fc-plugin-codbi/com.github.xima_formcycle_entwicklerkreis.fc.plugin.codbi.logic.cb.ai/LLAMA

LLAMA

abstract class LLAMA : AI

Base class for AI models served via a local LLAMA server process.

Phases

Intelligence — Detect OS (os.name) and architecture (os.arch)
Fetch — Download the correct LLAMA-Server binary + GGUF model (with resume support)
Ignition — Launch the server via ProcessBuilder on a local port
Request — All inference goes through http://127.0.0.1:{port}/v1/chat/completions.

Crash isolation

Because the AI runs in a separate OS process, if the model runs out of RAM, the OS kills the LLAMA-Server process but the Formcycle Tomcat JVM does not even feel a bump.

Plugin properties

Property	Default	Description
`Active_AI`	—	Must contain `llama_engine`
`AI_Remove`	—	If contains `llama_engine`, clean up all
`AI_LLAMA_ENGINE_Port`	`8392`	Local port for LLAMA-Server
`AI_LLAMA_ENGINE_Threads`	physical cores	Number of CPU threads
`AI_LLAMA_ENGINE_CtxSize`	`32768`	Context window size (shared across parallel slots)
`AI_LLAMA_ENGINE_GpuLayers`	auto-detect	Layers offloaded to GPU (-1 = auto)
`AI_LLAMA_ENGINE_Release`	`b8175`	llama.cpp release tag for downloads
`AI_LLAMA_ENGINE_ServerArgs`	—	Extra CLI args for LLAMA-Server
`AI_LLAMA_ENGINE_MaxConcurrent`	`2`	Maximum concurrent inferences allowed across all local servers
`AI_LLAMA_ENGINE_Parallel`	`4`	DEPRECATED Number of parallel inference slots per server (use MaxConcurrent instead)

Domains to whitelist

github.com — llama-server binary releases
objects.githubusercontent.com — GitHub release asset CDN

DSGVO / EU-AI Act

All data stays on the local machine.
No external API calls.
Same compliance advantages as all other CodBi AI implementations.

Inheritors

Standard

Constructors

LLAMA

constructor()

Types

Companion

object Companion

The companion for static members.

Functions

execute

abstract fun execute(p0: IPluginServletActionParams): IPluginServletActionRetVal

getDescription

open fun ~~getDescription~~(): String

open fun getDescription(p0: Locale): String

getDisplayName

open override fun getDisplayName(p0: Locale): String

getName

abstract override fun getName(): String

initialize

open override fun initialize(configData: IPluginInitializeData)

Initializes the LLAMA infrastructure: creates directories, reads plugin properties. Subclasses should call super.initialize(configData) then proceed with downloading and starting the server.

initPlugin

open fun ~~initPlugin~~()

install

open fun install(p0: IPluginInstallData)

shutdown

open override fun shutdown(shutdownData: IPluginShutdownData?)

Shuts down the LLAMA-Server process and releases resources.

open fun ~~shutdown~~()

startJanitor

fun startJanitor()

Initiates a task that removes unused images that're expired (msExpirationIDedImages) from the cache (cacheIDedImages).

uninstall

open fun uninstall(p0: IPluginUninstallData)

validateConfigurationData

open override fun validateConfigurationData(configData: IPluginValidationData): IPluginInitializeValidationResult?

Rejects tenant-level installation. CodBi must be installed as a system plugin because its AI services (Whisper, LLAMA) bind local server ports and manage heavyweight processes that would conflict when instantiated once per tenant.