diff --git a/tokenization/FST.js b/tokenization/FST.js
new file mode 100644
index 00000000..c9e5f12d
--- /dev/null
+++ b/tokenization/FST.js
@@ -0,0 +1,150 @@
+/**
+ * FST - Finite-state transducer data structure
+ *
+ * A graph structure which is very efficient for querying prefix (or suffix) matches.
+ *
+ * see: https://www.elastic.co/blog/you-complete-me
+ */
+
+const Graph = require('./Graph')
+const ETX = String.fromCharCode(3) // end-of-text
+
+class FST {
+  constructor () {
+    this.head = new Graph()
+    this.tail = new Graph()
+  }
+
+  // add a new token to the index
+  add (token) {
+    if (this.has(token)) { return }
+    this._index(this._split(token))
+  }
+
+  // remove token from index
+  delete (token) {
+    this._deindex(this._split(token))
+  }
+
+  // index contains token
+  has (token) {
+    let node = this._walk(this.head, '>', this._split(token))
+    return !!node && !!node.length(`>${ETX}`)
+  }
+
+  // index contains a token with this prefix
+  hasPrefix (prefix) {
+    let node = this._walk(this.head, '>', this._split(prefix))
+    return !!node && (!node.findOne(`>${ETX}`) || node.findOne('_meta')['>count'] > 1)
+  }
+
+  // index contains a token with this suffix
+  hasSuffix (suffix) {
+    let node = this._walk(this.tail, '<', this._split(suffix).reverse())
+    return !!node && (!node.findOne(`<${ETX}`) || node.findOne('_meta')['<count'] > 1)
+  }
+
+  // split token in to characters
+  _split (token) { return (token || '').split('') }
+
+  // walk the graph & return the last node
+  _walk (parent, direction, chars, create, each) {
+    let DIR = (direction === '<') ? '<' : '>'
+    let END = (DIR === '<') ? this.head : this.tail
+    let LAST = (chars.length - 1)
+    for (let i = 0; i < chars.length; i++) {
+      let child = parent.findOne(`${DIR}${chars[i]}`)
+      if (create === true) {
+        // create new graph node
+        if (!child) {
+          child = new Graph()
+          child.add('_meta', { '>count': 0, '<count': 0 })
+          parent.add(`${DIR}${chars[i]}`, child)
+        }
+
+        // update count
+        child.findOne('_meta')[`${DIR}count`]++
+
+        // create final edge
+        if (i === LAST && !child.length(`${DIR}${ETX}`)) {
+          child.add(`${DIR}${ETX}`, END)
+        }
+      }
+      if (!child) { return child }
+      if (typeof each === 'function') {
+        each(child, parent, chars[i])
+      }
+      parent = child
+    }
+
+    return parent
+  }
+
+  /**
+   * dump FST to stderr, eg:
+   * [3] >e>x
+   * [3] >e>x>a
+   * [2] >e>x>a>m
+   * [2] >e>x>a>m>p
+   * [2] >e>x>a>m>p>l
+   * [2] >e>x>a>m>p>l>e
+   * [2] >e>x>a>m>p>l>e>s
+   * [3] >e>x>c
+   * [1] >e>x>c>e
+   * [1] >e>x>c>e>s
+   * [1] >e>x>c>e>s>s
+   */
+  print (node, direction) {
+    this._recurse(node, direction, (path, count) => {
+      console.error(`[${count}] ${path}`)
+    })
+  }
+
+  _recurse (node, direction, each, prefix) {
+    if (!direction) { direction = '>' }
+    if (!prefix) { prefix = '' }
+    for (let key in node.edges) {
+      if (key[0] !== direction) { continue }
+      if (!node.length(key)) { continue }
+      if (node !== this.head && node !== this.tail) {
+        let count = node.findOne('_meta')[`${direction}count`]
+        each(prefix, count)
+      } else if (prefix.length > 0) { return }
+      this._recurse(node.findOne(key), direction, each, prefix + key)
+    }
+  }
+
+  // walk the graph & add characters to graph
+  _index (chars) {
+    this._walk(this.head, '>', chars, true)
+    this._walk(this.tail, '<', chars.reverse(), true)
+  }
+
+  // walk the graph & remove characters from graph
+  _deindex (chars) {
