Practical AI applications for Harbour development

396 views
Skip to first unread message

Lorenzo Fiorini

unread,
Mar 31, 2025, 11:17:36 AM3/31/25
to harbou...@googlegroups.com
I'm exploring practical AI applications for Harbour development.

Would anyone be willing to share their experiences and results?
  
Best regards,
Lorenzo

antonio....@gmail.com

unread,
Apr 1, 2025, 10:19:05 AM4/1/25
to Harbour Users
Dear Lorenzo,

MCP servers are the hotest topic in AI right now. Claude implemented the MCP protocol past november 2024, OpenAI is currently implementing it and Google, etc. are on the same track.

This is a great free and open source MCP manager tool to start using them. You can use it with Ollama so no need to pay for keys and you can start getting the MCP concept:

Model Context Protocol (MCP)
The Model Context Protocol is an open standard that enables developers to build secure, two-way connections between their data sources and AI-powered tools

MCP servers can be implemented in any language, so who knows, we could provide Harbour as a MCP server too... ;-)

best regards

Lautaro Moreira

unread,
Apr 1, 2025, 10:31:54 AM4/1/25
to harbou...@googlegroups.com, antonio....@gmail.com

Dear Antonio,

 Since you published the MCP source for Harbor I am trying to understand it, but I do not see what the AI ​​from Ollama to the MCP scheduled to request, for example, data from a database, this is obviously because of my ignorance of the matter.

There is some documentation/example (in any language) where you can see this implementation.

 My interest comes that I have an idea of ​​generation of an assistant, which depending on the request made, should consult 1 or 2 external data sources, one that is in a database and another that is a document file, each with its own form of consultation and this MCP comes very well for this, if I understand how to use it.

Greetings,


--- Spanish ----

Estimado Antonio,

Desde que publicaste el fuente del MCP para harbour estoy tratando de entenderlo, pero no veo de que forma se comunicara la IA desde ollama al MCP programado para solicitar, por ejemplo, datos de una base de datos, esto obviamente es por mi desconocimiento del asunto.

Hay alguna documentacion/ejemplo ( en cualquier lenguaje ) donde poder ver esta implementacion.

Mi interes viene sobre que tengo una idea de generacion de un asistente, que dependiendo de la solicitud que se le haga, debera consultar 1 o 2 fuentes externas de datos, una que esta en una base de datos y otra que es un archivo de documentos, cada uno con su forma propia de consulta y esto del MCP viene muy bien para esto, si es que entiendo como usarlo.

Saludos,

-------------------------

Lautaro Moreira

Lorenzo Fiorini

unread,
Apr 1, 2025, 12:16:56 PM4/1/25
to harbou...@googlegroups.com

Hi Antonio,

I appreciate your continued support of the Harbour community.

I'm working on a project that utilizes Node.js, Ollama, and LangChain to build a system for generating PRG code. 

I'm leveraging SQLite3 for local embedding storage and feeding the system with various source files (PRG, JSON, TS, HTML, CSS, MD). 

The aim is to fine-tune a model to produce practical PRG code.

At the moment I'm using Mistral 7B Instruct Q4 but it can be any other that is suited for a local usage.

I'd be interested in any insights or recommendations anyone might have.

Best regards,
Lorenzo

--
You received this message because you are subscribed to the Google Groups "Harbour Users" group.
Unsubscribe: harbour-user...@googlegroups.com
Web: https://groups.google.com/group/harbour-users
---
You received this message because you are subscribed to the Google Groups "Harbour Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to harbour-user...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/harbour-users/73b8baf8-3190-46f3-8c61-d491909917f9n%40googlegroups.com.
-- 
Il presente messaggio è indirizzato esclusivamente ai destinatari. Tutte le informazioni
contenute, compresi eventuali allegati, sono confidenziali ai sensi del Regolamento (UE)
2016/679 e del D. Lgs. 196/2003. Pertanto ne sono vietati l'inoltro, la divulgazione e la
messa a disposizione in qualunque forma o modo, in mancanza di preventiva autorizzazione
del mittente. Qualora il messaggio Le fosse pervenuto per errore, La invitiamo cortesemente
ad eliminarlo in modo definitivo dando immediato riscontro.

