++ b/src/main.cc
@@ -1,39 +1,17 @@
-#include "io/file.h"
 #include "pch.h"
 #include "vase/iterators/chunk.h"
 #include "vase/iterators/line.h"
-#include "vase/shard.h"
 #include "vase/vase.h"
 
 int main(int argc, char *argv[]) {
-
-#ifdef DEBUG
-
-  const char *text_o =
-    "abcdefghijklmnopqrstuvwxyzabcdefghijklmnopqrstuvwxyz1\n"
-    "abcdefghijklmnopqrstuvwxyzabcdefghijklmnopqrstuvwxyz2\n"
-    "abcdefghijklmnopqrstuvwxyzabcdefghijklmnopqrstuvwxyz3\n"
-    "abcdefghijklmnopqrstuvwxyzabcdefghijklmnopqrstuvwxyz4";
-
-  uint32_t len = strlen(text_o);
-  char *text = strdup(text_o);
-
-#else
-
-  const char *path;
-
-  if (argc >= 2)
-    path = argv[1];
-  else
+  if (argc < 2)
     throw std::runtime_error("Please give filename.");
 
-  uint32_t len;
-  char *text;
-  read_file(path, &text, &len);
+  Vase vase = Vase(std::string(argv[1]));
 
-#endif
-
-  Vase vase = Vase(text, len);
+  std::cout << "\n"
+            << vase.to_string()
+            << "\n";
 
   return 0;
 }
diff --git a/src/vase/buffer/append.cc b/src/vase/buffer/append.cc
index c6ccac4..e798ce1 100644
--- a/src/vase/buffer/append.cc
+++ b/src/vase/buffer/append.cc
@@ -1,7 +1,8 @@
 #include "vase/buffer/append.h"
 
 AppendBuffer::AppendBuffer() {
-  new_text_chunk();
+  t_current = (TChunk *)malloc(sizeof(TChunk));
+  buf.push_back(t_current);
 }
 
 AppendBuffer::~AppendBuffer() {
@@ -10,8 +11,11 @@ AppendBuffer::~AppendBuffer() {
 }
 
 uint32_t AppendBuffer::key(char c) {
-  if (t_offset == CHUNK_SIZE)
-    new_text_chunk();
+  if (t_offset == APPEND_CHUNK_SIZE) {
+    t_current = (TChunk *)malloc(sizeof(TChunk));
+    buf.push_back(t_current);
+    t_offset = 0;
+  }
   (*t_current)[t_offset++] = c;
   return current_offset++;
 }
@@ -19,9 +23,12 @@ uint32_t AppendBuffer::key(char c) {
 uint32_t AppendBuffer::append(const char *text, uint32_t length) {
   uint32_t start = current_offset;
   while (length > 0) {
-    if (t_offset == CHUNK_SIZE)
-      new_text_chunk();
-    uint32_t copy = std::min(length, CHUNK_SIZE - t_offset);
+    if (t_offset == APPEND_CHUNK_SIZE) {
+      t_current = (TChunk *)malloc(sizeof(TChunk));
+      buf.push_back(t_current);
+      t_offset = 0;
+    }
+    uint32_t copy = std::min(length, APPEND_CHUNK_SIZE - t_offset);
     memcpy((*t_current) + t_offset, text, copy);
     t_offset += copy;
     current_offset += copy;
@@ -34,21 +41,15 @@ uint32_t AppendBuffer::append(const char *text, uint32_t length) {
 const char *AppendBuffer::read(uint32_t pos, uint32_t *out_len) {
   if (pos >= current_offset)
     return nullptr;
-  uint32_t local_offset = pos % CHUNK_SIZE;
+  uint32_t local_offset = pos % APPEND_CHUNK_SIZE;
   if (out_len) {
     uint32_t remaining = current_offset - pos;
-    uint32_t until_chunk_end = CHUNK_SIZE - local_offset;
+    uint32_t until_chunk_end = APPEND_CHUNK_SIZE - local_offset;
     *out_len = std::min(remaining, until_chunk_end);
   }
-  return &(*buf[pos / CHUNK_SIZE])[local_offset];
+  return &(*buf[pos / APPEND_CHUNK_SIZE])[local_offset];
 }
 
 inline uint32_t AppendBuffer::length() {
   return current_offset;
 }
-
-inline void AppendBuffer::new_text_chunk() {
-  t_current = (TChunk *)malloc(sizeof(TChunk));
-  buf.push_back(t_current);
-  t_offset = 0;
-}
diff --git a/src/vase/buffer/original.cc b/src/vase/buffer/original.cc
index 9d49a29..716c6a2 100644
--- a/src/vase/buffer/original.cc
+++ b/src/vase/buffer/original.cc
@@ -1,6 +1,9 @@
 #include "vase/buffer/original.h"
+#include "io/file.h"
 
