The source code has different commit hash...

[ustk]

Now I'm getting this when trying to export my plugin... I just rebuilt Hise and I assure you the branch has not changed between the built and the current state...
Getting so, so many bugs these days, it's so frustrating spending most of the time trying to get Hise, the DLLs, and everything just working, that development of the plugin itself just became a side job...

[dannytaurus]

@ustk Hopefully we're in a transition phase while Christoph is doing lots of AI-related work.

[David Healey]

@dannytaurus I think that's what it is. We need a develop develop branch :)

[Christoph Hart]

@ustk yeah sorry, that happens if I forget to run a shell command that bakes in the commit hash or if you actually are using an old build that doesn't match your source code.

which commit are you on?

We need a develop develop branch

Actually with the new AI superpower of just telling Claude to cherrypick a commit into another branch we might actually try to use the branch concepts as they are intended.

[resonant]

@Christoph-Hart said in The source code has different commit hash...:

Actually with the new AI superpower of just telling Claude to cherrypick a commit into another branch we might actually try to use the branch concepts as they are intended.

For those completely new to this, where can we find information on the steps to follow for setup and operation for Agentinc Coding with HISE?

[Christoph Hart]

@resonant It's still a moving target so as long as there's not an established workflow I'd like to refrain from putting out an "official" guide. The current state is:

• there's an MCP server which contains the entire documentation & code examples in a LLM consumable way.
• there's an LSP server that can be invoked after each script edit and catches hallucinated API methods by sending a "parse" request to HISE which performs a "light" compilation step and returns all errors it finds.
• there's a CLI tool that can remotely control HISE and add / remove modules, build up UI component trees. This functionality is currently half-implemented in the MCP server, but I'll rip that out of the MCP server and move it to the CLI exclusively soon. That will be the meat and bone of the agentic integration since it enables the agent to perform most tasks automatically.
• there's the "art" of prompting - give the AI agent instructions / style guides so it uses a efficient workflow & produces good code. This is probably the most "moving" target of them all so it'll be the last thing I tackle after all the other tools fall into place.
• there are multiple agent tools (Opencode, Claude Code, Pi, Codex) - most of them support all the tools listed above in some way, but there are subtle details between them - eg. Claude never runs LSP servers after an edit so you have to manually patch it up.

If you want to dabble around with the current feature set you're best bet is to checkout the hise_mcp_server repo, grab a AI client and tell it to install it, then see how much it improves your workflow.

https://github.com/christoph-hart/hise_mcp_server

Just point the agent to the readme and let it figure out all the rest.