Automated switchboard using voice recognition and 46elks calls API

Voice recognition / Speech recognition / Speech to text is with the help of could services such as Azue , Bluemix , Google Cloud , etc. is gradually becoming a realiable way to get user input. And language support is also improving including support to understand different dialects like the Swedish dialect from Skåne (Scania), that dialect is even hard for some swedes to understand. I created a simple use case using our the Voice Call API. And the result was pretty good despite the limited sound quality of the Cellphone network.

Call flow

Here is an overview of how the systems are connected.

  1. Call arrives from someone.
  2. A welcome-sound file is played to the caller.
  3. A recording is then started.
  4. When 3 seconds of silence is detected, or when a digit is pressed on the phone, the recording stops.
  5. Once the recording is complete a request is sent to the set endpoint informing it that the recording is now available for download. At this point it’s possible to start the transcription of the recording.
  6. In parallel with the transcription, it is possible to make actions in the call, e.g. playing a sound saying - “wait a second”.
  7. The next step is to check the text content and take an appropriate action.
  8. In this example I connect the call to the correct number depending on the name in the recording.

Incoming call - Play Welcome - Record - Play Ok - Next - Connect to phone


To handle this I created two endpoints.


This is the endpoint that will be informed when new sound files are available for a download in the 46elks API in order to send it to the speech recognition API.


This it the endpoint that will be informed when the caller is ready to be forwarded to the telephone requested. This endpoint needs to wait for the recording to be transcribed.


I created 4 sound files:

Parts on of the process

First step is to add a start JSON to voice_start on the number for the incoming call.

Initial voice start

The call starts with the welcome.wav, after that the sound is recorded and then ok_wait.wav is played and lastly the request to the /whatnow endpoint is made.

voice_start on number:

  "play": "",
  "next": {
    "record": "",
    "next": {
      "play": "",
      "next": ""

Handle recording

The recording the handled by downloading the sound file and then converting it into a BASE64 string as required by the voice recognition API.

First part of recording endpoint

// Set auth header.
$opts = array(
    'http' => array(
        'method' => 'GET',
        'header'  => "Authorization: Basic ". 
        'timeout' => 180
$context = stream_context_create($opts);

// Download sound file content.
$sound = file_get_contents($_POST['wav'], false, $context);
$sound = base64_encode($sound);

Ask Speech API for text

When the BASE64 string is available the request to the Google Speech API is made. The reason for using the Google Speech API in this example is that is supports the format of the sound files as is received from the 46elks API. And also Swedish along with lots of other languages is supported.

Second part of recording endpoint

$apirequest = array(
  "config"=> array(
    "encoding"=> "LINEAR16",
    "sampleRate"=> 8000,
    "languageCode"=> "sv-SE",
    "speechContext" => array ( 
      "phrases" => array("call","martin","johannes")
  "audio" => array(

$data_string = json_encode($apirequest);

$ch = curl_init('<api-key>');
curl_setopt($ch, CURLOPT_CUSTOMREQUEST, "POST");
curl_setopt($ch, CURLOPT_POSTFIELDS, $data_string);
curl_setopt($ch, CURLOPT_RETURNTRANSFER, true);
curl_setopt($ch, CURLOPT_HTTPHEADER, array(
    'Content-Type: application/json',
    'Content-Length: ' . strlen($data_string))

$text = curl_exec($ch);

Handle the text from the API

The API will then reply with different options of what the sound could mean as text. A simple solution is to select one with the highest accuracy. And then store that text in a local file, database, queue, etc.

Third part of recording endpoint

$texts = json_decode($text,true);

$accuracy = 0;
$besttext = '';

foreach($texts['results'][0]['alternatives'] as $alternative{
    if($accuracy < $alternative['confidence']){
      $besttext = $alternative['transcript'];


Check that the recording is not mute

If the recording was completely quiet there will not be any recording data, in this case it would be useful to ask the user the repeat the request.

First part of /whatnow endpoint

$opts = array(
    'http' => array(
      'method' => 'GET',
      'header'  => "Authorization: Basic ".
                   base64_encode('<api-username>:<api-password>'). "\r\n",
      'timeout' => 180

$context = stream_context_create($opts);

// Get call information:
$calldata = file_get_contents($_POST['callid'], false, $context);
$calldata = json_decode($calldata); 

$latestreccordresult = "";
foreach($calldata['actions'] as $action){
        $latestreccordresult = $action['result'];
if($latestreccordresult ==! "ok"){
print <<<END
  "play": "",
  "next": {
    "record": "",
    "next": {
      "play": "",
      "next": ""

Wait for recording to be transcribed then handle text.

