diff --git a/README.md b/README.md
index fe8b908..753d5a4 100644
--- a/README.md
+++ b/README.md
@@ -296,6 +296,22 @@ $ sql-pipe orders.csv customers.csv \
      JOIN customers c ON o.cust_id = c.id GROUP BY c.name'
 ```
 
+Gzip-compressed inputs are handled transparently (gzip only). A file ending in `.gz`
+is decompressed automatically; the format is detected from the inner extension, and
+the table name comes from the inner basename:
+
+```sh
+$ sql-pipe data.csv.gz 'SELECT COUNT(*) FROM data'
+```
+
+Gzipped stdin is detected by its magic bytes, so both decompressed and compressed
+pipes work:
+
+```sh
+$ zcat huge.csv.gz | sql-pipe 'SELECT * FROM t'          # decompressed by zcat
+$ gzip -c data.ndjson | sql-pipe 'SELECT * FROM t'       # gzip stdin magic detection
+```
+
 Use `-I` to override auto-detection when the extension is wrong or ambiguous (`.txt`, `.dat`):
 
 ```sh
diff --git a/build.zig b/build.zig
index 62bed45..f1ea831 100644
--- a/build.zig
+++ b/build.zig
@@ -3708,6 +3708,84 @@ pub fn build(b: *std.Build) void {
     test_parquet_fuzz_empty.step.dependOn(b.getInstallStep());
     test_step.dependOn(&test_parquet_fuzz_empty.step);
 
+    // ─── Gzip integration tests ────────────────────────────────────────────────
+
+    // Integration test: gzip file input (tests/fixtures/sample.csv.gz)
+    const test_gzip_file = b.addSystemCommand(&.{
+        "bash", "-c",
+        \\result=$(./zig-out/bin/sql-pipe tests/fixtures/sample.csv.gz 'SELECT name FROM sample WHERE age > 27 ORDER BY name')
+        \\expected=$(printf 'Alice\nCarol')
+        \\[ "$result" = "$expected" ]
+    });
+    test_gzip_file.step.dependOn(b.getInstallStep());
+    test_step.dependOn(&test_gzip_file.step);
+
+    // Integration test: gzip file table name derived from inner name (sample.csv.gz → "sample")
+    const test_gzip_table_name = b.addSystemCommand(&.{
+        "bash", "-c",
+        \\result=$(./zig-out/bin/sql-pipe tests/fixtures/sample.csv.gz 'SELECT name FROM sample ORDER BY name')
+        \\expected=$(printf 'Alice\nBob\nCarol')
+        \\[ "$result" = "$expected" ]
+    });
+    test_gzip_table_name.step.dependOn(b.getInstallStep());
+    test_step.dependOn(&test_gzip_table_name.step);
+
+    // Integration test: gzip stdin magic sniffing
+    // sample.csv.gz is already gzipped; pipe it directly (re-gzipping would double-compress).
+    const test_gzip_stdin = b.addSystemCommand(&.{
+        "bash", "-c",
+        \\result=$(cat tests/fixtures/sample.csv.gz | ./zig-out/bin/sql-pipe 'SELECT name FROM t WHERE age < 30')
+        \\expected=$(printf 'Bob\n')
+        \\[ "$result" = "$expected" ]
+    });
+    test_gzip_stdin.step.dependOn(b.getInstallStep());
+    test_step.dependOn(&test_gzip_stdin.step);
+
+    // Integration test: non-gzip data on stdin is not misdetected as corrupt gzip
+    const test_gzip_non_gzip_stdin = b.addSystemCommand(&.{
+        "bash", "-c",
+        \\msg=$(printf 'hello\n' | ./zig-out/bin/sql-pipe --output-format json 'SELECT 1' 2>&1; echo "EXIT:$?")
+        \\echo "$msg" | grep -q 'EXIT:0'
+    });
+    test_gzip_non_gzip_stdin.step.dependOn(b.getInstallStep());
+    test_step.dependOn(&test_gzip_non_gzip_stdin.step);
+
+    // Integration test: non-gzip stdin unchanged (regular CSV via stdin still works)
+    const test_gzip_stdin_unchanged = b.addSystemCommand(&.{
+        "bash", "-c",
+        \\result=$(printf 'name,age\nAlice,30\n' | ./zig-out/bin/sql-pipe 'SELECT name FROM t')
+        \\[ "$result" = "Alice" ]
+    });
+    test_gzip_stdin_unchanged.step.dependOn(b.getInstallStep());
+    test_step.dependOn(&test_gzip_stdin_unchanged.step);
+
+    // Integration test: .ndjson.gz inner format detection
+    const test_gzip_ndjson = b.addSystemCommand(&.{
+        "bash", "-c",
+        \\printf '{"name":"Alice","age":30}\n{"name":"Bob","age":25}\n' | gzip -c > /tmp/test.ndjson.gz
+        \\result=$(./zig-out/bin/sql-pipe /tmp/test.ndjson.gz 'SELECT name FROM test WHERE age > 27')
+        \\rm -f /tmp/test.ndjson.gz
+        \\[ "$result" = "Alice" ]
+    });
+    test_gzip_ndjson.step.dependOn(b.getInstallStep());
+    test_step.dependOn(&test_gzip_ndjson.step);
+
+    // Integration test: corrupt/truncated gzip file exits non-zero with a
+    // gzip-specific message. The truncated fixture is copied to a real .gz name
+    // (its own name ends in `.gz_truncated`, which isGzipExtension rejects).
+    const test_gzip_corrupt = b.addSystemCommand(&.{
+        "bash", "-c",
+        \\set -euo pipefail
+        \\cp tests/fixtures/sample.csv.gz_truncated /tmp/sql-pipe-corrupt.csv.gz
+        \\trap 'rm -f /tmp/sql-pipe-corrupt.csv.gz' EXIT
+        \\msg=$(./zig-out/bin/sql-pipe /tmp/sql-pipe-corrupt.csv.gz 'SELECT 1' 2>&1; echo "EXIT:$?")
+        \\echo "$msg" | grep -q 'EXIT:2' || { echo "expected exit 2, got: $msg"; exit 1; }
+        \\echo "$msg" | grep -q 'gzip decompression failed: EndOfStream' || { echo "missing gzip error msg: $msg"; exit 1; }
+        \\
+    });
+    test_gzip_corrupt.step.dependOn(b.getInstallStep());
+    test_step.dependOn(&test_gzip_corrupt.step);
+
     // ─── --checksum integration tests (issue #204) ──────────────────────────────
     // 25 data-driven cases (204a-204y). Scripts run with `set -euo pipefail` so a
     // failed assertion actually fails the step (a bare `[` + trailing `rm -f`
diff --git a/docs/sql-pipe.1.scd b/docs/sql-pipe.1.scd
index 75e72cd..7108461 100644
--- a/docs/sql-pipe.1.scd
+++ b/docs/sql-pipe.1.scd
@@ -36,6 +36,13 @@ DESCRIPTION
 	missing extensions default to CSV. Use *-I* to override auto-detection
 	(e.g., when a TSV file has a *.txt* extension).
 