+    // left-to-right
+    let node = this._walk(this.head, '>', chars)
+    if (node && node.remove(`>${ETX}`, this.tail)) {
+      this._walk(this.head, '>', chars, false, (child, parent, char) => {
+        let meta = child.findOne('_meta')
+        if (meta && --meta['>count'] < 1) {
+          parent.remove(`>${char}`, child)
+        }
+      })
+    }
+
+    // right-to-left
+    let reversed = chars.slice().reverse()
+    node = this._walk(this.tail, '<', reversed)
+    if (node && node.remove(`<${ETX}`, this.head)) {
+      this._walk(this.tail, '<', reversed, false, (child, parent, char) => {
+        let meta = child.findOne('_meta')
+        if (meta && --meta['<count'] < 1) {
+          parent.remove(`<${char}`, child)
+        }
+      })
+    }
+  }
+}
+
+module.exports = FST
diff --git a/tokenization/FST.test.js b/tokenization/FST.test.js
new file mode 100644
index 00000000..c9b4c2e1
--- /dev/null
+++ b/tokenization/FST.test.js
@@ -0,0 +1,269 @@
+const FST = require('./FST')
+const ETX = String.fromCharCode(3) // end-of-text
+
+module.exports.tests = {}
+
+module.exports.tests.constructor = (test) => {
+  test('constructor', (t) => {
+    let fst = new FST()
+    t.deepEquals(fst.head.constructor.name, 'Graph')
+    t.deepEquals(fst.tail.constructor.name, 'Graph')
+    t.deepEquals(Object.keys(fst.head.edges).length, 0)
+    t.deepEquals(Object.keys(fst.tail.edges).length, 0)
+    t.end()
+  })
+}
+
+module.exports.tests.graph = (test) => {
+  test('graph', (t) => {
+    let fst = new FST()
+    fst.add('example')
+
+    // left-to-right
+    t.true(
+      fst.head
+        .findOne('>e')
+        .findOne('>x')
+        .findOne('>a')
+        .findOne('>m')
+        .findOne('>p')
+        .findOne('>l')
+        .findOne('>e')
+        .findOne(`>${ETX}`)
+    )
+
+    // right-to-left
+    t.true(
+      fst.tail
+        .findOne('<e')
+        .findOne('<l')
+        .findOne('<p')
+        .findOne('<m')
+        .findOne('<a')
+        .findOne('<x')
+        .findOne('<e')
+        .findOne(`<${ETX}`)
+    )
+
+    t.end()
+  })
+}
+
+module.exports.tests.has = (test) => {
+  test('has', (t) => {
+    let fst = new FST()
+
+    fst.add('example')
+    fst.add('exam')
+    t.true(fst.has('example'))
+    t.true(fst.has('exam'))
+
+    t.end()
+  })
+}
+
+module.exports.tests._meta = (test) => {
+  test('_meta', (t) => {
+    let fst = new FST()
+
+    // add term
+    fst.add('example')
+
+    // meta-data
+    t.equals(Object.keys(fst.head.edges).length, 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('e')).findOne('_meta')['>count'], 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('ex')).findOne('_meta')['>count'], 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('exa')).findOne('_meta')['>count'], 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('exam')).findOne('_meta')['>count'], 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('examp')).findOne('_meta')['>count'], 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('exampl')).findOne('_meta')['>count'], 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('example')).findOne('_meta')['>count'], 1)
+
+    // duplicate term
+    fst.add('example')
+
+    // meta-data
+    t.equals(Object.keys(fst.head.edges).length, 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('e')).findOne('_meta')['>count'], 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('ex')).findOne('_meta')['>count'], 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('exa')).findOne('_meta')['>count'], 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('exam')).findOne('_meta')['>count'], 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('examp')).findOne('_meta')['>count'], 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('exampl')).findOne('_meta')['>count'], 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('example')).findOne('_meta')['>count'], 1)
