Check out the People Discovery (kind 5301), done on the terminal because clients don't support it yet.

This DVM searches for pubkeys that another pubkey might want to follow based on its existing follows. If the "source" pubkey is not specified in the job input, the results are returned for the pubkey who signed the job request.

Example request:

```json
{
"content": "",
"kind": 5301,
"tags": [["p", "f45f5667ffe7649d7c9f648930c9498cda88974e7ea28929194d0167cdcbc124"]]
}
```

Example using nak:

```shell
# create kind 5301 job request to get follow recommendations for the dataMachine pubkey, and send to a relay
EVENT=$(nak event -k 5301 --tag param="f45f5667ffe7649d7c9f648930c9498cda88974e7ea28929194d0167cdcbc124;p" wss://relay.primal.net)

# extract event ID
EVENT_ID=$(jq -r '.id' <<< $EVENT)

# query relay for the response
nak req --stream -k 6301 --tag e="$EVENT_ID" wss://relay.primal.net | jq
```

## People Search (kind 5303)

This DVM searches for pubkeys by looking at the `name` and/or `display_name` fields in their profile metadata:

- It checks all profiles in our database, and calculates a similarity score based on the job input.
- To avoid returning spammy/scammy profiles, the similarity score is multplied by a ranking score derived from how many reputable profiles follow the matching profile.
- Given the input `jack`, there will be lots of perfect matches, but the first result will be Jack Dorsey's pubkey.

**Example request:**

```json
{
"content": "",
"kind": 5303,
"tags": [
["i", "jack", "text"],
["param", "max_results", "10"]
]
}
```

**Example using nak:**

```shell
# search for the dataMachine profile, add a typo on purpose
EVENT=$(nak event -k 5303 --tag i="datamchaine;text" wss://relay.primal.net)

# extract event ID
EVENT_ID=$(jq -r '.id' <<< $EVENT)

# query relay for the response
nak req --stream -k 6303 --tag e="$EVENT_ID" wss://relay.primal.net | jq
```

DataMachine.ai demo for kind 5301, this is the recommend people to follow.

Check this #DVM out using #Vendata

Recommends popular profiles based on your existing follow list.

Where have you gotten better results?

Yes!

Good morning. Time to list unleashed.chat, the grok for Nostr on nostrapps.com :)

Unleashed.Chat v0.1.21

-Mixtral 8x22B Instruct–Mistral AI's flagship FOSS model. This is not an uncensored variant like our other models (yet).

-The Mixtral 8x7B and 8x22B models are now capable of searching the Internet. Try to ask for the current bitcoin price or the population of a country.

-Added API endpoints for fetching the current balance and creating a Lightning invoice for adding funds - see https://unleashed.chat/api/docs
-Billing is now based on the length of the response (output tokens) rather than generation time.

3000+ users 🎉

Try it!

Unleashed.Chat v0.1.20

Nostr mode improvements:
- Ask “What’s happening on my feed today?” The AI can now access recent posts by people you follow. You need to have Nostr login enabled for this to work. If you have an existing account, go to the Account page (left side menu) and connect your npub there.
- You'll see some prompt suggestions when you start a new chat in Nostr mode.

Other:
- AI generated code blocks now have a button for copying their contents to the clipboard.

Shared infrastructure is on the roadmap but it's a big undertaking when keeping privacy as priority

People want to keep chat history and upload files

It's on the list

Unleashed.Chat v0.1.19

- We now have a Nostr #DVM that can summarize text inputs using our Mistral 7B model. More models and request types coming soon! You can test it out with a Nostr DVM client like https://vendata.io.
- You can now start a chat with a data source right from the "Files and Data Sources" table. Just click on the three dots at the end of the row and select "Start Chat".
- Other small bugfixes.