The transcription may not be fast enough, so some time might be needed for the file to be created. And then take action. In this example I waited for 13 seconds before concluding that the transaction failed.

Second part of /whatnow endpoint

for($i = 0; $i < 13; $i++){
        $text = file_get_contents('calls/'.$_POST['callid']);

        $to = False;
        elseif (stristr($text,'johannes')){
             print <<<END
                  "connect": "{$to}",
                  "bussy": {
                    "play": "",
                    "next": {
                        "record": "",
                        "next": {
                            "play": "",
                            "next": ""

If all fails say sorry try again.

And if all else fails ask the user for input again. Simply play the sound file to the user and make a new recording request.

Third part of /whatnow endpoint

print <<<END
  "play": "",
  "next": {
    "record": "",
    "next": {
      "play": "",
      "next": ""

Written 2017-06-30 by Martin Harari Thuresson

The switch to Mondido

In 2016 46elks decided to implement Mondido as a fallback payment gateway.

The idéa was to use Mondido if there were issues with the primary payment gateway. This was why the integration was started. The reason we selected Mondido as fallback gateway was:

  1. No fixed fees.

  2. Seemed simple to implement.

  3. We did know about them from the SUP46 community.

However after running our primary gateway for a while we did notice some horrible problems:

  1. They were unable to prevent the majority of payments with stolen cards and we had to cover the cost of sent messages with stolen credits by refunding the payments.

  2. The payment reports for the book-keeping where very hard to handle resulting in lots of work for the book-keeper.

  3. The primary gateway needed a complex SDK to work in a simple way.

During an error with an update of the SDK, we did an emergency switch to Mondido. Since the switch we have been using Mondido, and we are very happy with that.


The 3D Secure ( Verified by Visa , MastercardⓇSecureCode™ , American Express SafekeyⓇ , and JCB J/Secure™ ) provides the safety we need, since the number of stolen cards is reduced close to zero, and even when there is a stolen card we do not have to cover the cost. For us stolen cards are extra problematic since the messages sent are often phishing SMS or spam. This could have caused the operators to block our connection to the telecom network in the long run.

Real contract

With the original gateway we did have one contract that included PSP and card acquirer in one, this was easy to get but left us with very little control. The acquirer we have with Mondido does allow us to negotiate the contract for example the number of days we have to wait for the money. And the acquirer ( Teller ) is huge and even if we had to wait for the payment for twenty days, (in reality we wait for as short as two days) the economic risk is close to zero. Teller also gives simple reports for book keeping. This saves us at least 4 hours of bookkeeping monthly.

Best entry level price

When we decided to check the price details for Mondido we noticed that the price was extremely appealing. Even better than our primary gateway. Mindido does give different price levels for expensive cards like AMEX (1,9%-3%) and low cost for cards like Visa/Mastercard ( 1,89% ), instead of giving a high start cost percentage for all cards.

Great API

When implementing the API some other great benefits were found:


The payment system adds extra safety for us because:

  1. Only our Backend can initiate a payment process by issuing a payment order-ID. This way even payments failing anywhere in the process can be tracked and completed. This way we can follow failed attempts thus identify and block fraudulent accounts.

  2. We never store the card details not even a token. The data is sent directly from the page to the Mondido backend. Even for the stored cards we can just query Mondido for the token when needed, all we need to do is keep a secure API password.

  3. SSL with Extended Validation is used by Mondido, this can be used by payment window to ensure the user that the card details are handled by Mondido, if needed. Or you may even have a sales page with http (I do not recommend it but you could) or have a page with Let’s Encrypt for the page and hosted window for the payments.


It was simple to implement the API, there were SDK s available but using our own framework was equally simple. The API is closely integrated with our system and therefore it is a low amount for code and in turn less bugs. The endpoints are easy to understand. However one thing that was hard to understand was how to use the the filter functionality when querying data from the API.

Costumer with reference

We do supply our user id to the payment, this makes it easy to follow payments for a single user as well as a simple way to verify payments. When we created the integration we did think that this was a problem because adding the customer ID required 10-20 lines of extra code. But we are very happy that we did implement it because the benefits are worth it! It is possible to skip this by simply using the same ID all the time, but don´t, you would probably regret it.

