Jump to content

Recommended Posts

Posted
6 hours ago, Peter Knight said:

Lovely view. Am I way off if I guess Pembrokeshire or maybe more likely Cornwall? 

Head north about a thousand miles.... 😀

Posted
1 hour ago, Joss said:

Head north about a thousand miles.... 😀

Ok, at a stretch, I was going to say Scotland, but thought that was a big move from London.
Even further to Scottish Isles.
Anywho - lovely office setup. 

Posted
15 hours ago, Joss said:

The error that came up during install was that makedir failed to create /site/assets/at/jobs/done, and failed to create the htaccess within "done", and the same with "failed", "pending," and "running."

Note - I am using laragon on windows 11 rather than a unix server.

@Joss hmm sounds like it didn't have permission to create the directories. You might check that processwire has write permission to site/assets. But if I recall correctly AgentTools will attempt to create them again when/if you use the scheduled jobs feature. 

Posted

@ryan - On a slightly different note, I have just installed the qwen-2.5-32b instruct, and it is running fine on my system via ollama (better at balancing tham LM Studio, it appears).  But I have been setting it up for gramma checking, using modelfile to customise it. Basically, it tells it to act as an editor, not to get too creative, then gives some specific instructions over strange words, grammar, etc. It then creates a model for just that purpose.

Could something similar be run for AgentTools so that the actual model has specific hard-wired PW knowledge in advance? I don't know enough to say if this would speed things up or anything like that. But it certainly helps with Word as it knows in advance character names etc, without it having to be prompted at the time.

  • 2 weeks later...
Posted

My agent/engineer shows me a wrong path for migrations.
Double-checked it. Each try has the wrong path.
Running the migrations from within the dialogue itself works fine.

image.png.0a01ed487b6b71521e9eaacb4b21eacc.png

wrong:
site/assets/migrations/20260716203216_convert_innodb_utf8mb4.php

corrent:
site/assets/at/migrations/20260716203216_convert_innodb_utf8mb4.php

Posted

OpenAI Responses API agents fail when AgentTools Tasks use tools

An additional OpenAI agent configured with https://api.openai.com/v1/responses works for requests without tools, but built-in Tasks fail with:

API error (400): Missing required parameter: 'tools[0].name'

AgentTools appears to send Chat Completions function definitions using tools[].function.name. The Responses API expects tools[].name at the top level. Responses tool-call parsing and function_call_output continuation also appear to require separate handling, since the current loop expects choices[].message.tool_calls.

Chat Completions agents continue to work.

A fix would be very welcome 🙂 

  • Like 1
Posted

@Joss

I posed your question to the GPT 5.5 session that maintains AgentTools:

Quote

 

A custom Ollama Modelfile can absolutely bake in ProcessWire/AgentTools orientation, but I’d treat it as a convenience layer, not the source of truth. The safest recommendation is:

  • Point the agent/client at the site’s root AGENTS.md.
  • Also point it at /site/modules/AgentTools/AGENTS.md.
  • Optionally make an Ollama model whose system prompt says something like: “You are working on ProcessWire sites. Before making changes, read the project AGENTS.md and AgentTools AGENTS.md, then use AgentTools CLI/API docs as the current source of truth.”

The reason I wouldn’t hard-wire too much PW knowledge into the model itself is staleness. ProcessWire APIs, local site conventions, and AgentTools commands can change. The AGENTS.md files and API.md docs stay close to the code, so they’re better as live context. A Modelfile is useful for stable behavior and orientation, like “prefer migrations,” “read docs before guessing,” “be concise,” “don’t invent CLI flags,” etc.

It may help speed/quality a little by giving the model a better starting posture, but it won’t replace retrieval/tool use. For smaller local models especially, I’d keep the baked-in prompt short and make it mostly a pointer to the right files.

 

 

@wbmnfktr @psy The latest AgentTools version includes updates for both of the issues you mentioned. 

  • Like 4
  • Thanks 1

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
  • Recently Browsing   0 members

    • No registered users viewing this page.
×
×
  • Create New...