a small core for append-only log based programs
Clone or download
Latest commit c06ba77 Aug 19, 2018
Failed to load latest commit information.
README.md readme: mention substack! Aug 19, 2018
example.js feat: bind core and index May 15, 2018
index.js feat: support non-function api entries Aug 10, 2018
package-lock.json 2.3.0 Aug 10, 2018
package.json 2.3.0 Aug 10, 2018



a small core for append-only log based programs

A lot like flumedb, but using multifeed as an append-only log base, which is actually a set of append-only logs.

Pronounced "capricorn".


Experimental, but functional.


var kappa = require('kappa-core')
var memdb = require('memdb')

var core = kappa('./log', { valueEncoding: 'json' })
var idx = memdb()

var sum = 0

var sumview = {
  api: {
    get: function (core, cb) {
      this.ready(function () {
        cb(null, sum)
  map: function (msgs, next) {
    msgs.forEach(function (msg) {
      if (typeof msg.value === 'number') sum += msg.value

  // where to store and fetch the indexer's state (which log entries have been
  // processed so far)
  storeState: function (state, cb) { idx.put('state', state, cb) },
  fetchState: function (cb) { idx.get('state', cb) }

// the api will be mounted at core.api.sum
core.use('sum', sumview)

core.feed('default', function (err, feed) {
  feed.append(1, function (err) {
    core.api.sum.get(function (err, value) {
      console.log(value) // 1


var kappa = require('kappa-core')

var core = kappa(storage, opts)

Create a new kappa-core database.

  • storage is an instance of random-access-storage. If a string is given, random-access-file is used with the string as the filename.
  • Valid opts include:
    • valueEncoding: a string describing how the data will be encoded.

var feed = core.feed(name, cb)

Create or get a local writable feed called name. If it already existed, it is returned. A feed is an instance of hypercore.

core.use(name, view)

Install a view called name to the kappa-core instance. A view is an object of the form

  api: {
    someSyncFunction: function (core) { return ... },
    someAsyncFunction: function (core, cb) { process.nextTick(cb, ...) }

  map: function (msgs, next) {
    msgs.forEach(function (msg) {
      // ...

  fetchState: function (cb) { ... },
  storeState: function (state, cb) { ... }

The kappa-core instance core is always is bound to this in all of the api functions you define.

The {fetch,store}State functions are optional: they tell the view where to store its state information about what log entries have been indexed thus far. If not passed in, they will be stored in memory (i.e. reprocessed on each fresh run of the program). You can use any backend you want (like leveldb) to store the Buffer object state.

There are also the following optional opts:

  • inedxed: a function to run whenever a new batch of messages have been indexed & written to storage. Receives an array of messages.

core.ready(viewNames, cb)

Wait until all views named by viewNames are caught up. e.g.

// one
core.ready('sum', function () { ... })

// or several
core.ready(['kv', 'refs', 'spatial'], function () { ... })

If viewNames is [] or not included, all views will be waited on.


Create a duplex replication stream. opts are passed in to multifeed's API of the same name.


With npm installed, run

$ npm install kappa-core

Useful view modules

Here are some useful modules that play well with kappa-core for building views:


flumedb presents an ideal small core API for an append-only log: append new data, and build (versioned) views over it. kappa-core copies this gleefully, but with two major differences:

  1. hypercore is used for feed (append-only log) storage
  2. views are built in out-of-order sequence

hypercore provides some very useful superpowers:

  1. all data is cryptographically associated with a writer's public key
  2. partial replication: parts of feeds can be selectively sync'd between peers, instead of all-or-nothing, without loss of cryptographic integrity

Building views in arbitrary sequence is more challenging than when order is known to be topographic, but confers some benefits:

  1. most programs are only interested in the latest values of data; the long tail of history can be traversed asynchronously at leisure after the tips of the feeds are processed
  2. the views are tolerant of partially available data. Many of the modules listed in the section below depend on topographic completeness: all entries referenced by an entry must be present for indexes to function. This makes things like the equivalent to a shallow clone (think git), where a small subset of the full dataset can be used and built on without breaking anything.


kappa-core is built atop ideas from a huge body of others' work:

Further Reading