+	Gzip-compressed inputs are supported transparently (gzip only). Files ending
+	in *.gz* are decompressed automatically; the input format is then detected
+	from the inner extension (e.g., *data.csv.gz* is read as CSV) and the table
+	name comes from the inner basename (*data.csv.gz* becomes table *data*).
+	Gzipped data piped to standard input is detected by its gzip magic bytes
+	(*\x1f\x8b*), so compressed stdin works without any extra flags.
+
 	This tool is useful for quick data transformations, filtering, grouping,
 	aggregations, and multi-file joins without manual SQL database setup.
 
@@ -339,6 +346,18 @@ EXAMPLES
 
 		$ sql-pipe data.parquet 'SELECT name FROM data WHERE active = true'
 
+	Query a gzip-compressed CSV file (format detected from inner extension):
+
+		$ sql-pipe data.csv.gz 'SELECT COUNT(*) FROM data'
+
+	Pipe decompressed stdin (zcat decompresses, stdin reads as plain CSV):
+
+		$ zcat huge.csv.gz | sql-pipe 'SELECT * FROM t'
+
+	Pipe gzip-compressed stdin (detected by magic bytes):
+
+		$ gzip -c data.ndjson | sql-pipe 'SELECT * FROM t'
+
 	Override auto-detection when the extension is wrong:
 
 		$ sql-pipe -I tsv data.txt 'SELECT * FROM data'
diff --git a/src/args.zig b/src/args.zig
index 672fa2d..c4ee930 100644
--- a/src/args.zig
+++ b/src/args.zig
@@ -1003,7 +1003,11 @@ fn isValidHttpHeader(header: []const u8) bool {
 }
 
 /// Derive a table name from a file path (basename without extension).
+/// For .gz files, strips the .gz first so "data.csv.gz" → "data".
 fn tableNameFromPath(allocator: std.mem.Allocator, path: []const u8) (std.mem.Allocator.Error)![]const u8 {
-    const stem = std.fs.path.stem(path);
+    const inner = format.InputFormat.stripGzExtension(path);
+    // stripGzExtension can collapse a path named exactly ".gz" to "", whose
+    // stem is also empty and would produce an invalid empty table name.
+    const stem = if (std.fs.path.stem(inner).len == 0) std.fs.path.stem(path) else std.fs.path.stem(inner);
     return allocator.dupe(u8, stem);
 }
diff --git a/src/format.zig b/src/format.zig
index afa15d6..b11caeb 100644
--- a/src/format.zig
+++ b/src/format.zig
@@ -33,13 +33,26 @@ pub const InputFormat = enum {
 
