diff --git a/Sources/ForgeOffload/FileHasher.swift b/Sources/ForgeOffload/FileHasher.swift
new file mode 100644
index 0000000..f87edeb
--- /dev/null
+++ b/Sources/ForgeOffload/FileHasher.swift
@@ -0,0 +1,37 @@
+import Foundation
+
+public enum OffloadError: Error, Equatable {
+    case fileNotFound(String)
+    case readFailed(String)
+    case writeFailed(String)
+    case verifyFailed(path: String, expected: String, actual: String)
+    case destinationConflict(String)
+}
+
+/// Chunked file hashing — constant memory over multi-GB camera files.
+public enum FileHasher {
+    static let chunkSize = 4 * 1024 * 1024
+
+    public static func xxh64(path: String) throws -> String {
+        var hasher = XXHash64()
+        try stream(path: path) { hasher.update($0) }
+        return XXHash64.hexString(hasher.finalize())
+    }
+
+    public static func md5(path: String) throws -> String {
+        var hasher = MD5()
+        try stream(path: path) { hasher.update($0) }
+        return hasher.finalize().map { String(format: "%02x", $0) }.joined()
+    }
+
+    private static func stream(path: String, _ body: (Data) -> Void) throws {
+        guard let handle = FileHandle(forReadingAtPath: path) else {
+            throw OffloadError.fileNotFound(path)
+        }
+        defer { try? handle.close() }
+        while true {
+            guard let chunk = try handle.read(upToCount: chunkSize), !chunk.isEmpty else { break }
+            body(chunk)
+        }
+    }
+}
diff --git a/Sources/ForgeOffload/ForgeOffload.swift b/Sources/ForgeOffload/ForgeOffload.swift
deleted file mode 100644
index 35edea5..0000000
--- a/Sources/ForgeOffload/ForgeOffload.swift
+++ /dev/null
@@ -1 +0,0 @@
-// ForgeOffload
diff --git a/Sources/ForgeOffload/MD5.swift b/Sources/ForgeOffload/MD5.swift
new file mode 100644
index 0000000..aa9ce4c
--- /dev/null
+++ b/Sources/ForgeOffload/MD5.swift
@@ -0,0 +1,108 @@
+import Foundation
+
+/// MD5 (RFC 1321) — pure Swift. Legacy-post-workflow compatibility only;
+/// integrity verification uses xxHash64.
+public struct MD5 {
+    private var a0: UInt32 = 0x67452301
+    private var b0: UInt32 = 0xefcdab89
+    private var c0: UInt32 = 0x98badcfe
+    private var d0: UInt32 = 0x10325476
+    private var buffer = [UInt8]()
+    private var totalLength: UInt64 = 0
+
+    private static let s: [UInt32] = [
+        7, 12, 17, 22, 7, 12, 17, 22, 7, 12, 17, 22, 7, 12, 17, 22,
+        5, 9, 14, 20, 5, 9, 14, 20, 5, 9, 14, 20, 5, 9, 14, 20,
+        4, 11, 16, 23, 4, 11, 16, 23, 4, 11, 16, 23, 4, 11, 16, 23,
+        6, 10, 15, 21, 6, 10, 15, 21, 6, 10, 15, 21, 6, 10, 15, 21,
+    ]
+
+    private static let k: [UInt32] = (0..<64).map {
+        UInt32(truncatingIfNeeded: Int64(abs(sin(Double($0 + 1))) * 4294967296.0))
+    }
+
+    public init() {}
+
+    public mutating func update(_ data: Data) {
+        totalLength += UInt64(data.count)
+        buffer.append(contentsOf: data)
+        var offset = 0
+        while buffer.count - offset >= 64 {
+            processBlock(Array(buffer[offset..<offset + 64]))
+            offset += 64
+        }
+        if offset > 0 {
+            buffer.removeFirst(offset)
+        }
+    }
+
+    private mutating func processBlock(_ block: [UInt8]) {
+        var m = [UInt32](repeating: 0, count: 16)
+        for i in 0..<16 {
+            m[i] = UInt32(block[i * 4])
+                | (UInt32(block[i * 4 + 1]) << 8)
+                | (UInt32(block[i * 4 + 2]) << 16)
+                | (UInt32(block[i * 4 + 3]) << 24)
+        }
+        var a = a0, b = b0, c = c0, d = d0
+        for i in 0..<64 {
+            var f: UInt32
+            var g: Int
+            switch i {
+            case 0..<16:
+                f = (b & c) | (~b & d)
+                g = i
+            case 16..<32:
+                f = (d & b) | (~d & c)
