@elizaos/shared 2.0.0-beta.1 → 2.0.11-beta.7

package/local-inference/types.d.ts

@@ -0,0 +1,510 @@
+/**
+ * Local inference shared types.
+ *
+ * Shared contracts referenced by the server-side service in
+ * `@elizaos/app-core` and the UI client in `@elizaos/ui`.
+ *
+ * Server-only logic (KV cache management, native runtime lifecycle,
+ * conversation registry, metrics scraping) stays in `app-core`; only
+ * the type contracts live here.
+ */
+/** Agent slot ids the runtime maps to a local model. */
+export type AgentModelSlot = "TEXT_SMALL" | "TEXT_LARGE" | "TEXT_EMBEDDING" | "TEXT_TO_SPEECH" | "TRANSCRIPTION";
+/** Subset of `AgentModelSlot` that participates in text generation. */
+export type TextGenerationSlot = Extract<AgentModelSlot, "TEXT_SMALL" | "TEXT_LARGE">;
+export declare const AGENT_MODEL_SLOTS: AgentModelSlot[];
+export declare const TEXT_GENERATION_SLOTS: TextGenerationSlot[];
+/**
+ * Mapping of agent slot → installed model id. Persisted to disk by
+ * `assignments.ts` and consumed by both the runtime router and the UI
+ * model picker.
+ */
+export type ModelAssignments = Partial<Record<AgentModelSlot, string>>;
+/**
+ * Installed-model registry entry. The on-disk format is JSON.
+ */
+export interface InstalledModel {
+    /** Matches CatalogModel.id when installed from the curated catalog. */
+    id: string;
+    displayName: string;
+    /** Absolute path to the GGUF file on disk. */
+    path: string;
+    sizeBytes: number;
+    /**
+     * Eliza-1 bundle root when this installed model came from a multi-file
+     * manifest. `path` still points at the primary GGUF used for loading the
+     * model; sibling voice/cache files live under this root.
+     */
+    bundleRoot?: string;
+    /** Absolute path to the validated `eliza-1.manifest.json`, when present. */
+    manifestPath?: string;
+    /** SHA256 of the validated manifest file, when present. */
+    manifestSha256?: string;
+    /** Semver bundle version from the manifest, when present. */
+    bundleVersion?: string;
+    /** Total bytes installed under `bundleRoot`, including voice/cache files. */
+    bundleSizeBytes?: number;
+    /** HF repo this came from, when known. */
+    hfRepo?: string;
+    /** ISO timestamp of install completion. */
+    installedAt: string;
+    /** ISO timestamp of last activation (null if never loaded). */
+    lastUsedAt: string | null;
+    /** Where we got this model from. Determines whether Eliza owns the file. */
+    source: "eliza-download" | "external-scan";
+    /**
+     * When source === "external-scan", which tool the file belonged to.
+     * Prevents Eliza from deleting files other apps own.
+     */
+    externalOrigin?: "lm-studio" | "jan" | "ollama" | "huggingface" | "text-gen-webui";
+    /** SHA256 of the GGUF file recorded at install time. Optional for legacy entries. */
+    sha256?: string;
+    /** ISO timestamp of the last successful re-verification. Absent = never verified since install. */
+    lastVerifiedAt?: string;
+    /**
+     * ISO timestamp of the one-time on-device verify pass for an Eliza-1
+     * bundle (load → 1-token text gen → 1-phrase voice gen → barge-in cancel,
+     * per `packages/inference/AGENTS.md` §7). Absent = the bundle was
+     * materialized but the verify-on-device pass has not run; such a bundle
+     * must not be auto-selected as the recommended default.
+     */
+    bundleVerifiedAt?: string;
+    runtimeRole?: "chat" | "mtp-drafter";
+    companionFor?: string;
+}
+export type ModelBucket = "small" | "mid" | "large" | "xl";
+export type ModelCategory = "chat" | "code" | "tools" | "tiny" | "reasoning";
+export type LocalRuntimeBackend = "capacitor-llama" | "llama-cpp";
+export type OpenVinoDeviceKind = "CPU" | "GPU" | "NPU";
+export interface OpenVinoHardwareProbe {
+    /**
+     * True when an OpenVINO runtime install is discoverable from env vars or
+     * common Linux install paths. CPU execution still requires this runtime.
+     */
+    runtimeAvailable: boolean;
+    devices: OpenVinoDeviceKind[];
+    /**
+     * Intel GPU plugin candidates. On Linux these are `/dev/dri/renderD*`
+     * render nodes plus the Intel Compute Runtime userspace libraries.
+     */
+    gpu: {
+        renderNodes: string[];
+        computeRuntimeReady: boolean;
+        missingLinuxPackages: string[];
+    };
+    /**
+     * Intel NPU plugin candidates. On Linux these are `/dev/accel/accel*`
+     * nodes; their presence only means the device is wireable, not validated.
+     */
+    npu: {
+        accelNodes: string[];
+    };
+    /** Best-known OpenVINO device for static-graph ASR workloads. */
+    recommendedAsrDevice: OpenVinoDeviceKind | null;
+    warnings: string[];
+}
+/**
+ * Runtime kernel/capability handles advertised by the optimized llama.cpp
+ * runtime. Most entries are specialised kernels shipped by the
+ * buun-llama-cpp fork. `openvino` is the upstream ggml-openvino
+ * backend capability; it selects CPU/GPU/NPU at runtime via
+ * `GGML_OPENVINO_DEVICE` and does not satisfy the Eliza-1 W4-B kernel set.
