AI
Implementation-accurate guide for the Forge AI extension
Overview
github.com/xraph/forge/extensions/ai is a Forge extension that wires core AI services into the container.
It is intentionally a thin wrapper around github.com/xraph/ai-sdk and focuses on DI registration, not a full platform bootstrap.
What It Registers Automatically
aisdk.LLMManager(with key aliases:forge.ai.sdk.llmManager,forge.ai.llmManager)aisdk.StateStoreaisdk.VectorStore*ai.AgentFactory*ai.AgentManager- Training interfaces only when
config.Training.Enabled == true:ai.ModelTrainerai.DataManagerai.PipelineManager
What It Does Not Auto-Register
- No automatic route mounting. You must register
AgentControllerroutes yourself. - No automatic startup of inference engine, middleware, or monitoring packages.
- No automatic LLM provider registration from
config.LLM.Providers(providers are currently registered manually).
Important Runtime Notes
- State store creation falls back to in-memory store when configured backend creation fails.
- Vector store creation falls back to in-memory store when configured backend creation fails.
- Agent definitions are held in memory by
AgentManager. - Extension
Health()currently returnsniland does not aggregate component checks.
Read This Next
Getting Started
Build a working setup with manual provider registration and type-based DI.
Features
Deep feature behavior, template coverage, and current implementation limits.
Configuration
Field-level config behavior, defaults, and production-safe config patterns.
DI and Lifecycle
Constructor graph, aliases, eager registration, and preferred Vessel pattern.
HTTP API
Exact routes exposed by AgentController, payloads, and behavior.
Training and Inference
How the extension relates to the training and inference subpackages.
Functions
Grouped function-level reference for extension entry points and subpackages.
Troubleshooting
Fix common startup and DI issues quickly.
How is this guide?