One line code for storage of card

When we wanted to add recurring payments we spent quite a few hours planning how to implement it with the old payment gateway, and it became so cumbersome that we never really completed this task. With Mondido it was just one line of code:

<input class="storecard" type="checkbox" name="store_card" value="true">Store my card for future payments.

Resulting in this in the 46elks Dashboard:

The card is stored on the user ID and can be accessed and used in the API.

JS or Hosted page

One benefit we have not completely taken an advantage of is that there is a JS and a hosted page option for payment, this means that we could offer SMS payment for example by simply sending a link to a hosted payment in Mondido. We are only using the JS version this is great for giving the user the on page experience. But what is great with the in hosted page is that you can change everything on it. We have not used these options to a huge extent, but I'm sure there is lots of innovation to be done with these options.

More EU options available

Mondido is based in EU (Sweden) and understands that not all is about Credit Cards, so they also offer invoice payment, etc. They also support local payment methods like Swish. This is very interesting for us since it lowers the barrier to get started with our platform.

International support

The payments from countries around the world work great as well. We were not sure that this would be the case. But our worries were proven wrong and international credit cards have been working very well.

They do not market well for developers

For us the major benefits of using Mondido do not really match their primary marketing slogans, which are:

  1. We help you convert more visitors into paying customers.

  2. Increase Sales and lower your costs with Mondido.

  3. Mondido offers card fees of 1,89 SEK + 1.89% per payment.

  4. A/B testing support.

These are not very interesting things for a developer. I do think the major benefits with Mondido from our experience have been these:

  1. Mondido allows you to adapt your user experience in every detail.

  2. Mondido is simple to implement, keeping the integration cost low.

  3. Mondido does not have a “one size fits all” contract but a base one that is adapted to the customer needs along the way.

Written 2017-02-15 by Martin Harari Thuresson

Trainhack 2016

Behind the scenes the full story

How did this happen?

I work at a company called 46elks, one part of my job is to make events that give us the opportunity to show developers that we exist. I learned about 46elks in 2013 during an event called Travelhack, organised by Trafiklab . In the same event I also got to know the company Oxyfi.

The idea

In december 2014 I joined the 46elks, and in the spring 2015 we started to discuss good Hackathon themes. An idea to have a hackathon on a train was voiced by Andreas Kron. I contacted Oxyfi, who luckily were equally excited about the idea! I also talked to the train companies and Blå Tåget was able to help us. Sadly in 2015 there was not enough time to do the Trainhack .

The date was set

In the beginning of 2016 we felt that the time was right to push for a Trainhack 2016! I started contacted the Trafiklab , Oxyfi and Blå Tåget . The date was set to 2016-08-25 and the exact time (initially) was 08:55 because this was a date when the special conference wagon of Blå Tåget was available at a price that was safe enough even in case the event did not attract anyone.

The event is taking shape

The decision to make the event a public transit event was easy. We would depart from Gothenburg and go all the way to Uppsala to have more time on the train and also be able to use the 46elks office space upon arrival, in case we did not find a more suitable space. The food arrangement would be Breakfast and fika on the train and lunch upon arrival (once already in Uppsala)

Public launch

We decided to go public with the event before the summer, and the same day Dag König from Microsoft signed up and asked if they could be a part of the Trainhack . Some more private signups followed, diverse in ages and backgrounds. Some weeks later I found out about a company called eTraveli that was into community events as well. So I decided to ask them if they were interested in sending a team to the event. They responded with overwhelming support, they offered not only to send their team, but also organize the Uppsala end of the event. Everything felt perfect!

The surprise

The next day, we received information that Blå Tåget lost their permission to run trains. This was not encouraging at all, I halted the work of pushing the event on different channels. And started to get information from Blå Tåget , who promised that they would try to get the permission back. And luckily some days later they did it, great! However the timetable was taken from them and some of the departure times were given to another operator. The process of getting the times back was long but after some weeks the times were secured again. I started to push for the event again. Read more in GP and Di .

The 2nd surprise

Two weeks before the event the court took the permission from Blå Tåget again! And now We were really short on time And again great people at Blå Tåget did get another operator Tågab to agree to to do the job instead. Now the only thing left was transfering the time from Blå Tåget to Tågab with no virtually time to spare! The application was sent to the train traffic control officer. Read more in GP and Di

Great news for the open transit APIs

After all the setbacks the luck was on our side again and the ride started to seem less and less bumpy . The first good news was that Oxyfi did manage to convince the train companies Norrtåg and Värmlandstrafik to share their GPS data with a Push API. It’s super useful when it comes to improving the delayed information for the drivers and the travelers. The second good news was that eTraveli gave access to a price API for air travel so it would be possible for the developers to make an Air+Rail price-comparison solution.