+
+    // add term
+    fst.add('excess')
+
+    t.equals(Object.keys(fst.head.edges).length, 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('e')).findOne('_meta')['>count'], 2)
+    t.equals(fst._walk(fst.head, '>', fst._split('ex')).findOne('_meta')['>count'], 2)
+    t.equals(fst._walk(fst.head, '>', fst._split('exa')).findOne('_meta')['>count'], 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('exam')).findOne('_meta')['>count'], 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('examp')).findOne('_meta')['>count'], 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('exampl')).findOne('_meta')['>count'], 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('example')).findOne('_meta')['>count'], 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('exc')).findOne('_meta')['>count'], 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('exce')).findOne('_meta')['>count'], 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('exces')).findOne('_meta')['>count'], 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('excess')).findOne('_meta')['>count'], 1)
+
+    // add term
+    fst.add('examples')
+
+    // meta-data
+    t.equals(Object.keys(fst.head.edges).length, 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('e')).findOne('_meta')['>count'], 3)
+    t.equals(fst._walk(fst.head, '>', fst._split('ex')).findOne('_meta')['>count'], 3)
+    t.equals(fst._walk(fst.head, '>', fst._split('exa')).findOne('_meta')['>count'], 2)
+    t.equals(fst._walk(fst.head, '>', fst._split('exam')).findOne('_meta')['>count'], 2)
+    t.equals(fst._walk(fst.head, '>', fst._split('examp')).findOne('_meta')['>count'], 2)
+    t.equals(fst._walk(fst.head, '>', fst._split('exampl')).findOne('_meta')['>count'], 2)
+    t.equals(fst._walk(fst.head, '>', fst._split('example')).findOne('_meta')['>count'], 2)
+    t.equals(fst._walk(fst.head, '>', fst._split('examples')).findOne('_meta')['>count'], 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('exc')).findOne('_meta')['>count'], 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('exce')).findOne('_meta')['>count'], 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('exces')).findOne('_meta')['>count'], 1)
+    t.equals(fst._walk(fst.head, '>', fst._split('excess')).findOne('_meta')['>count'], 1)
+
+    t.end()
+  })
+}
+
+module.exports.tests.prefix = (test) => {
+  test('prefix', (t) => {
+    let fst = new FST()
+
+    fst.add('example')
+    t.true(fst.hasPrefix('e'))
+    t.true(fst.hasPrefix('ex'))
+    t.true(fst.hasPrefix('exa'))
+    t.true(fst.hasPrefix('exam'))
+    t.true(fst.hasPrefix('examp'))
+    t.true(fst.hasPrefix('exampl'))
+    t.false(fst.hasPrefix('example'))
+
+    fst.add('example')
+    t.true(fst.hasPrefix('e'))
+    t.true(fst.hasPrefix('ex'))
+    t.true(fst.hasPrefix('exa'))
+    t.true(fst.hasPrefix('exam'))
+    t.true(fst.hasPrefix('examp'))
+    t.true(fst.hasPrefix('exampl'))
+    t.false(fst.hasPrefix('example'))
+
+    fst.add('example@')
+    t.true(fst.hasPrefix('e'))
+    t.true(fst.hasPrefix('ex'))
+    t.true(fst.hasPrefix('exa'))
+    t.true(fst.hasPrefix('exam'))
+    t.true(fst.hasPrefix('examp'))
+    t.true(fst.hasPrefix('exampl'))
+    t.true(fst.hasPrefix('example'))
+    t.false(fst.hasPrefix('example@'))
+
+    t.end()
+  })
+}
+
+module.exports.tests.suffix = (test) => {
+  test('suffix', (t) => {
+    let fst = new FST()
+