+ *
+ * Models that declare a `requiresKernel` advertise that they only run
+ * correctly under optimized llama.cpp when the matching capability is present.
+ *
+ * The set must stay in sync with `inference/AGENTS.md` §3 mandatory
+ * optimizations and with the native runtime capability probe — the
+ * capability probe is the runtime gate that refuses to start if a required
+ * kernel is missing.
+ *
+ * This (the llama.cpp-handle layer) is *not* the same enum as the
+ * bundle-manifest layer's `Eliza1Kernel`
+ * (`@elizaos/app-core/src/services/local-inference/manifest/schema`):
+ * `turboquant_q3↔turbo3`, `turboquant_q4↔turbo4`, `qjl↔qjl_full`, with
+ * `polarquant` / `turbo3_tcq` shared by name. The translation is
+ * codified there by `ELIZA1_TO_RUNTIME_KERNEL` / `RUNTIME_TO_ELIZA1_KERNEL`.
+ * `openvino` intentionally has no manifest-level Eliza-1 kernel equivalent.
+ */
+export type LocalRuntimeKernel = "turbo3" | "turbo4" | "turbo3_tcq" | "qjl_full" | "polarquant" | "openvino";
+/**
+ * llama.cpp optimization knobs that the dispatcher can wire into the
+ * FFI runtime. Values come from catalog metadata (per-model) and
+ * environment overrides (per-process). The catalog is the source of truth
+ * for which knobs are *safe* on a given quant; env vars are the operator's
+ * escape hatch and override the catalog when set.
+ */
+export interface LocalRuntimeOptimizations {
+    /** Lookahead decoding window. */
+    lookahead?: number;
+    /**
+     * Built-in n-gram drafter (no separate drafter model). Maps to
+     * Built-in n-gram speculation. `minProb` is kept for older configs and
+     * ignored by current llama.cpp ngram-mod.
+     */
+    ngramDraft?: {
+        min: number;
+        max: number;
+        minProb: number;
+    };
+    /**
+     * `--parallel N` for continuous batching. The Cache Bridge agent may bump
+     * this default at runtime; the dispatcher reads but does not override.
+     */
+    parallel?: number;
+    /**
+     * Prompt-cache chunk reuse threshold. Useful for repeated tool/system prefixes where a
+     * full slot restore is not available.
+     */
+    cacheReuse?: number;
+    /**
+     * RAM budget for prompt/KV cache files, in MiB.
+     */
+    cacheRamMb?: number;
+    /** `--batch-size N` logical batch size. */
+    batchSize?: number;
+    /** `--ubatch-size N` physical micro-batch size. */
+    ubatchSize?: number;
+    /** Continuous batching toggle (`--cont-batching` / `--no-cont-batching`). */
+    contBatching?: boolean;
+    /** Unified KV cache toggle (`--kv-unified` / `--no-kv-unified`). */
+    kvUnified?: boolean;
+    /**
+     * Number of runtime context checkpoints the cache bridge should keep for
+     * interruption/resume.
+     */
+    ctxCheckpoints?: number;
+    /** Token interval between saved context checkpoints. */
+    ctxCheckpointInterval?: number;
+    /** Host tensor op offload toggle (`--op-offload` / `--no-op-offload`). */
+    opOffload?: boolean;
+    /**
+     * Mixture-of-experts expert-tensor offload target. `"cpu"` maps to
+     * `-ot ".*=CPU"` so expert tensors stay in CPU memory and only the
+     * shared layers occupy VRAM.
+     */
+    moeOffload?: "cpu" | "none";
+    /** `--mlock` — pin model pages in RAM. */
+    mlock?: boolean;
+    /** Inverse of `--mmap`; maps to `--no-mmap`. */
+    noMmap?: boolean;
+    /** Multimodal projector path; maps to `--mmproj <path>`. */
+    mmproj?: string;
+    /** `--alias <name>` for the OpenAI-compatible model id. */
+    alias?: string;
+    /** Flash attention. */
+    flashAttention?: boolean;
+    /** Use native MTP verifier events when the runtime advertises support. */
+    nativeMtpEvents?: boolean;
+    /**
+     * Specialised kernels this model requires from optimized llama.cpp.
+     * The dispatcher uses this to pick `llama-cpp` over `capacitor-llama`
+     * regardless of `preferredBackend`, since the legacy node binding cannot
+     * provide these kernels.
+     */
+    requiresKernel?: LocalRuntimeKernel[];
+    /**
+     * Kernels that must be disabled for this model. Used when a fused build
+     * co-compiles a backend (e.g. OpenVINO) that this model is known not to
+     * support.
+     */
+    unsupportedKernels?: LocalRuntimeKernel[];
+}
+export interface LocalRuntimeAcceleration {
+    /**
+     * Prefer the optimized llama.cpp path over the node binding when the
+     * required runtime is available.