-OriginalBuffer::OriginalBuffer(char *buf, uint32_t len) : buf(buf), len(len) {}
+OriginalBuffer::OriginalBuffer(std::string path) {
+  read_file(path.c_str(), &buf, &len);
+}
 
 OriginalBuffer::~OriginalBuffer() {
   free(buf);
diff --git a/src/vase/iterators/chunk.cc b/src/vase/iterators/chunk.cc
index 3c16bb0..9db32e4 100644
--- a/src/vase/iterators/chunk.cc
+++ b/src/vase/iterators/chunk.cc
@@ -1,6 +1,7 @@
 #include "vase/iterators/chunk.h"
 
-ChunkIterator::ChunkIterator(Shard *r) : root(r) {
+ChunkIterator::ChunkIterator(Shard *r, Direction dir)
+    : dir(dir), root(r) {
   if (!root)
     return;
   Shard::retain(root);
@@ -14,119 +15,171 @@ void ChunkIterator::seek_offset(uint32_t offset) {
   stack.clear();
   petal = nullptr;
   petal_offset = 0;
-  global_offset = 0;
-  if (!root || offset >= root->length)
+  if (!root)
     return;
+  if (offset >= root->length)
+    offset = root->length - 1;
   uint32_t target = offset;
   Shard *curr = root;
   while (curr) {
     if (curr->kind == Shard::ShardKind::Petal) {
       petal = (Petal *)curr;
       petal_offset = target;
+      global_offset += target;
       return;
     } else {
       auto *b = (Branch *)curr;
       uint32_t left_len = b->left->length;
       if (target < left_len) {
-        stack.push_back(b->right);
+        if (dir == Direction::Forward)
+          stack.push_back(b->right);
         curr = b->left;
       } else {
         target -= left_len;
-        global_offset += b->left->length;
+        if (dir == Direction::Backward)
+          stack.push_back(b->left);
+        global_offset += left_len;
         curr = b->right;
       }
     }
   }
+  std::unreachable();
 }
 
 void ChunkIterator::seek_line(uint32_t line) {
   stack.clear();
   petal = nullptr;
   petal_offset = 0;
-  global_offset = 0;
-  if (!root || line > root->lines)
+  bool last_line = false;
+  if (!root)
     return;
+  if (line > root->lines + 1)
+    line = root->lines + 1;
+  if (line == root->lines + 1)
+    last_line = true;
   Shard *curr = root;
   while (curr) {
     if (curr->kind == Shard::ShardKind::Petal) {
       petal = (Petal *)curr;
-      const char *text = nullptr;
-      uint32_t remaining = 0;
-      uint32_t offset = 0;
-      while (line) {
-        if (!text) {
-          uint32_t got = 0;
-          text = petal->source->read(petal->pos + offset, &got);
-          remaining = std::min(got, petal->length - offset);
+      if (last_line) {
+        petal_offset = petal->length;
+        global_offset += petal->length;
+      } else {
+        const char *text = nullptr;
+        uint32_t remaining = 0;
+        uint32_t offset = 0;
+        while (line) {
+          if (!text) {
+            uint32_t got = 0;
+            text = petal->source->read(petal->pos + offset, &got);
+            remaining = std::min(got, petal->length - offset);
+          }
+          const char *c = (const char *)memchr(text, '\n', remaining);
+          if (!c) {
+            offset += remaining;
+            text = nullptr;
+            continue;
+          }
+          remaining -= (c - text) + 1;
+          offset += (c - text) + 1;
+          text = c + 1;
+          line--;
         }
-        const char *c = (const char *)memchr(text, '\n', remaining);
-        if (!c) {
-          offset += remaining;
-          text = nullptr;
-          continue;
-        }
-        remaining -= (c - text) + 1;
-        offset += (c - text) + 1;
-        text = c + 1;
-        line--;
+        if (dir == Direction::Backward)
+          --offset;
+        petal_offset = offset;
+        global_offset += offset;
       }
-      petal_offset = offset;
       return;
     } else {
       auto *b = (Branch *)curr;
       uint32_t left_lines = b->left->lines;
       if (line <= left_lines) {
-        stack.push_back(b->right);
+        if (dir == Direction::Forward)
+          stack.push_back(b->right);
         curr = b->left;
       } else {
         line -= left_lines;
+        if (dir == Direction::Backward)
+          stack.push_back(b->left);
         global_offset += b->left->length;
         curr = b->right;
       }
     }
   }
+  std::unreachable();
 }
 