+    fst.add('example')
+    t.true(fst.hasSuffix('e'))
+    t.true(fst.hasSuffix('le'))
+    t.true(fst.hasSuffix('ple'))
+    t.true(fst.hasSuffix('mple'))
+    t.true(fst.hasSuffix('ample'))
+    t.true(fst.hasSuffix('xample'))
+    t.false(fst.hasSuffix('example'))
+
+    fst.add('example')
+    t.true(fst.hasSuffix('e'))
+    t.true(fst.hasSuffix('le'))
+    t.true(fst.hasSuffix('ple'))
+    t.true(fst.hasSuffix('mple'))
+    t.true(fst.hasSuffix('ample'))
+    t.true(fst.hasSuffix('xample'))
+    t.false(fst.hasSuffix('example'))
+
+    fst.add('@example')
+    t.true(fst.hasSuffix('e'))
+    t.true(fst.hasSuffix('le'))
+    t.true(fst.hasSuffix('ple'))
+    t.true(fst.hasSuffix('mple'))
+    t.true(fst.hasSuffix('ample'))
+    t.true(fst.hasSuffix('xample'))
+    t.true(fst.hasSuffix('example'))
+    t.false(fst.hasSuffix('@example'))
+
+    t.end()
+  })
+}
+
+module.exports.tests.delete = (test) => {
+  test('delete', (t) => {
+    let fst = new FST()
+    t.deepEquals(Object.keys(fst.head.edges), [])
+    t.deepEquals(Object.keys(fst.tail.edges), [])
+
+    fst.add('example')
+    t.deepEquals(Object.keys(fst.head.edges), ['>e'])
+    t.deepEquals(Object.keys(fst.tail.edges), ['<e'])
+
+    fst.add('exam')
+    t.deepEquals(Object.keys(fst.head.edges), ['>e'])
+    t.deepEquals(Object.keys(fst.tail.edges), ['<e', '<m'])
+
+    t.true(fst.has('example'))
+    t.true(fst.has('exam'))
+
+    t.true(fst.hasPrefix('e'))
+    t.true(fst.hasPrefix('ex'))
+    t.true(fst.hasPrefix('exa'))
+    t.true(fst.hasPrefix('exam'))
+    t.true(fst.hasPrefix('examp'))
+    t.true(fst.hasPrefix('exampl'))
+
+    fst.delete('example')
+    t.deepEquals(Object.keys(fst.head.edges), ['>e'])
+    t.deepEquals(Object.keys(fst.tail.edges), ['<m'])
+
+    t.false(fst.has('example'))
+    t.true(fst.has('exam'))
+
+    t.true(fst.hasPrefix('e'))
+    t.true(fst.hasPrefix('ex'))
+    t.true(fst.hasPrefix('exa'))
+    t.false(fst.hasPrefix('exam'))
+    t.false(fst.hasPrefix('examp'))
+    t.false(fst.hasPrefix('exampl'))
+
+    fst.delete('exam')
+    t.deepEquals(Object.keys(fst.head.edges), [])
+    t.deepEquals(Object.keys(fst.tail.edges), [])
+
+    t.false(fst.has('example'))
+    t.false(fst.has('exam'))
+
+    t.false(fst.hasPrefix('e'))
+    t.false(fst.hasPrefix('ex'))
+    t.false(fst.hasPrefix('exa'))
+    t.false(fst.hasPrefix('exam'))
+    t.false(fst.hasPrefix('examp'))
+    t.false(fst.hasPrefix('exampl'))
+
+    t.end()
+  })
+}
+
+module.exports.all = (tape, common) => {
+  function test (name, testFunction) {
+    return tape(`FST: ${name}`, testFunction)
+  }
+
+  for (var testCase in module.exports.tests) {
+    module.exports.tests[testCase](test, common)
+  }
+}
diff --git a/tokenization/Graph.js b/tokenization/Graph.js
index 66cc24e3..3ca548e9 100644
--- a/tokenization/Graph.js
+++ b/tokenization/Graph.js
@@ -16,6 +16,10 @@ class Graph {
     if (!this.edges[relationship]) { return false }
     let len = this.edges[relationship].length
     this.edges[relationship] = this.edges[relationship].filter(n => n !== node)
+    if (!this.edges[relationship].length) {
+      delete this.edges[relationship]
+      return true
+    }
     return this.edges[relationship].length !== len
   }
 
diff --git a/tokenization/Graph.test.js b/tokenization/Graph.test.js
index 223b16dd..c9a1b50a 100644
--- a/tokenization/Graph.test.js
+++ b/tokenization/Graph.test.js
@@ -57,7 +57,7 @@ module.exports.tests.remove = (test) => {
 
     // remove node from 'bar'
     let ok3 = graph.remove('bar', node1)
-    t.equal(graph.edges.bar.length, 0)
+    t.false(graph.edges.bar)
     t.true(ok3)
 
     // remove node from 'baz'