diff --git a/README.md b/README.md
index 6163a26..55f20ac 100644
--- a/README.md
+++ b/README.md
@@ -284,6 +284,36 @@ The TTS module doesn't directly handle speech output. Rather, its methods return
 
 The SSML attributes `ssml:ph` and `ssml:alphabet` are supported. There's no support for PLS and CSS Speech.
 
+### Offline Dictionaries
+
+The `dict.js` module can be used to load dictd and StarDict dictionaries. Usage:
+
+```js
+import { StarDict } from './dict.js'
+import { inflate } from 'your inflate implementation'
+
+const { ifo, dz, idx, syn } = { /* `File` (or `Blob`) objects */ }
+const dict = new StarDict()
+await dict.loadIfo(ifo)
+await dict.loadDict(dz, inflate)
+await dict.loadIdx(idx)
+await dict.loadSyn(syn)
+
+// look up words
+const query = '...'
+await dictionary.lookup(query)
+await dictionary.synonyms(query)
+```
+
+Note that you must supply your own `inflate` function. Here is an example using [fflate](https://github.com/101arrowz/fflate):
+```js
+const inflate = data => new Promise(resolve => {
+    const inflate = new fflate.Inflate()
+    inflate.ondata = data => resolve(data)
+    inflate.push(data)
+})
+```
+
 ### OPDS
 