     /// Detect input format from file extension.
     /// Returns null for unrecognized extensions.
+    /// A ".gz" extension is stripped first, so "data.csv.gz" maps to .csv.
     pub fn fromExtension(filename: []const u8) ?InputFormat {
-        const ext = std.fs.path.extension(filename);
+        const inner = stripGzExtension(filename);
+        const ext = std.fs.path.extension(inner);
         if (ext.len == 0) return null;
         const ext_no_dot = ext[1..]; // skip the leading '.'
         if (std.mem.eql(u8, ext_no_dot, "yml")) return .yaml;
         return std.meta.stringToEnum(InputFormat, ext_no_dot);
     }
+
+    /// Return true when filename ends with ".gz".
+    pub fn isGzipExtension(filename: []const u8) bool {
+        return std.mem.endsWith(u8, filename, ".gz");
+    }
+
+    /// Return filename without a trailing ".gz", or the original when it isn't a .gz file.
+    pub fn stripGzExtension(filename: []const u8) []const u8 {
+        if (!isGzipExtension(filename)) return filename;
+        return filename[0 .. filename.len - 3];
+    }
 };
 
 // ─── Output format ─────────────────────────────────────
diff --git a/src/main.zig b/src/main.zig
index ac07ea6..03265c7 100644
--- a/src/main.zig
+++ b/src/main.zig
@@ -114,6 +114,108 @@ fn writeStreaming(
     try out_writer.end(writer);
 }
 
+/// Wrapper around a gzip-decompressing reader. Heap-allocated so the returned
+/// `*std.Io.Reader` stays valid for the lifetime of the program. The wrapper
+/// shares the flate decompressor's buffer and mirrors its cursor state, so it
+/// behaves exactly like the flate reader; its only job is to translate the raw
+/// `error.ReadFailed` into a message naming the specific flate failure (stored
+/// in `decompress.err`) instead of a generic read error.
+///
+/// The wrapper forwards to flate's *vtable* methods (not the generic
+/// `Reader.readVec`/`stream`/`discard` helpers), because the generic helpers
+/// short-circuit on the empty `data` slice that `fill` passes and would return
+/// 0 without ever refilling the decompressor's buffer, spinning forever.
+const GzipReader = struct {
+    decompress: std.compress.flate.Decompress,
+    buffer: [std.compress.flate.max_window_len]u8,
+    stderr_writer: *std.Io.Writer,
+    reader: std.Io.Reader,
+
+    fn sync(r: *std.Io.Reader) *GzipReader {
+        const gz: *GzipReader = @fieldParentPtr("reader", r);
+        r.buffer = gz.decompress.reader.buffer;
+        r.seek = gz.decompress.reader.seek;
+        r.end = gz.decompress.reader.end;
+        return gz;
+    }
+
+    fn stream(r: *std.Io.Reader, w: *std.Io.Writer, limit: std.Io.Limit) std.Io.Reader.StreamError!usize {
+        _ = sync(r);
+        const gz: *GzipReader = @fieldParentPtr("reader", r);
+        const n = gz.decompress.reader.vtable.stream(&gz.decompress.reader, w, limit) catch |err| switch (err) {
+            error.WriteFailed => return error.WriteFailed,
+            error.EndOfStream => return error.EndOfStream,
+            error.ReadFailed => {
+                const detail = gz.decompress.err orelse return error.ReadFailed;
+                fatal("gzip decompression failed: {s}", gz.stderr_writer, .csv_error, .{@errorName(detail)});
+            },
+        };
+        _ = sync(r);
+        return n;
+    }
+
+    fn discard(r: *std.Io.Reader, limit: std.Io.Limit) std.Io.Reader.Error!usize {
+        _ = sync(r);
+        const gz: *GzipReader = @fieldParentPtr("reader", r);
+        const n = gz.decompress.reader.vtable.discard(&gz.decompress.reader, limit) catch |err| switch (err) {
+            error.EndOfStream => return error.EndOfStream,
+            error.ReadFailed => {
+                const detail = gz.decompress.err orelse return error.ReadFailed;
+                fatal("gzip decompression failed: {s}", gz.stderr_writer, .csv_error, .{@errorName(detail)});
+            },
+        };
+        _ = sync(r);
+        return n;
+    }
+
+    fn readVec(r: *std.Io.Reader, data: [][]u8) std.Io.Reader.Error!usize {
+        _ = sync(r);
+        const gz: *GzipReader = @fieldParentPtr("reader", r);
+        const n = gz.decompress.reader.vtable.readVec(&gz.decompress.reader, data) catch |err| switch (err) {
+            error.EndOfStream => return error.EndOfStream,
+            error.ReadFailed => {
+                const detail = gz.decompress.err orelse return error.ReadFailed;
+                fatal("gzip decompression failed: {s}", gz.stderr_writer, .csv_error, .{@errorName(detail)});
+            },
+        };
+        _ = sync(r);
+        return n;
+    }
+
+    fn rebase(r: *std.Io.Reader, capacity: usize) std.Io.Reader.RebaseError!void {
+        _ = sync(r);
+        const gz: *GzipReader = @fieldParentPtr("reader", r);
+        try gz.decompress.reader.vtable.rebase(&gz.decompress.reader, capacity);
+        _ = sync(r);
+    }
+
+    const vtable: std.Io.Reader.VTable = .{
+        .stream = stream,
+        .discard = discard,
+        .readVec = readVec,
+        .rebase = rebase,
+    };
+};
+
+/// Wrap `reader` in a gzip decompressing reader and return its reader.
+pub fn makeGzipReader(allocator: std.mem.Allocator, reader: *std.Io.Reader, stderr_writer: *std.Io.Writer) !*std.Io.Reader {
+    const gz = try allocator.create(GzipReader);
+    // ponytail: leaked for program lifetime; input pointer valid only
+    // within loadPipelineInputs scope. CLI, single-use.
+    gz.* = .{
+        .decompress = std.compress.flate.Decompress.init(reader, .gzip, &gz.buffer),
+        .buffer = undefined,
+        .stderr_writer = stderr_writer,
+        .reader = .{
+            .vtable = &GzipReader.vtable,
+            .buffer = &gz.buffer,
+            .seek = 0,
+            .end = 0,
+        },
+    };
+    return &gz.reader;
+}
+
 const progress_interval = loader.progress_interval;
 const fatal = sqlite_mod.fatal;
 