 uint32_t ChunkIterator::byte_offset() {
-  return global_offset + petal_offset;
+  return global_offset;
 }
 
 bool ChunkIterator::next(const char **data, uint32_t *out_len) {
-  while (true) {
-    if (petal) {
-      if (petal_offset < petal->length) {
-        uint32_t remaining = petal->length - petal_offset;
-        uint32_t read_pos = petal->pos + petal_offset;
-        uint32_t got = 0;
-        const char *chunk = petal->source->read(read_pos, &got);
-        if (got == 0) {
-          petal_offset = petal->length;
+  if (dir == Direction::Forward) {
+    while (true) {
+      if (petal) {
+        if (petal_offset == petal->length) {
           petal = nullptr;
           continue;
         }
+        uint32_t remaining = petal->length - petal_offset;
+        uint32_t got = 0;
+        const char *chunk = petal->source->read(petal->pos + petal_offset, &got);
+        if (got == 0) {
+          petal = nullptr;
+          petal_offset = 0;
+          return false;
+        }
         uint32_t take = std::min(got, remaining);
         *data = chunk;
         *out_len = take;
         petal_offset += take;
         return true;
       }
-      global_offset += petal->length;
-      petal = nullptr;
-      petal_offset = 0;
-    }
-    if (stack.empty())
-      return false;
-    Shard *s = stack.back();
-    stack.pop_back();
-    if (!s)
-      continue;
-    if (s->kind == Shard::ShardKind::Petal) {
+      if (stack.empty())
+        return false;
+      auto s = stack.back();
+      stack.pop_back();
+      while (s->kind == Shard::ShardKind::Branch) {
+        Branch *b = (Branch *)s;
+        stack.push_back(b->right);
+        s = b->left;
+      }
       petal = (Petal *)s;
       petal_offset = 0;
-    } else {
-      auto *b = (Branch *)s;
-      stack.push_back(b->right);
-      stack.push_back(b->left);
+    }
+  } else {
+    while (true) {
+      if (petal) {
+        if (petal_offset == 0) {
+          petal = nullptr;
+          continue;
+        }
+        uint32_t got = 0;
+        *data = petal->source->read(petal->pos, &got);
+        if (petal_offset > got) {
+          uint32_t got2 = 0;
+          *data = petal->source->read(petal->pos + got, &got2);
+          *out_len = std::min(got2, petal_offset - got);
+          petal_offset = got;
+          return true;
+        } else {
+          *out_len = std::min(got, petal_offset);
+          petal_offset = 0;
+          return true;
+        }
+      }
+      if (stack.empty())
+        return false;
+      auto s = stack.back();
+      stack.pop_back();
+      while (s->kind == Shard::ShardKind::Branch) {
+        Branch *b = (Branch *)s;
+        stack.push_back(b->left);
+        s = b->right;
+      }
+      petal = (Petal *)s;
+      petal_offset = petal->length;
     }
   }
 }
diff --git a/src/vase/iterators/line.cc b/src/vase/iterators/line.cc
index 3710b91..17cad93 100644
--- a/src/vase/iterators/line.cc
+++ b/src/vase/iterators/line.cc
@@ -1,40 +1,65 @@
 #include "vase/iterators/line.h"
 
