registry-entry.json

{
	"id": "local-inference",
	"name": "Local Inference",
	"description": "Unified Eliza-1 local inference provider — text generation, embeddings, voice (TTS/ASR), and image description via the elizaOS llama.cpp fork. Replaces the deprecated plugin-local-ai / plugin-local-embedding pair.",
	"npmName": "@elizaos/plugin-local-inference",
	"version": "2.0.0-beta.2",
	"source": "bundled",
	"tags": [
		"ai-provider",
		"llm",
		"local-models",
		"self-hosted",
		"local",
		"embedding",
		"voice",
		"tts",
		"transcription",
		"eliza-1"
	],
	"config": {
		"ELIZA_PII_SWAP_ENABLED": {
			"type": "boolean",
			"required": false,
			"sensitive": false,
			"default": "false",
			"label": "Enable PII Swap",
			"help": "Master switch (read by @elizaos/core). When on, named-entity PII is pseudonymized before every model call and restored at the execution boundary. Detection runs as an extraction prompt on the resident local Eliza-1 backend, so PII never leaves the device. Off by default — zero behavior change.",
			"advanced": false
		},
		"ELIZA_PII_SWAP_EXEMPT_VALUES": {
			"type": "string",
			"required": false,
			"sensitive": false,
			"label": "Never-Swap Values",
			"help": "Comma-separated exact values that must never be pseudonymized (false-positive opt-out), on top of the built-in framework/brand blocklist.",
			"advanced": true
		},
		"ELIZA_PII_SWAP_DISABLED_KINDS": {
			"type": "string",
			"required": false,
			"sensitive": false,
			"label": "Skip Entity Kinds",
			"help": "Comma-separated entity kinds to skip swapping, e.g. location,address.",
			"advanced": true
		},
		"MODELS_DIR": {
			"type": "file-path",
			"required": false,
			"sensitive": false,
			"label": "Models Dir",
			"help": "Filesystem path to the directory where GGUF models are stored. Defaults to <stateDir>/models.",
			"advanced": false
		},
		"CACHE_DIR": {
			"type": "file-path",
			"required": false,
			"sensitive": false,
			"label": "Cache Dir",
			"help": "Filesystem path to the cache directory for model assets (tokenizers, ONNX caches).",
			"advanced": false
		},
		"LOCAL_SMALL_MODEL": {
			"type": "string",
			"required": false,
			"sensitive": false,
			"default": "text/eliza-1-2b-128k.gguf",
			"label": "Small Model",
			"help": "Filename of the small local model (TEXT_SMALL handler).",
			"advanced": false
		},
		"LOCAL_LARGE_MODEL": {
			"type": "string",
			"required": false,
			"sensitive": false,
			"default": "text/eliza-1-4b-128k.gguf",
			"label": "Large Model",
			"help": "Filename of the large local model (TEXT_LARGE handler).",
			"advanced": false
		},
		"LOCAL_EMBEDDING_MODEL": {
			"type": "string",
			"required": false,
			"sensitive": false,
			"default": "text/eliza-1-2b-128k.gguf",
			"label": "Embedding Model",
			"help": "Filename of the embedding model used for vector embeddings.",
			"advanced": false
		},
		"LOCAL_EMBEDDING_DIMENSIONS": {
			"type": "number",
			"required": false,
			"sensitive": false,
			"default": "1024",
			"label": "Embedding Dimensions",
			"help": "Number of dimensions the embedding model outputs.",
			"advanced": false
		},
		"CUDA_VISIBLE_DEVICES": {
			"type": "string",
			"required": false,
			"sensitive": false,
			"label": "CUDA Visible Devices",
			"help": "Set to restrict which CUDA-enabled GPUs the runtime sees (e.g. `0,1`).",
			"advanced": true
		},
		"ELIZA_LOCAL_LLAMA": {
			"type": "string",
			"required": false,
			"sensitive": false,
			"label": "AOSP FFI Loader",
			"help": "Set to `1` on Android (elizaOS build) to enable the in-process bun:ffi loader (libllama.so + libeliza-llama-shim.so) via the @elizaos/plugin-native-inference companion plugin.",
			"advanced": true
		},
		"ELIZA_MODEL": {
			"type": "select",
			"required": false,
			"sensitive": false,
			"label": "Eliza-1 Tier",
			"help": "Select a published Eliza-1 fine-tune tier. The runtime auto-picks the right quant flavor (gguf/polarquant/fp8/bf16) for your detected GPU and pulls it from HuggingFace on first use.",
			"options": [
				{
					"value": "eliza-1-2b",
					"label": "Eliza-1 2B (mobile / entry)",
					"description": "Smallest/entry tier; fits Android/iOS on-device inference and 16 GB consumer GPUs at Q4_K_M."
				},
				{
					"value": "eliza-1-4b",
					"label": "Eliza-1 4B (16-24 GB GPU)",
					"description": "Desktop-tier fine-tune for everyday workstation use."
				},
				{
					"value": "eliza-1-9b",
					"label": "Eliza-1 9B (24-48 GB GPU)",
					"description": "Workstation-tier fine-tune; Q4_K_M for 16 GB cards, PolarQuant for 24+."
				},
				{
					"value": "eliza-1-27b",
					"label": "Eliza-1 27B (cloud / 48 GB+)",
					"description": "Cloud-tier fine-tune; Q6_K GGUF on RTX 5090, fp8/bf16 on H200/B200."
				}
			],
			"advanced": false
		}
	},
	"render": {
		"visible": true,
		"pinTo": [],
		"style": "card",
		"icon": "Monitor",
		"group": "ai-provider",
		"groupOrder": 0,
		"actions": ["enable", "configure"]
	},
	"resources": {
		"homepage": "https://github.com/elizaOS/eliza/tree/main/plugins/plugin-local-inference#readme",
		"repository": "https://github.com/elizaOS/eliza",
		"setupGuideUrl": "https://docs.eliza.ai/plugin-setup-guide#local-inference"
	},
	"dependsOn": [],
	"kind": "plugin",
	"subtype": "ai-provider",
	"launch": {
		"type": "server-launch",
		"capabilities": [],
		"bootHook": {
			"specifier": "@elizaos/plugin-local-inference/runtime",
			"exportName": "registerLocalInferenceBoot"
		}
	}
}