2025-20

[michielbdejong]

I'll create a unit test for adding a Drop to a Tub.

[michielbdejong]

----------onMessage-----------
file:///Users/michiel/gh/michielbdejong/automerge-basic/build/src/drops.js:3
        tubsId: from.foreignIds.tubs || identifierToInternal(from.model, from.localId),
                                ^

TypeError: Cannot read properties of undefined (reading 'tubs')

[michielbdejong]

A message is added on Solid, it is synced to Slack. Metadata is added there but not fed back to Solid. But then automerge is restarted, and the message is synced again. Two possible remedies:

• write back the foreignId to Solid
• query existing messages before creating a new one on Slack

[michielbdejong]

It's all still a bit messy.

[michielbdejong]

Tomorrow I'll record the screencast for the RfP using https://github.com/tubsproject/reflector

[michielbdejong]

• adopt cambria
• sdm-related functions into sdm
• update SolidOS Chat pane

[michielbdejong]

const todayDoc = 'https://michielbdejong.solidcommunity.net/IndividualChats/nlnet-demo/2025/05/13/chat.ttl';
const streamingUrl = `https://solidcommunity.net/.notifications/StreamingHTTPChannel2023/${encodeURIComponent(todayDoc)}`;
const res = await fetch(streamingUrl);
const textStream = res.body.pipeThrough(new TextDecoderStream());
let first = true;
for await (const notificationText of textStream) {
  console.log(notificationText);
  if (first) {
    first = false;
  } else {
    window.location = '';
  }
}

[michielbdejong]

Cambria operations for Slack and Solid:

• rename localId -> ts / uri
• copy text - text
• rename authorId - user / authorWebId
• rename channelId - channel / chatUri
• in Slack, add metadata.event_type: 'from_tubs'
• in Slack, drop foreignIds into metadata.event_payload.foreignIds
• in Solid, add an owl:sameAs triple for each foreignId.

[michielbdejong]

Finished the Solid-Slack bridge, including screencast demo: https://github.com/tubsproject/reflector
Next task: add tasks, from the work we did in https://github.com/federatedbookkeeping/github-dxos/issues

[michielbdejong]

The code is in https://github.com/federatedbookkeeping/bridgeBot
It's good to read the readme there.
Caching API responses is nice.
https://github.com/federatedbookkeeping/bridgeBot/blob/main/src/client/github.ts will be useful.
But I need to take into account that the milestone is for a Solid Data Module about Tasks. So adding the GitHub issues one makes sense, but other than that I should be working in solid-contrib/data-modules#74

Tomorrow is (at least partially) a surf day.

[michielbdejong]

Now may be a good time to think about adoption. I should:

• define a service in terms of functionality
• do a risk analysis of possible failure modes and security risks
• set up a software QA pipeline
• implement its requirements and security mitigations
• start using it myself
• offer a private alpha

Then I can see if it works and if people like it. Once I have that, I can create an operational plan, consider investing in something more than a bedroom operation, grow, take over the world

[michielbdejong]

The definition of the functionality can be incremental. I can start with syncing a copy of various types of data to my own Dropbox account. That's nice in itself, especially with for instance Slack's 90-day disappearing messages.

If I don't post to Slack then I can also not post the wrong thing.
I could think of ways to test what it posts and making it more reliable, then trying it myself in a channel of my own.

For hosting the OAuth keys I'll just try to follow best practices, maybe ask Mahdi to help.

The service can also include things like Peppol forwarding and LedgerLoops agency.

In any case, I will need to do a lot of software engineering and unit testing to build this properly.

[michielbdejong]

Could reflector also act as a CreditCommons node?
Well, if you combine it with some storage then maybe yes, although there also is the business logic.

In your bookkeeping system, you can write down that you want to make a transfer, and then reflector can make the API calls to neighbour CreditCommons servers to make that happen.

Apart from schema translations, I also need to work out how to implement constraints on the data that is synced, that might cause a node to reject an update based on business logic.

[michielbdejong]

Anything I do that's not aimed at helping humans stay in control against the forces of capitalist AI is a waste of time. In the light of my duty to help in what could be called an emergency, it would even be negligence to work on anything else.

What can I find out about open source web agents?

[michielbdejong]

The vendors would run their agents.
Copilot becomes the interface between the user and the vendors. It's web, search, and browser and operating system, all in one.
So we need to have an open MCP web, where the user has an open source MCP browser.
Is that maybe more important to work on than Tubs? Can it be part of Tubs?

If your data is on Tubs, it becomes easier to invoke MCP servers on it.

Two lock-in questions:

• where is a user's data?
• where can the vendors be contacted?

For vendors to offer their services in multiple places is pretty easy. The device a user buys (and this holds especially if educational institutes tell their students to bring a Windows computer to class!) has a strong lock-in factor into which user agent they will use (i.e., all students will be using CoPilot as their user agent that is at the same time a vendor agent).

Should I work more on letting Tubs interface with MCP? I should read the launch blogpost

[michielbdejong]

Wow, amazing how the list of MCP servers has grown since November.
But they are really just 100-line wrappers around APIs? https://github.com/modelcontextprotocol/servers/blob/main/src/gitlab/index.ts

At least they will be very useful as coding examples for API access in Tubs! :)

I should be able to quickly create a Tub for any MCP from that repo.