-LineIterator::LineIterator(Shard *r, uint32_t start_line) : it(r) {
-  it.seek_line(start_line);
+LineIterator::LineIterator(Shard *r, uint32_t start_line, Direction dir)
+    : it(r, dir), dir(dir) {
+  it.seek_line(start_line + (dir == Direction::Backward));
+  if (!it.next(&chunk, &len))
+    chunk = nullptr;
+}
+
+uint32_t LineIterator::byte_offset() {
+  return 0;
 }
 
 bool LineIterator::next(std::string &line) {
   line.clear();
-  line_offset = it.byte_offset() - remaining;
-  while (true) {
-    if (remaining == 0) {
-      uint32_t got;
-      if (!it.next(&cursor, &got)) {
-        eof = true;
-        break;
+  if (!chunk)
+    return false;
+  if (dir == Direction::Forward) {
+    while (true) {
+      const char *nl = (const char *)memchr(chunk, '\n', len);
+      if (!nl) {
+        if (len && chunk[len - 1] == '\r')
+          --len;
+        line.append(chunk, len);
+        if (!it.next(&chunk, &len)) {
+          chunk = nullptr;
+          return true;
+        }
+        continue;
       }
-      line_offset = it.byte_offset() - remaining;
-      remaining = got;
-    }
-    const char *nl = (const char *)memchr(cursor, '\n', remaining);
-    if (nl) {
       const char *end = nl;
-      uint32_t len = nl - cursor;
-      if (len && cursor[len - 1] == '\r')
-        --len;
-      line.append(cursor, len);
-      cursor = end + 1;
-      remaining -= (uint32_t)(len + 1);
+      if (end > chunk && *(end - 1) == '\r')
+        --end;
+      line.append(chunk, end);
+      uint32_t consumed = (nl - chunk) + 1;
+      chunk += consumed;
+      len -= consumed;
+      return true;
+    }
+  } else {
+    while (true) {
+      const char *nl = (const char *)memrchr(chunk, '\n', len);
+      if (!nl) {
+        if (len && chunk[len - 1] == '\n')
+          --len;
+        if (len && chunk[len - 1] == '\r')
+          --len;
+        line.insert(0, chunk, len);
+        if (!it.next(&chunk, &len)) {
+          chunk = nullptr;
+          return true;
+        }
+        continue;
+      }
+      const char *end = chunk + len;
+      const char *start = nl + 1;
+      const char *line_end = end;
+      if (line_end > start && *(line_end - 1) == '\r')
+        --line_end;
+      line.insert(0, start, line_end - start);
+      len = nl - chunk;
       return true;
     }
-    line.append(cursor, remaining);
-    cursor += remaining;
-    remaining = 0;
   }
-  return !line.empty();
-}
-
-uint32_t LineIterator::byte_offset() {
-  return line_offset;
 }
diff --git a/src/vase/search.cc b/src/vase/search.cc
index 54c2032..1c9a699 100644
--- a/src/vase/search.cc
+++ b/src/vase/search.cc
@@ -83,7 +83,7 @@ std::vector<RegexMatch> regex_search(
 
   pcre2_match_data *match_data = pcre2_match_data_create_from_pattern(re, NULL);
 
-  ChunkIterator it(root);
+  ChunkIterator it(root, Direction::Forward);
   it.seek_offset(start_offset);
 
   const char *data;
diff --git a/src/vase/shard.cc b/src/vase/shard.cc
index 7546602..8587dad 100644
--- a/src/vase/shard.cc
+++ b/src/vase/shard.cc
@@ -195,7 +195,7 @@ Shard *merge_leaves(Shard *a, Shard *b) {
 Shard *append_leaf(Shard *root, Shard *leaf) {
   if (!root)
     return leaf;
-  if (root->kind == Shard::ShardKind::Petal && root->length < 32 * 1024)
+  if (root->kind == Shard::ShardKind::Petal && root->length < PETAL_SIZE_MAX)
     return merge_leaves(root, leaf);
   Branch *b = (Branch *)root;
   auto new_right = append_leaf(b->right, leaf);
@@ -221,16 +221,15 @@ Shard *build_balanced(Shard **pieces, uint32_t lo, uint32_t hi) {
 }
 
 Shard *create_shards(Buffer *o) {
-  constexpr uint32_t PETAL_SIZE = 32 * 1024;
   std::vector<Shard *> pieces;
   uint32_t pos = 0;
   const uint32_t total = o->length();
-  pieces.reserve((total + PETAL_SIZE - 1) / PETAL_SIZE);
+  pieces.reserve((total + PETAL_SIZE_MAX - 1) / PETAL_SIZE_MAX);
 
   while (pos < total) {
     uint32_t len = 0;
     const char *data = o->read(pos, &len);
-    uint32_t take = std::min(len, PETAL_SIZE);
+    uint32_t take = std::min(len, PETAL_SIZE_MAX);
     uint32_t lines = 0;
     const char *p = data;
     const char *end = data + take;
diff --git a/src/vase/vase.cc b/src/vase/vase.cc
index fa399a8..5efd044 100644
--- a/src/vase/vase.cc
+++ b/src/vase/vase.cc
@@ -3,12 +3,18 @@
 #include "vase/iterators/line.h"
 #include "vase/search.h"
 