+     */
+    preferredBackend?: LocalRuntimeBackend;
+    /** Optimization knobs declared per-model. See `LocalRuntimeOptimizations`. */
+    optimizations?: LocalRuntimeOptimizations;
+    mtp?: {
+        /** Native llama.cpp MTP speculative mode. */
+        specType: "draft-mtp";
+        /**
+         * Bundle-relative path to a separate MTP drafter GGUF. Omitted for
+         * same-file MTP, where the NextN head is embedded in the main text
+         * GGUF (`qwen35.nextn_predict_layers > 0` + `blk.N.nextn.*` tensors)
+         * and no separate drafter download exists.
+         */
+        drafterFile?: string;
+        /** Default draft range passed to the native MTP runner. */
+        draftMin: number;
+        draftMax: number;
+        /** GPU layer placement for MTP heads when the runtime exposes it. */
+        gpuLayers: number | "auto";
+    };
+    kvCache?: {
+        /**
+         * llama.cpp KV cache type overrides. Stock builds support f16/q8_0;
+         * TurboQuant-capable forks add tbq3_0/tbq4_0.
+         */
+        typeK?: string;
+        typeV?: string;
+        requiresFork?: "apothic-turboquant" | "buun-llama-cpp";
+    };
+}
+/**
+ * Tokenizer family identifier for local GGUF catalog entries. Add new
+ * families here as the catalog grows.
+ */
+export type TokenizerFamily = "qwen35" | "eliza1" | "sentencepiece" | (string & {});
+export type CatalogHub = "huggingface" | "modelscope";
+export type CatalogQuantizationId = "q3_k_m" | "q4_0" | "q4_k_m" | "q5_k_m" | "q6_k" | "q8_0";
+export interface CatalogQuantizationVariant {
+    id: CatalogQuantizationId;
+    label: "3-bit" | "4-bit" | "5-bit" | "6-bit" | "8-bit";
+    ggufFile: string;
+    sizeGb: number;
+    minRamGb: number;
+    status: "published" | "planned";
+}
+export interface CatalogQuantization {
+    defaultVariantId: CatalogQuantizationId;
+    variants: CatalogQuantizationVariant[];
+}
+export type CatalogQuantizationMatrix = CatalogQuantization;
+export interface CatalogModel {
+    /** Stable Eliza id — used as the primary key. */
+    id: string;
+    displayName: string;
+    /** Hosting backend. Defaults to Hugging Face when omitted. */
+    hub?: CatalogHub;
+    /** HuggingFace repo slug, e.g. "elizaos/eliza-1". */
+    hfRepo: string;
+    /**
+     * Optional path prefix inside `hfRepo`. Eliza-1 publishes every tier under
+     * one Hugging Face model repo, so app downloads resolve bundle-relative
+     * paths like `text/eliza-1-2b-128k.gguf` under `bundles/2b/`.
+     */
+    hfPathPrefix?: string;
+    /** Exact GGUF filename in the repo. */
+    ggufFile: string;
+    /**
+     * Optional Eliza-1 bundle manifest in the same HF repo. When present, the
+     * downloader installs every file listed in the manifest and uses
+     * `ggufFile` as the primary text GGUF inside that bundle.
+     */
+    bundleManifestFile?: string;
+    /**
+     * Optional SHA-256 for `bundleManifestFile`. Publish tooling should fill
+     * this once the manifest is finalized so desktop downloads get the same
+     * manifest authenticity check as mobile. Omitted only for local/dev
+     * catalogs whose manifests are still being staged.
+     */
+    bundleManifestSha256?: string;
+    params: "360M" | "0.1B" | "0.3B" | "0.5B" | "0.8B" | "1B" | "2B" | "3B" | "4B" | "7B" | "8B" | "9B" | "14B" | "16B" | "22B" | "24B" | "27B" | "32B";
+    /** Optional human-facing parameter label when `params` is normalized. */
+    parameterLabel?: string;
+    quant: string;
+    sizeGb: number;
+    /** Minimum system RAM (GB) we recommend before offering this model. */
+    minRamGb: number;
+    category: ModelCategory;
+    bucket: ModelBucket;
+    blurb: string;
+    /**
+     * Hidden entries are installable by id and can be downloaded as companions,
+     * but are omitted from the visible Model Hub catalog.
+     */
+    hiddenFromCatalog?: boolean;
+    /** Runtime role for non-standard entries. */
+    runtimeRole?: "chat" | "mtp-drafter";
+    /** Parent chat model id when this entry is a hidden companion. */
+    companionForModelId?: string;
+    /** Extra catalog model ids to download alongside this model. */
+    companionModelIds?: string[];
+    /** Maximum context length supported by the underlying GGUF, in tokens. */
+    contextLength?: number;
+    /** Default GPU offload strategy for this model. */
+    gpuLayers?: "auto" | number;
+    /** Optional recommended hardware profile id for copy/UI sorting. */
+    gpuProfile?: string;
+    /** Available text quantization variants for this tier. */
+    quantization?: CatalogQuantization;
+    /**
+     * Tokenizer/vocabulary family this GGUF emits.
+     */
+    tokenizerFamily?: TokenizerFamily;
+    /** Voice backends whose artifacts ship in this Eliza-1 bundle. */
+    voiceBackends?: ReadonlyArray<"kokoro" | "omnivoice">;
+    /**
+     * Provenance for the Eliza-1 v1 release shape (`releaseState=base-v1`):
+     * Eliza-1 v1 is the upstream BASE models — GGUF-converted via the
+     * elizaOS/llama.cpp fork and fully Eliza-optimized (every quant/kernel
+     * trick in `packages/inference/AGENTS.md` §3) — but NOT fine-tuned.