The 3rd surprize

After 4 days -2 days before the event- I got an email from Blå Tåget , that we did still not have any timetable! There was an option to wait for 2 more days, and hope for a timetable. If that would fail we would still have to pay for the train without being able to go anywhere. This was two days before the event so hotel and train tickets for traveling the other direction were already booked! I was in despair, and did all I could to get a timetable, but without success.

The flexible train companies

After talking to Blå Tåget , a possible solution was found, however, it would require a maneuver not commonly used in Sweden today. The solution was that the wagons would be connected to a normal Tågab train that would depart from Gothenburg at 06:15 (instead of 08:55 as planned in the beginning). And at the station of Kristinehamn, the wagons would be disconnected, left in the station and the other part of train would continue without us. About 40 minutes later another train would have a 7 minute long stop at the station. During this time our wagons would be connected to that train and it would depart to Stockholm. Blå Tåget and Tågab worked together and were able to make this possible! And there was hope again!


The new travel time resulted in new planning for the food on the train. But more importantly, it required that everyone attending the event needed to replan their mornings! So I stated sending information to all the participants about the new time. As an additional issue the pre-event place also cancelled, so I would also have to find the pre-event location, for the meetup, alongside the food and beverages. Lots and lots of things to plan. How were we going to arrange the food for the train? Nice people at the Creative Loop in Gothenburg provided a solution for the Pre-event space and supported us with how we could solve the issue with the food. As it turned out, the train would be standing on the platform during the night so we could leave the food on the train in the evening. So now we did have a plan that could work. Creative Loop also helped with sleeping arrangements for the Photographer from Nerdy Makers.

The heart attack

The day before the event, Blå Tåget calls me and says "We have a problem...". My heart rate jumped to 220 bpm and blood pressure to 230/180! No, I can't take one more setback! The message in the phonecall continued "...but we were able to solve it quickly. ". So my body dropped the heart rate and blood pressure. However the adrenaline in my body would stay at levels I never have felt before.

The missing prices

Due to all the issues with the train and the timetable, some other parts of the planning had lagged behind. One quite important part was to arrange prices for the winners of the event, so the day before the pre-event we only had one price and that was a ticket and boat trip to Junction hack in Helsinki (in November). I decided to buy several Raspberry Pi3s to have them as prices. I remembered that I knew someone from Webhallen so I decided to call him. As a result, Webhallen sponsored the event with some Raspberry Pis! This was great! Unfortunately, I was already on my way to Gothenburg but my colleague, who was sick and therefore stayed in Uppsala, could pick up the Raspberry Pis. Also, Trafiklab promised to arrange one more price for the team that would impress them the most.

The water boys

The water on the train cannot be used to make coffee because the quality cannot be secured, so all water needed for drinking and coffee had to be brought along by us. So me and Rob took all the empty PET bottles from my apartment in Gothenburg, filled them up and put them in an IKEA bag to carry them to the train in the evening.

The Pre-event

Then it was time to take care of the pre-event. Food needed to be arranged, thus, a quick run to the local supermarket for cheese pie was the solution. Later on, the travelers arrived and started to socialize. When the night was getting closer it was time to go and buy the food for the train. Everyone from the event decided to help so it turned out to be very easy to arrange.

The food odyssey

The Trainhack ers participating in the pre-event went together to the train station in Gothenburg, bringing the water bottles. There we met Joel, the train manager, who was going to help us along the way. We then went to Hemköp and started to buy things from the long shopping list. With the help of Matilda, from Microsoft, and Rob, from Nerdy Makers. The rest of the people stayed outside of the store and waited for us to return. After a while we were interrupted by a security officer. Apparently, the shop had been closed for 10 minutes already. So once more some adrenaline and pulse to finish shopping. When we came out of the closed store, the company had left because the guards said that the store was closed. As a consequence, we needed to carry the heavy shopping bags all the way to the train, where luckily Joel was ready to help us put the food in the refrigerators.

The late night planning.

After the food odyssey, me and Rob went back to Creative Loop , where we met Victoria also from 46elks. She was already working on the Trainhack logos to put on the doors and on the walls to make the event more visible. We also created the diplomas for the winners. After some hours we were finished and went to bed.