-Vase::Vase(char *data, uint32_t length)
-    : original(data, length), append(),
-      root(create_shards(&original)) {}
+Vase::Vase(std::string path)
+    : original(path), append(),
+      root(create_shards(&original)) {
+  history.push_back(root);
+  Shard::retain(root);
+  history_top = 0;
+}
 
 Vase::~Vase() {
   Shard::release(root);
+  for (auto s : history)
+    Shard::release(s);
 }
 
 uint32_t Vase::length() {
@@ -21,13 +27,56 @@ std::string Vase::to_string() {
   return out;
 }
 
-void Vase::input(Point *point, char key) {
+LineIterator Vase::iterate(uint32_t line) {
+  return LineIterator(root, line, Direction::Forward);
+}
+
+bool Vase::undo() {
+  if (history_top == 0)
+    return false;
+  Shard::release(root);
+  history_top--;
+  root = history[history_top];
+  Shard::retain(root);
+  return true;
+}
+
+bool Vase::redo() {
+  if (history_top + 1 >= history.size())
+    return false;
+  Shard::release(root);
+  history_top++;
+  root = history[history_top];
+  Shard::retain(root);
+  return true;
+}
+
+void Vase::snapshot() {
+  if (history[history_top] == root)
+    return;
+  while (history.size() > history_top + 1) {
+    Shard::release(history.back());
+    history.pop_back();
+  }
+  Shard::retain(root);
+  history.push_back(root);
+  history_top++;
+}
+
+void Vase::prune_history(uint32_t n) {
+  n = std::min(n, history_top);
+  for (uint32_t i = 0; i < n; ++i)
+    Shard::release(history[i]);
+  history.erase(history.begin(), history.begin() + n);
+  history_top -= n;
+}
+
+void Vase::insert(Point *point, char key) {
   if (key == '\n')
     *point = {point->row + 1, 0};
   else
     point->col++;
   uint32_t pos = append.key(key);
-  // limit petal size to 32KiB here later.
   Shard *inserted = new Petal(1, key == '\n', &append, pos);
   auto [left, right] = split_shard(root, offset_of(*point));
   Shard *left2 = append_leaf(left, inserted);
@@ -41,6 +90,15 @@ void Vase::input(Point *point, char key) {
 }
 
 void Vase::insert(Point *point, const char *data, uint32_t len) {
+  while (len) {
+    uint32_t chunk_size = std::min<uint32_t>(len, PETAL_SIZE_MAX);
+    _insert(point, data, chunk_size);
+    len -= chunk_size;
+    data += chunk_size;
+  }
+}
+
+void Vase::_insert(Point *point, const char *data, uint32_t len) {
   if (len == 0)
     return;
   uint32_t offset = offset_of(*point);
@@ -80,20 +138,20 @@ void Vase::insert(Point *point, const char *data, uint32_t len) {
 }
 
 void Vase::erase(Point *point, int64_t amount) {
-  // wrong: havta make amount be in clusters not bytes
-  uint32_t offset = offset_of(*point);
   if (amount == 0)
     return;
-  uint32_t start;
-  uint32_t count;
-  if (amount < 0) {
-    count = std::min<uint32_t>(-amount, offset);
-    start = offset - count;
-  } else {
-    start = offset;
-    count = amount;
-  }
-  auto [a, b] = split_shard(root, start);
+  Point start = *point;
+  Point end = *point;
+  if (amount < 0)
+    move_clusters(&start, amount);
+  else
+    move_clusters(&end, amount);
+  uint32_t start_offset = offset_of(start);
+  uint32_t end_offset = offset_of(end);
+  if (start_offset > end_offset)
+    std::swap(start_offset, end_offset);
+  uint32_t count = end_offset - start_offset;
+  auto [a, b] = split_shard(root, start_offset);
   auto [d, c] = split_shard(b, count);
   Shard *new_root = concat_shard(a, c);
   Shard::release(a);
@@ -102,6 +160,30 @@ void Vase::erase(Point *point, int64_t amount) {
   Shard::release(d);
   Shard::release(root);
   root = new_root;
+  if (amount < 0)
+    *point = start;
+}
+
+void Vase::erase(Range range) {
+  Point start = range.start;
+  Point end = range.end;
+  uint32_t start_offset = offset_of(start);
+  uint32_t end_offset = offset_of(end);
+  uint32_t count = end_offset - start_offset;
+  auto [a, b] = split_shard(root, start_offset);
+  auto [d, c] = split_shard(b, count);
+  Shard *new_root = concat_shard(a, c);
+  Shard::release(a);
+  Shard::release(b);
+  Shard::release(c);
+  Shard::release(d);
+  Shard::release(root);
+  root = new_root;
+}
+
+void Vase::replace(Range range, const char *data, uint32_t len) {
+  erase(range);
+  insert(&range.start, data, len);
 }
 