 The `opds.js` module can be used to implement OPDS clients. It can convert OPDS 1.x documents to OPDS 2.0:
diff --git a/dict.js b/dict.js
new file mode 100644
index 0000000..9afa7ea
--- /dev/null
+++ b/dict.js
@@ -0,0 +1,241 @@
+const decoder = new TextDecoder()
+const decode = decoder.decode.bind(decoder)
+
+const concatTypedArray = (a, b) => {
+    const result = new a.constructor(a.length + b.length)
+    result.set(a)
+    result.set(b, a.length)
+    return result
+}
+
+const strcmp = (a, b) => {
+    a = a.toLowerCase(), b = b.toLowerCase()
+    return a < b ? -1 : a > b ? 1 : 0
+}
+
+class DictZip {
+    #chlen
+    #chunks
+    #compressed
+    inflate
+    async load(file) {
+        const header = new DataView(await file.slice(0, 12).arrayBuffer())
+        if (header.getUint8(0) !== 31 || header.getUint8(1) !== 139
+        || header.getUint8(2) !== 8) throw new Error('Not a DictZip file')
+        const flg = header.getUint8(3)
+        if (!flg & 0b100) throw new Error('Missing FEXTRA flag')
+
+        const xlen = header.getUint16(10, true)
+        const extra = new DataView(await file.slice(12, 12 + xlen).arrayBuffer())
+        if (extra.getUint8(0) !== 82 || extra.getUint8(1) !== 65)
+            throw new Error('Subfield ID should be RA')
+        if (extra.getUint16(4, true) !== 1) throw new Error('Unsupported version')
+
+        this.#chlen = extra.getUint16(6, true)
+        const chcnt = extra.getUint16(8, true)
+        this.#chunks = []
+        for (let i = 0, chunkOffset = 0; i < chcnt; i++) {
+            const chunkSize = extra.getUint16(10 + 2 * i, true)
+            this.#chunks.push([chunkOffset, chunkSize])
+            chunkOffset = chunkOffset + chunkSize
+        }
+
+        // skip to compressed data
+        let offset = 12 + xlen
+        const max = Math.min(offset + 512, file.size)
+        const strArr = new Uint8Array(await file.slice(0, max).arrayBuffer())
+        if (flg & 0b1000) { // fname
+            const i = strArr.indexOf(0, offset)
+            if (i < 0) throw new Error('Header too long')
+            offset = i + 1
+        }
+        if (flg & 0b10000) { // fcomment
+            const i = strArr.indexOf(0, offset)
+            if (i < 0) throw new Error('Header too long')
+            offset = i + 1
+        }
+        if (flg & 0b10) offset += 2 // fhcrc
+        this.#compressed = file.slice(offset)
+    }
+    async read(offset, size) {
+        const chunks = this.#chunks
+        const startIndex = Math.trunc(offset / this.#chlen)
+        const endIndex = Math.trunc((offset + size) / this.#chlen)
+        const buf = await this.#compressed.slice(chunks[startIndex][0],
+            chunks[endIndex][0] + chunks[endIndex][1]).arrayBuffer()
+        let arr = new Uint8Array()
+        for (let pos = 0, i = startIndex; i <= endIndex; i++) {
+            const data = new Uint8Array(buf, pos, chunks[i][1])
+            arr = concatTypedArray(arr, await this.inflate(data))
+            pos += chunks[i][1]
+        }
+        const startOffset = offset - startIndex * this.#chlen
+        return arr.subarray(startOffset, startOffset + size)
+    }
+}
+
+class Index {
+    strcmp = strcmp
+    // binary search
+    bisect(query, start = 0, end = this.words.length - 1) {
+        if (end - start === 1) {
+            if (!this.strcmp(query, this.getWord(start))) return start
+            if (!this.strcmp(query, this.getWord(end))) return end
+            return null
+        }
+        const mid = Math.floor(start + (end - start) / 2)
+        const cmp = this.strcmp(query, this.getWord(mid))
+        if (cmp < 0) return this.bisect(query, start, mid)
+        if (cmp > 0) return this.bisect(query, mid, end)
+        return mid
+    }
+    // check for multiple definitions
+    checkAdjacent(query, i) {
+        if (i == null) return []
+        let j = i
+        const equals = i => {
+            const word = this.getWord(i)
+            return word ? this.strcmp(query, word) === 0 : false
+        }
+        while (equals(j - 1)) j--
+        let k = i
+        while (equals(k + 1)) k++
+        return j === k ? [i] : Array.from({ length: k + 1 - j }, (_, i) => j + i)
+    }
+    lookup(query) {
+        return this.checkAdjacent(query, this.bisect(query))
+    }
+}
+
+const decodeBase64Number = str => {
+    const { length } = str
+    let n = 0
+    for (let i = 0; i < length; i++) {
+        const c = str.charCodeAt(i)
+        n += (c === 43 ? 62     // "+"
+            : c === 47 ? 63     // "/"
+            : c < 58 ? c + 4    // 0-9 -> 52-61
+            : c < 91 ? c - 65   // A-Z -> 0-25
+            : c - 71            // a-z -> 26-51
+        ) * 64 ** (length - 1 - i)
+    }
+    return n
+}
+
+class DictdIndex extends Index {
+    getWord(i) {
+        return this.words[i]
+    }
+    async load(file) {
+        const words = []
+        const offsets = []
+        const sizes = []
+        for (const line of decode(await file.arrayBuffer()).split('\n')) {
+            const a = line.split('\t')
+            words.push(a[0])
+            offsets.push(decodeBase64Number(a[1]))
+            sizes.push(decodeBase64Number(a[2]))
+        }
+        this.words = words
+        this.offsets = offsets
+        this.sizes = sizes
+    }
+}
+
+export class DictdDict {
+    #dict = new DictZip()
+    #idx = new DictdIndex()
+    loadDict(file, inflate) {
+        this.#dict.inflate = inflate
+        return this.#dict.load(file)
+    }
+    async #readWord(i) {
+        const word = this.#idx.getWord(i)
+        const offset = this.#idx.offsets[i]
+        const size = this.#idx.sizes[i]
+        return { word, data: ['m', this.#dict.read(offset, size)] }
+    }
+    #readWords(arr) {
+        return Promise.all(arr.map(this.#readWord.bind(this)))
+    }
+    lookup(query) {
+        return this.#readWords(this.#idx.lookup(query))
+    }
+}
+
+class StarDictIndex extends Index {
+    isSyn
+    #arr
+    getWord(i) {
+        const word = this.words[i]
+        if (!word) return
+        return decode(this.#arr.subarray(word[0], word[1]))
+    }
+    async load(file) {
+        const { isSyn } = this
+        const buf = await file.arrayBuffer()
+        const arr = new Uint8Array(buf)
+        this.#arr = arr
+        const view = new DataView(buf)
+        const words = []
+        const offsets = []
+        const sizes = []
+        for (let i = 0; i < arr.length;) {
+            const newI = arr.subarray(0, i + 256).indexOf(0, i)
+            if (newI < 0) throw new Error('Word too big')
+            words.push([i, newI])
+            offsets.push(view.getUint32(newI + 1))
+            if (isSyn) i = newI + 5
+            else {
+                sizes.push(view.getUint32(newI + 5))
+                i = newI + 9
+            }
+        }
+        this.words = words
+        this.offsets = offsets
+        this.sizes = sizes
+    }
+}
+
+export class StarDict {
+    #dict = new DictZip()
+    #idx = new StarDictIndex()
+    #syn = Object.assign(new StarDictIndex(), { isSyn: true })
+    async loadIfo(file) {
+        const str = decode(await file.arrayBuffer())
+        this.ifo = Object.fromEntries(str.split('\n').map(line => {
+            const sep = line.indexOf('=')
+            if (sep < 0) return
+            return [line.slice(0, sep), line.slice(sep + 1)]
+        }).filter(x => x))
+    }
+    loadDict(file, inflate) {
+        this.#dict.inflate = inflate
+        return this.#dict.load(file)
+    }
+    loadIdx(file) {
+        return this.#idx.load(file)
+    }
+    loadSyn(file) {
+        if (file) return this.#syn.load(file)
+    }
+    async #readWord(i) {
+        const word = this.#idx.getWord(i)
+        const offset = this.#idx.offsets[i]
+        const size = this.#idx.sizes[i]
+        const data = await this.#dict.read(offset, size)
+        const seq = this.ifo.sametypesequence
+        if (!seq) throw new Error('TODO')
+        if (seq.length === 1) return { word, data: [[seq[0], data]] }
+        throw new Error('TODO')
+    }
+    #readWords(arr) {
+        return Promise.all(arr.map(this.#readWord.bind(this)))
+    }
+    lookup(query) {
+        return this.#readWords(this.#idx.lookup(query))
+    }
+    synonyms(query) {
+        return this.#readWords(this.#syn.lookup(query).map(i => this.#syn.offsets[i]))
+    }
+}