I woke up early in the morning, no alarm needed, still lots of adrenaline in the body apparently. I took the remaining things and walked to the station. At the station I did see the train still there. And to my surprise the hackers started to arrive, and they all seemed happy and ready to go. Some minutes later Joel arrived and opened the wagons for us. Oxyfi arrived to support with the on board wifi. The day before all the antennas did not fit on top of the conference wagon so Blå Tåget added one extra wagon for us, just like that! So we did get a 45 ton 3/4G antenna on the hock. The extra wagon was also quite good for comfort since traveling sideways for hours is not for everybody.


When the train departed from Gothenburg station, I could not believe it! I was in shock. My colleague Victoria reminded me to welcome everyone. At that point I understood that Trainhack 2016 was happening for real. During this time, Jakob from eTraveli , Viktoria from Oxyfi and Matilda from Microsoft , as well as Victoria from 46elks where arranging the breakfast buffet. With the buffet, the first Trainhack in the nordics was started. Victoria stated the group presentation where everyone presented themselves and what they would like to do.

The stolen train

The onboard atmosphere quickly turned to relaxed. I felt that everyone knew each other, or was at least eager to do so. Some minutes into the trip the train was delayed some minutes due to a slower train ahead. To regain the lost minutes the driver did the best he could to regain the time, i.e. full acceleration and full braking and max allowed speed all the time. So basically driving like he stole the train, so we all sat tight and looked for things to hold on everytime the train would start to slow down. This gave some more experience and motion sickness to some of the hackers.

The Wifi starts to fail

At some point our fast internet connection started to malfunction. Luckily Oxyfi was on the train and started to work on figuring out what was wrong. The issue was first believed to be due to one of the new antennas. But later on it seemed like the issue was more complicated, with TCP errors that prevented new sockets to be created. The issues were never really solved completely during the trip. However, Oxyfi got lots of information of how the system could be improved in future. Especially when it comes to lots of short SSH sessions to Github, at the same time as there is live streaming going on from the same network. So for next year the connection will be greatly improved. Later on they found errors that does only apear when the train is running att speed. The result of this is that internet onboard from Oxyfi for current trains and the future trains will become more stable.

SJ helps out

When the train was getting closer to Stockholm, I did remember we needed to get all the way to Uppsala. I needed to buy 24 tickets from Stockholm to Uppsala. However, that was not too easy to do on the SJ webpage. Luckily we did have a SJ developer on the train and he helped me with buying 24 tickets both quickly and easy. He also managed to get Lounge access for everyone while we were waiting for the train to Uppsala.

The countdown

After arriving to the eTraveli office in Uppsala, we were welcomed by Jakob Rödström who presented the place and informed everyone that the food was served. All Trainhack ers started eating and kept on hacking. It got pretty quiet while everyone was trying to finish their projects. The time flew by, and at 16:30 the presentations began.


Matilda from Microsoft was broadcasting the event, you can see more clips at Livearena

The video shows the presentations of the projects.

The winners can be found on the Trainhack site.

The future

For next year I hope we can have much fewer surprises. There have been some great things happening after the event, SJ showed an interest in being a co-organizer and all the organicers from this year have also been very eager to keep cooperating which makes me hope that everyone will join us next year as well. For the future prehaps a cooperation with other Trainhacks like could become a possibility. Next year the rout might be different perhaps Mamö - Stockholm - Gothenburg could be an interesting route. If you are interested you can signup for information for the next year at Trainhack .

Written 2016-10-15 by Martin Harari Thuresson

Share soundfiles from Dropbox to 46elks IVR

To play sound files to the users with the 46elks voice API you need to have somewhere to store your sound files. One solution is to store the files in your Dropbox and use the download link in the API.

Add a file to your dropbox folder

Open the folder in your Dropbox (where you want to store the sound file). Upload the sound file by selecting upload:

Share the soundfile

Share the soundfile by clicking on Share :

Create link

In the popup select Create link :

Copy the link

Then copy the link:

Change the link for direct download:

Change from: to by changing www to dl and removeing ?dl=0

You may allso change the: to by changing ?dl=0 to ?dl=1

Update your JSON in the dashboard:

The next step is to save the url in the JSON for the number:

You may use the file in any JSON for the Voice calls with 46elks.