@@ -307,7 +409,16 @@ pub fn loadPipelineInputs(
             fatal("cannot open file '{s}': {s}", stderr_writer, .csv_error, .{ file_input.path, @errorName(err) });
         defer std.Io.File.close(file, io);
         var file_reader = std.Io.File.reader(file, io, &file_buf);
-        const rows = loadInput(allocator, io, db, file_input.table_name, file_input.format, &file_reader.interface, parsed, stderr_writer);
+        const is_gz = InputFormat.isGzipExtension(file_input.path);
+        // For .gz files: wrap in gzip decompressor, auto-detect inner format if not explicit
+        const effective_reader: *std.Io.Reader = if (is_gz)
+            makeGzipReader(allocator, &file_reader.interface, stderr_writer) catch
+                fatal("out of memory allocating gzip reader for '{s}'", stderr_writer, .csv_error, .{file_input.path})
+        else
+            &file_reader.interface;
+        // Format already resolved at parse time (args.zig uses fromExtension which strips .gz)
+        const effective_format = file_input.format;
+        const rows = loadInput(allocator, io, db, file_input.table_name, effective_format, effective_reader, parsed, stderr_writer);
         if (rows == 0) {
             fatal("empty input file: '{s}'", stderr_writer, .csv_error, .{file_input.path});
         }
@@ -318,8 +429,18 @@ pub fn loadPipelineInputs(
     if (parsed.has_stdin) {
         var stdin_buf: [4096]u8 = undefined;
         var stdin_reader = std.Io.File.reader(std.Io.File.stdin(), io, &stdin_buf);
+        // Sniff gzip magic by peeking first 2 bytes (peek fills buffer and
+        // returns without consuming, so we can wrap in place).
+        var effective: *std.Io.Reader = &stdin_reader.interface;
+        const peeked = stdin_reader.interface.peek(2) catch null;
+        if (peeked) |b| {
+            if (b.len >= 2 and b[0] == 0x1f and b[1] == 0x8b) {
+                effective = makeGzipReader(allocator, &stdin_reader.interface, stderr_writer) catch
+                    fatal("out of memory allocating gzip reader for stdin", stderr_writer, .csv_error, .{});
+            }
+        }
         const stdin_table = if (parsed.urls.len > 0 or parsed.files.len > 0) "stdin" else "t";
-        const rows = loadInput(allocator, io, db, stdin_table, parsed.input_format, &stdin_reader.interface, parsed, stderr_writer);
+        const rows = loadInput(allocator, io, db, stdin_table, parsed.input_format, effective, parsed, stderr_writer);
         total_rows += rows;
     }
 
diff --git a/tests/fixtures/sample.csv.gz b/tests/fixtures/sample.csv.gz
new file mode 100644
index 0000000..c2d22ba
Binary files /dev/null and b/tests/fixtures/sample.csv.gz differ
diff --git a/tests/fixtures/sample.csv.gz_truncated b/tests/fixtures/sample.csv.gz_truncated
new file mode 100644
index 0000000..e08934e
Binary files /dev/null and b/tests/fixtures/sample.csv.gz_truncated differ