+                g = (5 * i + 1) % 16
+            case 32..<48:
+                f = b ^ c ^ d
+                g = (3 * i + 5) % 16
+            default:
+                f = c ^ (b | ~d)
+                g = (7 * i) % 16
+            }
+            f = f &+ a &+ Self.k[i] &+ m[g]
+            a = d
+            d = c
+            c = b
+            b = b &+ ((f << Self.s[i]) | (f >> (32 - Self.s[i])))
+        }
+        a0 = a0 &+ a
+        b0 = b0 &+ b
+        c0 = c0 &+ c
+        d0 = d0 &+ d
+    }
+
+    public mutating func finalize() -> Data {
+        let bitLength = totalLength * 8
+        // Padding: 0x80 then zeros to 56 mod 64, then length LE64.
+        var padding: [UInt8] = [0x80]
+        let remainder = (buffer.count + 1) % 64
+        let zeros = remainder <= 56 ? 56 - remainder : 120 - remainder
+        padding.append(contentsOf: [UInt8](repeating: 0, count: zeros))
+        var lenLE = bitLength.littleEndian
+        withUnsafeBytes(of: &lenLE) { padding.append(contentsOf: $0) }
+
+        buffer.append(contentsOf: padding)
+        var offset = 0
+        while buffer.count - offset >= 64 {
+            processBlock(Array(buffer[offset..<offset + 64]))
+            offset += 64
+        }
+        buffer.removeAll()
+
+        var out = Data()
+        for v in [a0, b0, c0, d0] {
+            var le = v.littleEndian
+            withUnsafeBytes(of: &le) { out.append(contentsOf: $0) }
+        }
+        return out
+    }
+
+    public static func hexString(_ data: Data) -> String {
+        var md5 = MD5()
+        md5.update(data)
+        return md5.finalize().map { String(format: "%02x", $0) }.joined()
+    }
+}
diff --git a/Sources/ForgeOffload/XXHash64.swift b/Sources/ForgeOffload/XXHash64.swift
new file mode 100644
index 0000000..e19844d
--- /dev/null
+++ b/Sources/ForgeOffload/XXHash64.swift
@@ -0,0 +1,126 @@
+import Foundation
+
+/// xxHash64 (seed 0) — pure Swift, streaming. Spec: github.com/Cyan4973/xxHash.
+public struct XXHash64 {
+    private static let p1: UInt64 = 0x9E3779B185EBCA87
+    private static let p2: UInt64 = 0xC2B2AE3D27D4EB4F
+    private static let p3: UInt64 = 0x165667B19E3779F9
+    private static let p4: UInt64 = 0x85EBCA77C2B2AE63
+    private static let p5: UInt64 = 0x27D4EB2F165667C5
+
+    private var v1: UInt64
+    private var v2: UInt64
+    private var v3: UInt64
+    private var v4: UInt64
+    private var buffer = [UInt8]()
+    private var totalLength: UInt64 = 0
+    private let seed: UInt64
+
+    public init(seed: UInt64 = 0) {
+        self.seed = seed
+        v1 = seed &+ Self.p1 &+ Self.p2
+        v2 = seed &+ Self.p2
+        v3 = seed
+        v4 = seed &- Self.p1
+    }
+
+    @inline(__always)
+    private static func rotl(_ x: UInt64, _ r: UInt64) -> UInt64 {
+        (x << r) | (x >> (64 - r))
+    }
+
+    @inline(__always)
+    private static func round(_ acc: UInt64, _ input: UInt64) -> UInt64 {
+        rotl(acc &+ input &* p2, 31) &* p1
+    }
+
+    @inline(__always)
+    private static func mergeRound(_ acc: UInt64, _ val: UInt64) -> UInt64 {
+        (acc ^ round(0, val)) &* p1 &+ p4
+    }
+
+    @inline(__always)
+    private static func readLE64(_ bytes: [UInt8], _ offset: Int) -> UInt64 {
+        var v: UInt64 = 0
+        for i in 0..<8 {
+            v |= UInt64(bytes[offset + i]) << (8 * UInt64(i))
+        }
+        return v
+    }
+
+    @inline(__always)
+    private static func readLE32(_ bytes: [UInt8], _ offset: Int) -> UInt64 {
+        var v: UInt64 = 0
+        for i in 0..<4 {
+            v |= UInt64(bytes[offset + i]) << (8 * UInt64(i))
+        }
+        return v
+    }
+
+    public mutating func update(_ data: Data) {
+        totalLength += UInt64(data.count)
+        buffer.append(contentsOf: data)
+        // Consume full 32-byte stripes, keep remainder buffered.