Dropbox guide with parameter change

Similar way of doing it

Written 2016-07-11 by Martin Harari Thuresson

Hackathon 3.0

I feel concerned with the hackathon climate in Sweden right now. I keep hearing comments about Hackathons being flawed, while I don't think hackathons are perfect the reasons that are often mentioned to highlight the imperfections are in my opinion quite inaccurate.

The flaws that are often mentioned are:

I don't share this point of view. From my perspective these are not issues with Hackathons. For me a Hackathon is something different.

I believe that Hackathons are great because:

Hackathons and Solutions

A Hackathon does not always need to arrive at a final solution. Often it is enough to raise attention around a certain problem and develop plausible solutions that can be taken to the next level by the society.

Hackathon is a marathon of hacking, no one would ask a sports-star to solve an issue with low milk price in the framing business in Sweden. But they are often expected to inspire and be a role model for future generations. I believe that hackathons should be an inspiration for others rather than the final solution.

In my experience issues often are not technical, usually the problem is in our minds, here are some examples:

The trains in Sweden had a top speed of 90 Km/h early on but were limited to 40 km/h because politicians believed that the speed would be harmful.

Internet was not promoted by Telia (Swedish state telecom company) in the beginning because they did not believe it was the future, and created their own product instead. Tele2 did and was a part of swipnet, the beginning of IP networks in Sweden.

When 46elks was created it was extremely hard to have SMS, Voice & MMS over the internet sent to any application. And this was not mainly due to technical reasons - it took years to get the agreements with the operators but just months to get the service running.

My conclusion is that the issues often only require relatively small amount of work from developers. Instead the big part is often the bureaucracy in trying to explain to the general society what needs to be done in a given situation.

What could Hackathon 3.0 be?

Vinova will announce development funds for improving the Hackations in Sweden. Vinova is an agency that is created to support innovations is Sweden.

I'm also a teacher by profession and all I have learnt, when it comes to how the mind works, tells me that it is super important that hackathons are kept as an open event and are not turned into a forced way of creating a company.

I believe that Hackathons would have the best role in innovations by supporting the community through networking and providing a commonplace for learning new technology. And most importantly show how we can improve the environment we live in without slow bureaucratic systems limiting us

Written 2016-06-28 by Martin Harari Thuresson

Including telephone number in short links

Sending out SMS links is very common, and it is often interesting to know who clicked the link in the SMS. Since the web browser does not know what phone it is running on - a unique ID is needed for each SMS. It is possible to use an URL shortner like ( that makes links like this .

URL-Shortener limitations

However using URL-Shorteners has some limitations:

Encode number as short string

To solve this issues and still keep the number information - you may encode the number as an encoded string. This way a number like 723175800 can be coded as "SWcuk" and put in an example url like: http://sortu.rl/SWcuk

"SWcuk" can then be decoded to the original number with a simple self contained function. No access to a database is needed.



Example of implementation in python

alphabet = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz_-~=+/.!&"
charscount = len(alphabet)

def encode(n):
    n = int(n) 
    encoded = ''
    while n > 0:
        n, r = divmod(n, charscount)
        encoded = alphabet[r] + encoded
    return encoded

def decode(s):
    decoded = 0
    while len(s) > 0:
        decoded = decoded * charscount + alphabet.find(s[0])
        s = s[1:]
    return str(decoded)

Example of implementation in PHP

function encode($number){
    $alphabet = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz_-~=+/.!&";
    $charscount = strlen($alphabet);
    $number = intval($number);
    while($number > 0){
        $reminder = $number % $charscount;
        $number = intval($number/$charscount);
        $encoded = substr($alphabet,$reminder,1).$encoded;
    return $encoded;

function decode($string){
    $alphabet = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz_-~=+/.!&";
    $charscount = strlen($alphabet);
    $number = 0;
    while(strlen($string) > 0){
        $number = $number * $charscount + strpos($alphabet,substr($string,0,1));
        $string = substr($string,1);

    return $number;

Hashids generate short unique ids from integers in a similar way.

Written 2016-06-13 by Martin Harari Thuresson

Zapier and 46elks automated workflow

46elks is now public on Zapier the beta period is history. Zapier has also added some nice features like Delays and multiple actions on one trigger.

Great for holidays

I wanted to have a holiday some weeks ago, and i did not want to check my email during that time. However I still wanted to get an update every Friday by SMS regarding the subjects of the emails sent to me exclusively.

I connected the Zapier to my Gmail and connected it to a spreadsheet to have an overview of the messages that were relevant, there I made a filter to avoid getting emails where I was CC’ed or one for several recipients.