+     * Each entry records which upstream HF repo every shipped bundle
+     * component comes from. The keys are the bundle's component slots; the
+     * values are upstream repo ids (and an optional file). This is the
+     * "base, not fine-tuned" provenance in the catalog — it must match the
+     * `provenance.sourceModels` block in the tier's `eliza-1.manifest.json`.
+     * `finetuned: false` until v2.
+     */
+    sourceModel?: {
+        finetuned: false;
+        components: Partial<Record<"text" | "voice" | "asr" | "vad" | "embedding" | "vision" | "mtp", {
+            repo: string;
+            file?: string;
+        }>>;
+    };
+    /** Runtime-specific acceleration metadata. */
+    runtime?: LocalRuntimeAcceleration;
+    /**
+     * Whether this tier's bundle is published
+     * on Hugging Face yet. Defaults to `"published"` when omitted, which
+     * preserves prior behaviour for any catalog entry that doesn't set it.
+     *
+     *   - `"published"`: HF repo has a real manifest + weights.
+     *     The recommender may route first-run users here.
+     *   - `"pending"`: catalog points at a tier whose HF repo is not
+     *     usable yet (404, empty manifest, or `releaseState=local-standin`).
+     *     `recommendForFirstRun` falls through
+     *     to the next ladder candidate. Used to keep the catalog reflecting
+     *     the intended product shape while the publish pipeline catches up.
+     *
+     * See elizaOS/eliza#7629.
+     */
+    publishStatus?: "published" | "pending";
+}
+export type HardwareFitLevel = "fits" | "tight" | "wontfit";
+export interface MobileHardwareProbe {
+    platform: "ios" | "android" | "web";
+    deviceModel?: string;
+    machineId?: string;
+    osVersion?: string;
+    isSimulator?: boolean;
+    availableRamGb?: number | null;
+    freeStorageGb?: number | null;
+    lowPowerMode?: boolean;
+    thermalState?: "nominal" | "fair" | "serious" | "critical" | "unknown";
+    gpuSupported?: boolean;
+    mtpSupported?: boolean;
+    mtpReason?: string;
+    source?: "native" | "adapter-fallback";
+}
+export interface CpuFeatureProbe {
+    /** Arm NEON/Advanced SIMD (`asimd` in Linux cpuinfo). */
+    neon?: boolean;
+    /** Arm dot-product instructions (`asimddp`/`dotprod`). */
+    dotprod?: boolean;
+    /** Arm int8 matrix multiply instructions. */
+    i8mm?: boolean;
+    /** Arm Scalable Vector Extension. */
+    sve?: boolean;
+    /** Arm Scalable Vector Extension v2. */
+    sve2?: boolean;
+}
+export interface HardwareProbe {
+    totalRamGb: number;
+    freeRamGb: number;
+    /** Null when no supported GPU is available (CPU-only). */
+    gpu: {
+        backend: "cuda" | "metal" | "vulkan";
+        totalVramGb: number;
+        freeVramGb: number;
+    } | null;
+    cpuCores: number;
+    /** Optional CPU ISA features detected from the host OS. */
+    cpuFeatures?: CpuFeatureProbe;
+    platform: NodeJS.Platform;
+    arch: NodeJS.Architecture;
+    /** True on Apple Silicon (shared memory — large models are viable on 16GB+). */
+    appleSilicon: boolean;
+    /** Recommended default bucket based on available memory. */
+    recommendedBucket: ModelBucket;
+    /**
+     * Source of the probe. `"capacitor-llama"` when GPU values come from a loaded
+     * Capacitor-llama context (mobile binding or desktop bun:ffi). `"os-fallback"`
+     * means the probe was synthesized from OS-level CPU/RAM only because no
+     * llama.cpp context was available to query for GPU/VRAM state.
+     */
+    source: "capacitor-llama" | "os-fallback";
+    /** OpenVINO CPU/GPU/NPU availability hints for Intel hosts. */
+    openvino?: OpenVinoHardwareProbe;
+    /** Mobile-only details used for minspec, storage, and native runtime gating. */
+    mobile?: MobileHardwareProbe;
+}
+export type DownloadState = "queued" | "downloading" | "completed" | "failed" | "cancelled";
+export interface DownloadJob {
+    jobId: string;
+    modelId: string;
+    state: DownloadState;
+    /** Bytes transferred so far. */
+    received: number;
+    /** Total bytes expected (from Content-Length or HEAD). */
+    total: number;
+    /** Moving-average bytes/sec over the last few seconds. */
+    bytesPerSec: number;
+    /** Milliseconds remaining based on current rate. Null when unknown. */
+    etaMs: number | null;
+    startedAt: string;
+    updatedAt: string;
+    /** Set when state === "failed". */
+    error?: string;
+}
+export interface LocalInferenceDownloadStatus {
+    state: DownloadState | "missing";
+    receivedBytes: number;
+    totalBytes: number;
+    percent: number | null;
+    bytesPerSec: number;
+    etaMs: number | null;
+    updatedAt: string | null;
+    errors: string[];
+}
+export interface ActiveModelState {
+    modelId: string | null;
+    loadedAt: string | null;
+    /**
+     * Human-readable load status. "idle" means nothing loaded.