+        var offset = 0
+        while buffer.count - offset >= 32 {
+            v1 = Self.round(v1, Self.readLE64(buffer, offset))
+            v2 = Self.round(v2, Self.readLE64(buffer, offset + 8))
+            v3 = Self.round(v3, Self.readLE64(buffer, offset + 16))
+            v4 = Self.round(v4, Self.readLE64(buffer, offset + 24))
+            offset += 32
+        }
+        if offset > 0 {
+            buffer.removeFirst(offset)
+        }
+    }
+
+    public func finalize() -> UInt64 {
+        var h: UInt64
+        if totalLength >= 32 {
+            h = Self.rotl(v1, 1) &+ Self.rotl(v2, 7) &+ Self.rotl(v3, 12) &+ Self.rotl(v4, 18)
+            h = Self.mergeRound(h, v1)
+            h = Self.mergeRound(h, v2)
+            h = Self.mergeRound(h, v3)
+            h = Self.mergeRound(h, v4)
+        } else {
+            h = seed &+ Self.p5
+        }
+        h &+= totalLength
+
+        var offset = 0
+        let tail = buffer
+        while tail.count - offset >= 8 {
+            h ^= Self.round(0, Self.readLE64(tail, offset))
+            h = Self.rotl(h, 27) &* Self.p1 &+ Self.p4
+            offset += 8
+        }
+        if tail.count - offset >= 4 {
+            h ^= Self.readLE32(tail, offset) &* Self.p1
+            h = Self.rotl(h, 23) &* Self.p2 &+ Self.p3
+            offset += 4
+        }
+        while offset < tail.count {
+            h ^= UInt64(tail[offset]) &* Self.p5
+            h = Self.rotl(h, 11) &* Self.p1
+            offset += 1
+        }
+
+        h ^= h >> 33
+        h &*= Self.p2
+        h ^= h >> 29
+        h &*= Self.p3
+        h ^= h >> 32
+        return h
+    }
+
+    /// One-shot.
+    public static func hash(_ data: Data, seed: UInt64 = 0) -> UInt64 {
+        var h = XXHash64(seed: seed)
+        h.update(data)
+        return h.finalize()
+    }
+
+    public static func hexString(_ value: UInt64) -> String {
+        String(format: "%016lx", value)
+    }
+}
diff --git a/Tests/ForgeOffloadTests/HashingTests.swift b/Tests/ForgeOffloadTests/HashingTests.swift
new file mode 100644
index 0000000..3404715
--- /dev/null
+++ b/Tests/ForgeOffloadTests/HashingTests.swift
@@ -0,0 +1,82 @@
+import XCTest
+import Foundation
+@testable import ForgeOffload
+
+final class HashingTests: XCTestCase {
+
+    // xxHash64 official spec vectors (seed 0).