The next step was to add the great function of Delay, (a new Zapier native function). That allows you to select in my case "next friday at 15:00".

This is what the flow looks like:


My experience

I’ve learned that it is important to remember to charge the 46elks account with credit before going for a vacation.

I was surprised, how noticeable it was not to have an e-mail notification most of the time. 😎

Written 2016-05-30 by Martin Harari Thuresson

Google Apps Scripts - Error timetrigger not running

We at 46elks are supporting Kodcentrum with an application. They use Google Drive and Google Apps Scripts.

I created an endpoint that would be called by the Google Apps' Script every day. Everything was working fine when I was running the script manually, but when I tried to run it with a time-trigger it did not work.

I saw that the trigger did start the script by checking the script-log. But no request was made to the endpoint. Google reported that the service was unavailable whitout even trying to make a request.

I made sure that IPv4 & IPv6 was not the reason. After an extensive research I found the issue. Apparently Google service is using different systems to run manual and time-triggered tasks.

Finally I found that when saving the url to the endpoint in the script, Google server tries to read /robots.txt from the endpoint server. The endpoint server I created responded with 50X for that url. That caused Google to think the endpoint was down completely! And it would not try to make any request to any other path on the server.

To solve the issue I modified the server of the endpoint so that it would respond with 404 to all paths except the correct one, and updating the url in the Google Apps' Script solved the issue!

Written 2016-04-29 by Martin Harari Thuresson

IPSEC - Transport Mode - Debian

Do you need to secure the traffic between two nodes on a network you don´t trust? There are many VPN solutions as well as SSL and SSH. They all require settings for either the application or for the link itself to work.

IPSEC encrypts on Level 3 (in the IP stack) because of this the application does not need to know about the encryption.

There are two types of IPSEC modes, Tunnel and Transport. I use Tunnel mode in many cases for standard VPN. Transport mode is not used as often since it has more limited usecases.

I needed to secure traffic between two servers. The servers were initaly on a trusted network but needed to be moved to a network that I do not have full controll of. In my case the data sent over the link is occasionaly sensitive, so I needed a way to make it harder to listen to the link.

The nodes have direct IP access to each other. This example is with internal IPs but it works the same way with public IPs on both machines.

This guide will show you how to setup a static link, wich has a limited amount of packets that can bet sent. The maximum number of packets that can be sent over the link without a reset is 2^32. But if the kernels support 64 bit index the max amount of packets is 2^64.

This is how i did it

Install ipsec-tools

I started by installing ipsec-tools on the two machines.

apt-get install ipsec-tools

Flush IPSEC databases

I changed ipsec-tools.conf to flush the ipsec databases when running the setup:

# Flush the SAD and SPD


I created a conf file: /etc/ipsec-tools.d/transport-link1.conf on the host with ip Allso make sure that the data in this file is secret since it contains the crypto-keys.

# for host

# Put the ESP SAs in the SAD, 192 bit long keys
add esp 31031 -E rijndael-cbc
add esp 43241 -E rijndael-cbc

# Put the IPs in the Security Policy Database 
# to require the traffic to be encrypted
spdadd any -P out ipsec

spdadd any -P in ipsec

I created a conf file: /etc/ipsec-tools.d/transport-link1.conf on the host with ip as well.

Note that the only difference between the files is that the IPs in the "spdadd" lines are swapped

# for host

# Put the ESP SAs in the SAD, 192 bit long keys
add esp 31031 -E rijndael-cbc
add esp 43241 -E rijndael-cbc

# Put the IPs in the Security Policy Database 
# to require the traffic to be encrypted
spdadd any -P out ipsec

spdadd any -P in ipsec

To generate a 192 bit key I used this command:

echo "0x$(hexdump -e '1/1 "%02x"' '/dev/random' -n 24)"

Enable the link

I started the link by the setkey init script on both machines:

/etc/init.d/setkey start

Make sure that the link is working

I tested the link by pinging between the machines and using tcpdump to see the traffic, when the IPSEC is working you get ESP packets when it is not you get ICMP packets in tcp-dump.


When things go wrong you will notice by the fact that no normal traffic reaches the other host.

I use the tcpdump to listen for esp traffic:

tcpdump -i eth0 esp

To listen to ah traffic, not in the first conf example:

tcpdump -i eth0 ah

There are som basic issues that I have encountered.