+     * "loading" is set while we're swapping models.
+     */
+    status: "idle" | "loading" | "ready" | "error";
+    error?: string;
+    /**
+     * Effective KV-cache configuration the loader applied. Populated on
+     * `status === "ready"`; null while loading or on error. The benchmark
+     * harness relies on these to verify per-load overrides actually took
+     * effect (a 128k contextSize request that silently fell back to 8k is
+     * exactly the bug the per-load override path exists to prevent).
+     */
+    loadedContextSize?: number | null;
+    loadedCacheTypeK?: string | null;
+    loadedCacheTypeV?: string | null;
+    loadedGpuLayers?: number | null;
+}
+export interface DownloadEvent {
+    type: "progress" | "completed" | "failed" | "cancelled";
+    job: DownloadJob;
+}
+export interface LocalInferenceSlotReadiness {
+    slot: TextGenerationSlot;
+    assigned: boolean;
+    assignedModelId: string | null;
+    displayName: string | null;
+    primaryDownloaded: boolean;
+    downloaded: boolean;
+    active: boolean;
+    ready: boolean;
+    state: "unassigned" | "missing" | "downloading" | "downloaded" | "active" | "failed" | "cancelled";
+    requiredModelIds: string[];
+    missingModelIds: string[];
+    installedBytes: number;
+    expectedBytes: number;
+    download: LocalInferenceDownloadStatus;
+    errors: string[];
+}
+export interface LocalInferenceReadiness {
+    updatedAt: string;
+    slots: Record<TextGenerationSlot, LocalInferenceSlotReadiness>;
+}
+export interface ModelHubSnapshot {
+    catalog: CatalogModel[];
+    installed: InstalledModel[];
+    active: ActiveModelState;
+    downloads: DownloadJob[];
+    hardware: HardwareProbe;
+    assignments: ModelAssignments;
+    textReadiness: LocalInferenceReadiness;
+}
+//# sourceMappingURL=types.d.ts.map

package/local-inference/types.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/local-inference/types.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAEH,wDAAwD;AACxD,MAAM,MAAM,cAAc,GACtB,YAAY,GACZ,YAAY,GACZ,gBAAgB,GAChB,gBAAgB,GAChB,eAAe,CAAC;AAEpB,uEAAuE;AACvE,MAAM,MAAM,kBAAkB,GAAG,OAAO,CACtC,cAAc,EACd,YAAY,GAAG,YAAY,CAC5B,CAAC;AAEF,eAAO,MAAM,iBAAiB,EAAE,cAAc,EAM7C,CAAC;AAEF,eAAO,MAAM,qBAAqB,EAAE,kBAAkB,EAGrD,CAAC;AAEF;;;;GAIG;AACH,MAAM,MAAM,gBAAgB,GAAG,OAAO,CAAC,MAAM,CAAC,cAAc,EAAE,MAAM,CAAC,CAAC,CAAC;AAEvE;;GAEG;AACH,MAAM,WAAW,cAAc;IAC7B,uEAAuE;IACvE,EAAE,EAAE,MAAM,CAAC;IACX,WAAW,EAAE,MAAM,CAAC;IACpB,8CAA8C;IAC9C,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,MAAM,CAAC;IAClB;;;;OAIG;IACH,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,4EAA4E;IAC5E,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,2DAA2D;IAC3D,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,6DAA6D;IAC7D,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,6EAA6E;IAC7E,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,0CAA0C;IAC1C,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,2CAA2C;IAC3C,WAAW,EAAE,MAAM,CAAC;IACpB,+DAA+D;IAC/D,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,4EAA4E;IAC5E,MAAM,EAAE,gBAAgB,GAAG,eAAe,CAAC;IAC3C;;;OAGG;IACH,cAAc,CAAC,EACX,WAAW,GACX,KAAK,GACL,QAAQ,GACR,aAAa,GACb,gBAAgB,CAAC;IACrB,qFAAqF;IACrF,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,mGAAmG;IACnG,