 void Vase::flatten(Shard *s, std::string &out) {
@@ -125,7 +207,7 @@ void Vase::flatten(Shard *s, std::string &out) {
 }
 
 uint32_t Vase::offset_of(Point point) {
-  LineIterator it(root, point.row);
+  LineIterator it(root, point.row, Direction::Forward);
   std::string line;
   uint32_t offset = 0;
   if (it.next(line)) {
@@ -142,14 +224,100 @@ uint32_t Vase::offset_of(Point point) {
   return it.byte_offset() + offset;
 }
 
-bool Vase::jump(Point *, int64_t) {
-  // todo.
-  return false;
+void Vase::move_clusters(Point *point, int64_t amount) {
+  if (amount == 0)
+    return;
+  if (amount < 0) {
+    amount = -amount;
+    LineIterator it(root, point->row, Direction::Backward);
+    while (amount > 0) {
+      std::string line;
+      if (!it.next(line))
+        return;
+      std::vector<uint32_t> clusters;
+      const char *ptr = line.data();
+      uint32_t remaining = line.size();
+      uint32_t byte = 0;
+      while (remaining) {
+        clusters.push_back(byte);
+        uint32_t len =
+          grapheme_next_character_break_utf8(ptr, remaining);
+        ptr += len;
+        remaining -= len;
+        byte += len;
+      }
+      while (amount > 0 && point->col > 0) {
+        point->col--;
+        amount--;
+      }
+      if (amount == 0)
+        return;
+      if (point->row == 0)
+        return;
+      point->row--;
+      point->col = clusters.size();
+      amount--;
+    }
+  } else {
+    LineIterator it(root, point->row, Direction::Forward);
+    while (amount > 0) {
+      std::string line;
+      if (!it.next(line))
+        return;
+      if (point->col > 0 || amount > 0) {
+        const char *ptr = line.data();
+        uint32_t remaining = line.size();
+        uint32_t col = 0;
+        while (col < point->col && remaining) {
+          uint32_t len = grapheme_next_character_break_utf8(ptr, remaining);
+          ptr += len;
+          remaining -= len;
+          col++;
+        }
+        while (amount > 0 && remaining) {
+          uint32_t len = grapheme_next_character_break_utf8(ptr, remaining);
+          ptr += len;
+          remaining -= len;
+          point->col++;
+          amount--;
+        }
+      }
+      if (amount > 0) {
+        point->row++;
+        point->col = 0;
+        amount--;
+      }
+    }
+  }
 }
 
-bool Vase::clamp(Point *) {
-  // todo.
-  return false;
+void Vase::clamp(Point *point) {
+  if (point->row > root->lines)
+    point->row = root->lines;
+  LineIterator it(root, point->row, Direction::Forward);
+  std::string line;
+  if (!it.next(line)) {
+    point->col = 0;
+    return;
+  }
+  uint32_t clusters = 0;
+  const char *ptr = line.data();
+  uint32_t remaining = line.size();
+  while (remaining) {
+    uint32_t len = grapheme_next_character_break_utf8(ptr, remaining);
+    ptr += len;
+    remaining -= len;
+    clusters++;
+  }
+  if (point->col > clusters)
+    point->col = clusters;
+}
+
+void Vase::move_lines(Point *point, int64_t amount) {
+  if (point->row + amount < 0)
+    return;
+  point->row += amount;
+  clamp(point);
 }
 
 struct ReplacePart {