+    func testXXHash64SpecVectors() {
+        XCTAssertEqual(XXHash64.hash(Data()), 0xEF46_DB37_51D8_E999)
+        XCTAssertEqual(XXHash64.hash(Data([0x9E])), 0x4FCE_394C_C88952D8)
+        // "Nobody inspects the spammish repetition" — widely published vector.
+        let spam = Data("Nobody inspects the spammish repetition".utf8)
+        XCTAssertEqual(XXHash64.hash(spam), 0xFBCE_A83C_8A378BF1)
+    }
+
+    // Known ASCII vector.
+    func testXXHash64Ascii() {
+        XCTAssertEqual(XXHash64.hash(Data("xxhash".utf8)), 0x32DD_38952C_4BC720)
+    }
+
+    // Streaming hash == whole-buffer hash for multi-MB input crossing block sizes.
+    func testStreamingMatchesWhole() {
+        var data = Data()
+        var seed: UInt64 = 0x12345678
+        for _ in 0..<(3 * 1024 * 1024 / 8) {
+            seed = seed &* 6364136223846793005 &+ 1442695040888963407
+            withUnsafeBytes(of: seed) { data.append(contentsOf: $0) }
+        }
+        let whole = XXHash64.hash(data)
+
+        var streamer = XXHash64()
+        // Uneven chunk sizes to exercise buffering.
+        var offset = 0
+        let chunks = [1, 31, 32, 33, 4096, 1_000_000]
+        var i = 0
+        while offset < data.count {
+            let size = min(chunks[i % chunks.count], data.count - offset)
+            streamer.update(data.subdata(in: offset..<offset + size))
+            offset += size
+            i += 1
+        }
+        XCTAssertEqual(streamer.finalize(), whole)
+    }
+
+    // Hex formatting: 16 lowercase hex chars, zero-padded.
+    func testHexFormat() {
+        let h = XXHash64.hexString(XXHash64.hash(Data("xxhash".utf8)))
+        XCTAssertEqual(h, "32dd38952c4bc720")
+        // Zero-padding for small values.
+        XCTAssertEqual(XXHash64.hexString(0xAB), "00000000000000ab")
+    }
+
+    // MD5 known vectors (RFC 1321).
+    func testMD5Vectors() {
+        XCTAssertEqual(MD5.hexString(Data()), "d41d8cd98f00b204e9800998ecf8427e")
+        XCTAssertEqual(MD5.hexString(Data("abc".utf8)), "900150983cd24fb0d6963f7d28e17f72")
+        XCTAssertEqual(
+            MD5.hexString(Data("message digest".utf8)),
+            "f96b697d7cb7938d525a2f31aaf161d0")
+    }
+
+    // File hashing: chunked reader over a temp file matches in-memory hash.
+    func testFileHashing() throws {
+        let path = NSTemporaryDirectory() + "forge-hash-\(UUID().uuidString).bin"
+        defer { try? FileManager.default.removeItem(atPath: path) }
+        var data = Data()
+        for i in 0..<100_000 {
+            data.append(UInt8(truncatingIfNeeded: i))
+        }
+        try data.write(to: URL(fileURLWithPath: path))
+
+        let fileHash = try FileHasher.xxh64(path: path)
+        XCTAssertEqual(fileHash, XXHash64.hexString(XXHash64.hash(data)))
+
+        let fileMD5 = try FileHasher.md5(path: path)
+        XCTAssertEqual(fileMD5, MD5.hexString(data))
+    }
+
+    func testMissingFileThrows() {
+        XCTAssertThrowsError(try FileHasher.xxh64(path: "/nonexistent/nope.bin"))
+    }
+}
diff --git a/Tests/ForgeOffloadTests/PlaceholderTests.swift b/Tests/ForgeOffloadTests/PlaceholderTests.swift
deleted file mode 100644
index 9f08b2b..0000000
--- a/Tests/ForgeOffloadTests/PlaceholderTests.swift
+++ /dev/null
@@ -1,5 +0,0 @@
-import XCTest
-
-final class ForgeOffloadPlaceholderTests: XCTestCase {
-    func testPlaceholder() { XCTAssertTrue(true) }
-}