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB;;;;;;OAMG;IACH,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,WAAW,CAAC,EAAE,MAAM,GAAG,aAAa,CAAC;IACrC,YAAY,CAAC,EAAE,MAAM,CAAC;CACvB;AAED,MAAM,MAAM,WAAW,GAAG,OAAO,GAAG,KAAK,GAAG,OAAO,GAAG,IAAI,CAAC;AAE3D,MAAM,MAAM,aAAa,GAAG,MAAM,GAAG,MAAM,GAAG,OAAO,GAAG,MAAM,GAAG,WAAW,CAAC;AAE7E,MAAM,MAAM,mBAAmB,GAAG,iBAAiB,GAAG,WAAW,CAAC;AAElE,MAAM,MAAM,kBAAkB,GAAG,KAAK,GAAG,KAAK,GAAG,KAAK,CAAC;AAEvD,MAAM,WAAW,qBAAqB;IACpC;;;OAGG;IACH,gBAAgB,EAAE,OAAO,CAAC;IAC1B,OAAO,EAAE,kBAAkB,EAAE,CAAC;IAC9B;;;OAGG;IACH,GAAG,EAAE;QACH,WAAW,EAAE,MAAM,EAAE,CAAC;QACtB,mBAAmB,EAAE,OAAO,CAAC;QAC7B,oBAAoB,EAAE,MAAM,EAAE,CAAC;KAChC,CAAC;IACF;;;OAGG;IACH,GAAG,EAAE;QACH,UAAU,EAAE,MAAM,EAAE,CAAC;KACtB,CAAC;IACF,iEAAiE;IACjE,oBAAoB,EAAE,kBAAkB,GAAG,IAAI,CAAC;IAChD,QAAQ,EAAE,MAAM,EAAE,CAAC;CACpB;AAED;;;;;;;;;;;;;;;;;;;;;;GAsBG;AACH,MAAM,MAAM,kBAAkB,GAC1B,QAAQ,GACR,QAAQ,GACR,YAAY,GACZ,UAAU,GACV,YAAY,GACZ,UAAU,CAAC;AAEf;;;;;;GAMG;AACH,MAAM,WAAW,yBAAyB;IACxC,iCAAiC;IACjC,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB;;;;OAIG;IACH,UAAU,CAAC,EAAE;QAAE,GAAG,EAAE,MAAM,CAAC;QAAC,GAAG,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAA;KAAE,CAAC;IAC3D;;;OAGG;IACH,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB;;;OAGG;IACH,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB;;OAEG;IACH,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,2CAA2C;IAC3C,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,mDAAmD;IACnD,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,6EAA6E;IAC7E,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,oEAAoE;IACpE,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB;;;OAGG;IACH,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,wDAAwD;IACxD,qBAAqB,CAAC,EAAE,MAAM,CAAC;IAC/B,0EAA0E;IAC1E,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB;;;;OAIG;IACH,UAAU,CAAC,EAAE,KAAK,GAAG,MAAM,CAAC;IAC5B,0CAA0C;IAC1C,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,gDAAgD;IAChD,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,4DAA4D;IAC5D,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,2DAA2D;IAC3D,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,uBAAuB;IACvB,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,0EAA0E;IAC1E,eAAe,CAAC,EAAE,OAAO,CAAC;IAC1B;;;;;OAKG;IACH,cAAc,CAAC,EAAE,kBAAkB,EAAE,CAAC;IACtC;;;;OAIG;IACH,kBAAkB,CAAC,EAAE,kBAAkB,EAAE,CAAC;CAC3C;AAED,MAAM,WAAW,wBAAwB;IACvC;;;OAGG;IACH,gBAAgB,CAAC,EAAE,mBAAmB,CAAC;IACvC,8EAA8E;IAC9E,aAAa,CAAC,EAAE,yBAAyB,CAAC;IAC1C,GAAG,CAAC,EAAE;QACJ,6CAA6C;QAC7C,QAAQ,EAAE,WAAW,CAAC;QACtB;;;;;WAKG;QACH,WAAW,CAAC,EAAE,MAAM,CAAC;QACrB,2DAA2D;QAC3D,QAAQ,EAAE,MAAM,CAAC;QACjB,QAAQ,EAAE,MAAM,CAAC;QACjB,qEAAqE;QACrE,SAAS,EAAE,MAAM,GAAG,MAAM,CAAC;KAC5B,CAAC;IACF,OAAO,CAAC,EAAE;QACR;;;WAGG;QACH,KAAK,CAAC,EAAE,MAAM,CAAC;QACf,KAAK,CAAC,EAAE,MAAM,CAAC;QACf,YAAY,CAAC,EAAE,oBAAoB,GAAG,gBAAgB,CAAC;KACxD,CAAC;CACH;AAED;;;GAGG;AACH,MAAM,MAAM,eAAe,GACvB,QAAQ,GACR,QAAQ,GACR,eAAe,GACf,CAAC,MAAM,GAAG,EAAE,CAAC,CAAC;AAElB,MAAM,MAAM,UAAU,GAAG,aAAa,GAAG,YAAY,CAAC;AAEtD,MAAM,MAAM,qBAAqB,GAC7B,QAAQ,GACR,MAAM,GACN,QAAQ,GACR,QAAQ,GACR,MAAM,GACN,MAAM,CAAC;AAEX,MAAM,WAAW,0BAA0B;IACzC,EAAE,EAAE,qBAAqB,CAAC;IAC1B,KAAK,EAAE,OAAO,GAAG,OAAO,GAAG,OAAO,GAAG,OAAO,GAAG,OAAO,CAAC;IACvD,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,WAAW,GAAG,SAAS,CAAC