This message is exclusively addressed to the recipients. All the information contained in this
message, including any attachments, is confidential in compliance with Regulation (UE)
2016/679 and Legislative Decree 196/2003. Therefore, forwarding, disclosing and making
the above-mentioned information available without prior authorization from the sender is
forbidden in any form or manner. If you have received this message in error, we kindly invite
you to delete it permanently and to notify the sender.
--

antonio....@gmail.com

unread,
Apr 1, 2025, 1:13:21 PM4/1/25
to Harbour Users
Dear Lautaro,

To get used to the MCPs concept and use I do recommend you to install and use https://github.com/OpenAgentPlatform/Dive

Para acostumbrarte al concepto y uso de los MCPs te recomiendo que instales y uses https://github.com/OpenAgentPlatform/Dive 
 
saludos

antonio....@gmail.com

unread,
Apr 1, 2025, 1:18:01 PM4/1/25
to Harbour Users
Dear Lorenzo,

My advise for PRG generation is Grok 3. To me it is by far the best model for Harbour coding.

best regards

hmpaquito

unread,
Apr 1, 2025, 1:37:03 PM4/1/25
to Harbour Users
Estimado Antonio,

Pero Grok-3 no tiene API y no se sabe cuando estará disponible.
Alguien ha hecho alguna API apócrifa pero ignoro sus limitaciones y si necesita de "mantenimiento periódico"

Salu2

Lautaro Moreira

unread,
Apr 1, 2025, 2:18:24 PM4/1/25
to harbou...@googlegroups.com, antonio....@gmail.com

Dear Antonio,

Many thanks for the links

Saludos,

Lautaro Moreira
--
You received this message because you are subscribed to the Google Groups "Harbour Users" group.
Unsubscribe: harbour-user...@googlegroups.com
Web: https://groups.google.com/group/harbour-users
---
You received this message because you are subscribed to the Google Groups "Harbour Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to harbour-user...@googlegroups.com.

antonio....@gmail.com

unread,
Apr 2, 2025, 6:01:48 AM4/2/25
to Harbour Users
Enhanced Class TOllama with Agents support:


Mostly of the code has been developed by Grok 3, I just supervised and tested it

best regards

HBQuerier

unread,
Apr 2, 2025, 8:57:48 PM4/2/25
to Harbour Users
Thanks Antonio, I'll look into it.  Most of the AIs I've tried with Harbour, think there's and ENDPROC or ENDFUNCTION statement at the end of every procedure or function.

Looks like Grok 3 is more expensive per month than Claude.  Will have to check out their annual rate.

Francesco Perillo

unread,
Apr 3, 2025, 5:37:22 AM4/3/25
to harbou...@googlegroups.com
I originally read your request as "what is your use of AI INSIDE your programs ? "  - not to generate code but actually helping users with the tasks.

So I ask the question explicitely: are you using AI to "help" users completing tasks ? Forecasting ? behaviour analysis ?
Francesco

--
You received this message because you are subscribed to the Google Groups "Harbour Users" group.
Unsubscribe: harbour-user...@googlegroups.com
Web: https://groups.google.com/group/harbour-users
---
You received this message because you are subscribed to the Google Groups "Harbour Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to harbour-user...@googlegroups.com.

Lorenzo Fiorini

unread,
Apr 3, 2025, 10:01:01 AM4/3/25
to harbou...@googlegroups.com

Hi Francesco,

I'm currently experimenting with Retrieval Augmented Generation (RAG).
I've observed that applying RAG to relational business data is challenging due to the normalized structure of tables, relations, keys, and values.
This complexity makes it difficult to produce data directly usable by an LLM.

However, RAG or fine-tuning for code generation yields more promising results.
The generated code can be easily linted, compiled, and tested, and the source code's text format is readily suitable for tokenization, embedding, and retrieval.

Actually my goal is to setup a AI environment for RAG (and/or fine-tuning): find out the tools and models to use and  understand hardware and resources required.

I'm still in the experimental phase, and these are my initial findings.

Best regards,
Lorenzo

Reply all
Reply to author
Forward
0 new messages