  1. The data is sent but not encrypted. When the data is sent unencrypted it will not be accepted by the other host since esp/transport//require requires the traffic to be esp encrypted. The add to the SPD might be the same on the two hosts.

  2. The data is sent encrypted but the receiver does not decrypt it. The receiver might not be running IPSEC, or does miss the SPD record for incoming traffic. The receiver/sender have the wrong keys, the keys might be swapped in se SAD. The SAD records should be exactly the same on the two machines. The parameters 31031 & 43241 is called SPI and it is sent in the IPSEC header to tell the the receiver what key to use to decrypt the message, it the SPI is not the same the kernel will not be able find the correct key.

  3. Staring setkey gives incorrect key length error If you forget the ; at the end of the add statement the setkey gives the error incorrect key lenght. If you forget the 0x before the hex the key is interpreted as ASCI and will therefore also have the wrong lenght.

SAD Security Association Database

The SAD is storing the keys and methods to use for a certain combiantion of IPs. And a SPI in this case 31031, is the ID of the combination of encryptionmethod and key to use.

add esp 31031 -E rijndael-cbc 0xf86f44...;

The SAD is direction aware this means it understands what direction a packet is traveling, and thanks to this the add statement needs to be exactly the same both nodes.

SPD Security Policy Database

The SPD is storing when the kernel should encrypt the traffic. The SPD is not direction aware this means it has to be adopted to the node it is on.

That means that one node needs to say:

spdadd any -P in ipsec esp/transport//require;

And the other node needs to say:

spdadd any -P in ipsec esp/transport//require;

If it is mixed the data on one node will not be encrypted.


I found the setkey manpage very helpfull and easy to follow!



For the ESP I use the rijndael-cbc algorithm also know as AES.


When I want AH as well I use the hmac-sha256 algorithm.

Authentication Header

I did not use Authentication Header(AH), AH does not encrypt the content of the message but it makes sure that the sender is known and that the IP-header is untouched.

Encapsulating Security Payloads (ESP) used in my example, is used for encrypting the data in the packet.

This is a example of AH and ESP in combination:


# for host
# AH SAs using 160 bit long keys
add ah 33362 -A hmac-sha256
add ah 30759 -A hmac-sha256

# Put the ESP SAs in the DB, 192 bit long keys
add esp 31031 -E rijndael-cbc
add esp 43241 -E rijndael-cbc

# Put the IPs in the Security Policy Database 
# to require the traffic to be encrypted
# with esp and ah
spdadd any -P out ipsec

spdadd any -P in ipsec

This is a example of AH and ESP in combination for the other host:


# for host
# AH SAs using 160 bit long keys
add ah 33362 -A hmac-sha256
add ah 30759 -A hmac-sha256

# Put the ESP SAs in the DB, 192 bit long keys
add esp 31031 -E rijndael-cbc
add esp 43241 -E rijndael-cbc

# Put the IPs in the Security Policy Database 
# to require the traffic to be encrypted
# with esp and ah
spdadd any -P out ipsec

spdadd any -P in ipsec


Written 2016-04-06 by Martin Harari Thuresson

Frontends are not my thing

I'm a typical backend programmer. When I deliver my JSON struct or data stream from the server I think that everything is great. But for some reason a lot of users are not happy with that. I would like everyone to change and accept a JSON struct instead of using a GUI. This does not seems to be happening anytime soon.

Since it seems to take a long time, before I have turned the general public to my way of thinking I will have to make some kind of frontend for people to use. I wanted to make a APP to check the balance of my public transportation card ( SL ACCESS ). I did not have the energy nor the real skills to create a GUI for that. At Travelhack 2013 I got to know about 46elks , it was my salvation for this use case.

Instead of creating a complete UI I just connected the backend to incoming and outgoing SMS. The backend loads the account data from the SL webpage to get the balance of each card and sends it to the user. It may look like this:


To make it possible for the backend to access the SL data, the user credentials for SL are needed. This is why the user need to set the credentials as the first SMS to the service. All SMS are sent over the air unencrypted so it is not the best solution. However it is better to send it once instead of sending it every time you need to know the balance.

The next time the user sends anything to the service it responds with the balances of each card. This is very simple and works on any phone anywhere. And hey I did not have to do any GUI work!

However it would be great if SL would add support for OAuth 2.0 for example. To make it possible to avoid sending credentials over SMS. However it would then be required to have access to a web browser on the phone or access to a web browser on a computer.

Written 2016-02-29 by Martin Harari Thuresson