;CACjC;AAED,MAAM,WAAW,mBAAmB;IAClC,gBAAgB,EAAE,qBAAqB,CAAC;IACxC,QAAQ,EAAE,0BAA0B,EAAE,CAAC;CACxC;AAED,MAAM,MAAM,yBAAyB,GAAG,mBAAmB,CAAC;AAE5D,MAAM,WAAW,YAAY;IAC3B,iDAAiD;IACjD,EAAE,EAAE,MAAM,CAAC;IACX,WAAW,EAAE,MAAM,CAAC;IACpB,8DAA8D;IAC9D,GAAG,CAAC,EAAE,UAAU,CAAC;IACjB,qDAAqD;IACrD,MAAM,EAAE,MAAM,CAAC;IACf;;;;OAIG;IACH,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,uCAAuC;IACvC,QAAQ,EAAE,MAAM,CAAC;IACjB;;;;OAIG;IACH,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B;;;;;OAKG;IACH,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAC9B,MAAM,EACF,MAAM,GACN,MAAM,GACN,MAAM,GACN,MAAM,GACN,MAAM,GACN,IAAI,GACJ,IAAI,GACJ,IAAI,GACJ,IAAI,GACJ,IAAI,GACJ,IAAI,GACJ,IAAI,GACJ,KAAK,GACL,KAAK,GACL,KAAK,GACL,KAAK,GACL,KAAK,GACL,KAAK,CAAC;IACV,yEAAyE;IACzE,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,uEAAuE;IACvE,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,aAAa,CAAC;IACxB,MAAM,EAAE,WAAW,CAAC;IACpB,KAAK,EAAE,MAAM,CAAC;IACd;;;OAGG;IACH,iBAAiB,CAAC,EAAE,OAAO,CAAC;IAC5B,6CAA6C;IAC7C,WAAW,CAAC,EAAE,MAAM,GAAG,aAAa,CAAC;IACrC,kEAAkE;IAClE,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,gEAAgE;IAChE,iBAAiB,CAAC,EAAE,MAAM,EAAE,CAAC;IAC7B,0EAA0E;IAC1E,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,mDAAmD;IACnD,SAAS,CAAC,EAAE,MAAM,GAAG,MAAM,CAAC;IAC5B,oEAAoE;IACpE,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,0DAA0D;IAC1D,YAAY,CAAC,EAAE,mBAAmB,CAAC;IACnC;;OAEG;IACH,eAAe,CAAC,EAAE,eAAe,CAAC;IAClC,kEAAkE;IAClE,aAAa,CAAC,EAAE,aAAa,CAAC,QAAQ,GAAG,WAAW,CAAC,CAAC;IACtD;;;;;;;;;;;OAWG;IACH,WAAW,CAAC,EAAE;QACZ,SAAS,EAAE,KAAK,CAAC;QACjB,UAAU,EAAE,OAAO,CACjB,MAAM,CACJ,MAAM,GAAG,OAAO,GAAG,KAAK,GAAG,KAAK,GAAG,WAAW,GAAG,QAAQ,GAAG,KAAK,EACjE;YAAE,IAAI,EAAE,MAAM,CAAC;YAAC,IAAI,CAAC,EAAE,MAAM,CAAA;SAAE,CAChC,CACF,CAAC;KACH,CAAC;IACF,8CAA8C;IAC9C,OAAO,CAAC,EAAE,wBAAwB,CAAC;IACnC;;;;;;;;;;;;;;OAcG;IACH,aAAa,CAAC,EAAE,WAAW,GAAG,SAAS,CAAC;CACzC;AAED,MAAM,MAAM,gBAAgB,GAAG,MAAM,GAAG,OAAO,GAAG,SAAS,CAAC;AAE5D,MAAM,WAAW,mBAAmB;IAClC,QAAQ,EAAE,KAAK,GAAG,SAAS,GAAG,KAAK,CAAC;IACpC,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,OAAO,CAAC;IACtB,cAAc,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC/B,aAAa,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC9B,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,YAAY,CAAC,EAAE,SAAS,GAAG,MAAM,GAAG,SAAS,GAAG,UAAU,GAAG,SAAS,CAAC;IACvE,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,MAAM,CAAC,EAAE,QAAQ,GAAG,kBAAkB,CAAC;CACxC;AAED,MAAM,WAAW,eAAe;IAC9B,yDAAyD;IACzD,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,0DAA0D;IAC1D,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,6CAA6C;IAC7C,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,qCAAqC;IACrC,GAAG,CAAC,EAAE,OAAO,CAAC;IACd,wCAAwC;IACxC,IAAI,CAAC,EAAE,OAAO,CAAC;CAChB;AAED,MAAM,WAAW,aAAa;IAC5B,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,MAAM,CAAC;IAClB,0DAA0D;IAC1D,GAAG,EAAE;QACH,OAAO,EAAE,MAAM,GAAG,OAAO,GAAG,QAAQ,CAAC;QACrC,WAAW,EAAE,MAAM,CAAC;QACpB,UAAU,EAAE,MAAM,CAAC;KACpB,GAAG,IAAI,CAAC;IACT,QAAQ,EAAE,MAAM,CAAC;IACjB,2DAA2D;IAC3D,WAAW,CAAC,EAAE,eAAe,CAAC;IAC9B,QAAQ,EAAE,MAAM,CAAC,QAAQ,CAAC;IAC1B,IAAI,EAAE,MAAM,CAAC,YAAY,CAAC;IAC1B,gFAAgF;IAChF,YAAY,EAAE,OAAO,CAAC;IACtB,4DAA4D;IAC5D,iBAAiB,EAAE,WAAW,CAAC;IAC/B;;;;;OAKG;IACH,MAAM,EAAE,iBAAiB,GAAG,aAAa,CAAC;IAC1C,+DAA+D;IAC/D,QAAQ,CAAC,EAAE,qBAAqB,CAAC;IACjC,gFAAgF;IAChF,MAAM,CAAC,EAAE,mBAAmB,CAAC;CAC9B;AAED,MAAM,MAAM,aAAa,GACrB,QAAQ,GACR,aAAa,GACb,WAAW,GACX,QAAQ,GACR,WAAW,CAAC;AAEhB,MAAM,WAAW,WAAW;IAC1B,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,EAAE,aAAa,CAAC;IACrB,gCAAgC;IAChC,QAAQ,EAAE,MAAM,CAAC;IACjB,0DAA0D;IAC1D,KAAK,EAAE,MAAM,CAAC;IACd,0DAA0D;IAC1D,WAAW,EAAE,MAAM,CAAC;IACpB,uEAAuE;IACvE,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,mCAAmC;IACnC,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,4BAA4B;IAC3C,KAAK,EAAE,aAAa,GAAG,SAAS,CAAC;IACjC,aAAa,EAAE,MAAM,CAAC;IACtB,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IACvB,WAAW,EAAE,MAAM,CAAC;IACpB,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,MAAM,EAAE,MAAM,EAAE,CAAC;CAClB;AAED,MAAM,WAAW,gBAAgB;IAC/B,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IACvB,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB;;;OAGG;IACH,MAAM,EAAE,MAAM,GAAG,SAAS,GAAG,OAAO,GAAG,OAAO,CAAC;IAC/C,KAAK,CAAC,EAAE,MAAM,CAAC;IACf;;;;;;OAMG;IACH,iBAAiB,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAClC,gBAAgB,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACjC,gBAAgB,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACjC,eAAe,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;CACjC;AAED,MAAM,WAAW,aAAa;IAC5B,IAAI,EAAE,UAAU,GAAG,WAAW,GAAG,QAAQ,GAAG,WAAW,CAAC;IACxD,GAAG,EAAE,WAAW,CAAC;CAClB;AAED,MAAM,WAAW,2BAA2B;IAC1C,IAAI,EAAE,kBAAkB,CAAC;IACzB,QAAQ,EAAE,OAAO,CAAC;IAClB,eAAe,EAAE,MAAM,GAAG,IAAI,CAAC;IAC/B,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B,iBAAiB,EAAE,OAAO,CAAC;IAC3B,UAAU,EAAE,OAAO,CAAC;IACpB,MAAM,EAAE,OAAO,CAAC;IAChB,KAAK,EAAE,OAAO,CAAC;IACf,KAAK,EACD,YAAY,GACZ,SAAS,GACT,aAAa,GACb,YAAY,GACZ,QAAQ,GACR,QAAQ,GACR,WAAW,CAAC;IAChB,gBAAgB,EAAE,MAAM,EAAE,CAAC;IAC3B,eAAe,EAAE,MAAM,EAAE,CAAC;IAC1B,cAAc,EAAE,MAAM,CAAC;IACvB,aAAa,EAAE,MAAM,CAAC;IACtB,QAAQ,EAAE,4BAA4B,CAAC;IACvC,MAAM,EAAE,MAAM,EAAE,CAAC;CAClB;AAED,MAAM,WAAW,uBAAuB;IACtC,SAAS,EAAE,MAAM,CAAC;IAClB,KAAK,EAAE,MAAM,CAAC,kBAAkB,EAAE,2BAA2B,CAAC,CAAC;CAChE;AAED,MAAM,WAAW,gBAAgB;IAC/B,OAAO,EAAE,YAAY,EAAE,CAAC;IACxB,SAAS,EAAE,cAAc,EAAE,CAAC;IAC5B,MAAM,EAAE,gBAAgB,CAAC;IACzB,SAAS,EAAE,WAAW,EAAE,CAAC;IACzB,QAAQ,EAAE,aAAa,CAAC;IACxB,WAAW,EAAE,gBAAgB,CAAC;IAC9B,aAAa,EAAE,uBAAuB,CAAC;CACxC"}

package/local-inference/types.js

@@ -0,0 +1,22 @@
+/**
+ * Local inference shared types.
+ *
+ * Shared contracts referenced by the server-side service in
+ * `@elizaos/app-core` and the UI client in `@elizaos/ui`.
+ *
+ * Server-only logic (KV cache management, native runtime lifecycle,
+ * conversation registry, metrics scraping) stays in `app-core`; only
+ * the type contracts live here.
+ */
+export const AGENT_MODEL_SLOTS = [
+    "TEXT_SMALL",
+    "TEXT_LARGE",
+    "TEXT_EMBEDDING",
+    "TEXT_TO_SPEECH",
+    "TRANSCRIPTION",
+];
+export const TEXT_GENERATION_SLOTS = [
+    "TEXT_SMALL",
+    "TEXT_LARGE",
+];
+//# sourceMappingURL=types.js.map

package/local-inference/types.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"types.js","sourceRoot":"","sources":["../../src/local-inference/types.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAgBH,MAAM,CAAC,MAAM,iBAAiB,GAAqB;IACjD,YAAY;IACZ,YAAY;IACZ,gBAAgB;IAChB,gBAAgB;IAChB,eAAe;CAChB,CAAC;AAEF,MAAM,CAAC,MAAM,qBAAqB,GAAyB;IACzD,YAAY;IACZ,YAAY;CACb,CAAC"}