import type { BaseDirsConfigError } from '../utils/base-dirs.js';
import type { ContentFormat } from '../utils/raw-data-utils.js';
import type { SourceContext } from '../utils/source-context.js';
import type { QualityProfile } from '../utils/visual-profile.js';
import type { GroupingMode } from '../vectordb/index.js';
/**
 * Fields shared by both `RAGServerConfig` shapes (legacy single-root and
 * multi-root). Extracted so the union below only needs to describe the
 * `baseDir` / `baseDirs` axis.
 */
interface RAGServerConfigBase {
    /** LanceDB database path */
    dbPath: string;
    /** Transformers.js model path */
    modelName: string;
    /** Model cache directory */
    cacheDir: string;
    /** Maximum file size (100MB) */
    maxFileSize: number;
    /** Compute device (cpu, webgpu, dml, etc) */
    device?: string;
    /** Embedding quantization dtype (fp32, fp16, q8, int8, ...). Unset → fp32. */
    dtype?: string;
    /** Maximum distance threshold for quality filtering (optional) */
    maxDistance?: number;
    /** Grouping mode for quality filtering (optional) */
    grouping?: GroupingMode;
    /** Hybrid search weight for BM25 (0.0 = vector only, 1.0 = BM25 only, default 0.6) */
    hybridWeight?: number;
    /** Maximum number of files to keep in search results (optional) */
    maxFiles?: number;
    /** Minimum chunk length in characters (optional, default: 50) */
    chunkMinLength?: number;
    /** Embed each chunk behind its document title (optional, default: false) */
    titlePrefix?: boolean;
    /** Add section paths to chunk embeddings when they fit (default: false). */
    headingPrefix?: boolean;
    /** Store bounded PDF regions and Mammoth-produced DOCX images. */
    storeImages?: boolean;
    /** External reranker command, as configured. Unset disables reranking. */
    rerankCommand?: string;
    /** Per-rerank-call budget in milliseconds (default: 10000). */
    rerankTimeoutMs?: number;
    /**
     * Normal-path roots, index-aligned with the realpath'd `baseDirs` boundary,
     * used for user-facing scan and display so paths match the stored DB keys.
     * Legacy `{ baseDir }` callers fall back to `baseDirs`.
     */
    rawBaseDirs?: readonly string[];
    /** `IGNORE_PATHS` globs, relative to the root containing each file. */
    ignorePatterns?: readonly string[];
    /** Configuration validation warnings to surface to users via MCP annotations */
    configWarnings?: string[];
    /**
     * When present the server is in degraded mode: `status` stays callable for
     * diagnosis, while root-dependent tools surface this first.
     */
    configError?: BaseDirsConfigError;
}
/**
 * RAGServer configuration. Exactly one of `baseDir` (legacy) or `baseDirs`
 * must be supplied; the constructor normalizes both to `baseDirs` and derives
 * the legacy accessor as `baseDirs[0]`.
 */
export type RAGServerConfig = (RAGServerConfigBase & {
    /** Document base directory (legacy single-root shape). */
    baseDir: string;
    baseDirs?: undefined;
}) | (RAGServerConfigBase & {
    /** One or more allowed document base directories (multi-root shape). */
    baseDirs: string[];
    baseDir?: undefined;
});
/**
 * query_documents tool input
 */
export interface QueryDocumentsInput {
    /** Natural language query */
    query: string;
    /** Number of results to retrieve (default 10) */
    limit?: number;
    /** Path prefix scope (one or a list); the parser normalizes to `string[]`. */
    scope?: string | string[];
}
/**
 * list_files tool input
 */
export interface ListFilesInput {
    /** Path prefix scope (one or a list); the parser normalizes to `string[]`. */
    scope?: string | string[];
}
/**
 * ingest_file tool input
 */
export interface IngestFileInput {
    /** File path */
    filePath: string;
    /**
     * When true and `filePath` is a PDF, VLM captioning runs. Silently coerced to
     * text-only for non-PDFs. The handler still checks at runtime, because MCP
     * arguments arrive as `unknown`.
     */
    visual?: boolean;
    /**
     * Visual-quality profile when `visual` is true. Some MCP clients send the
     * empty string for unspecified optional parameters; the transport decoder
     * normalizes it before this type reaches the handler.
     */
    visualQuality?: QualityProfile;
}
/**
 * ingest_data tool input metadata
 */
interface IngestDataMetadata {
    /** Source identifier: URL ("https://...") or custom ID ("clipboard://2024-12-30") */
    source: string;
    /** Content format */
    format: ContentFormat;
}
/**
 * ingest_data tool input
 */
export interface IngestDataInput {
    /** Content to ingest (text, HTML, or Markdown) */
    content: string;
    /** Content metadata */
    metadata: IngestDataMetadata;
}
/**
 * delete_file tool input
 * Either filePath or source must be provided
 */
export type DeleteFileInput = {
    filePath: string;
    source?: never;
} | {
    source: string;
    filePath?: never;
};
/**
 * delete_file tool output
 */
export interface DeleteFileResult {
    /** Resolved file path used for the delete operation */
    filePath: string;
    /** True when the delete operation completed (idempotent; not "something was removed") */
    deleted: true;
    /** Number of vector chunks removed from the database */
    removedChunks: number;
    /** True when ingested chunks and/or raw-data artifacts existed before delete */
    existed: boolean;
    /** Timestamp */
    timestamp: string;
}
/**
 * ingest_file tool output
 */
export interface IngestResult {
    /** File path */
    filePath: string;
    /** Chunk count */
    chunkCount: number;
    /** Timestamp */
    timestamp: string;
    /** Document title extracted from file content (display-only, not used for scoring) */
    fileTitle: string | null;
}
/**
 * One file found under an effective base directory. `baseDir` is always
 * present, including single-root configs — additive over the legacy shape, so
 * clients that ignore it keep working.
 */
export type FileEntry = {
    filePath: string;
    baseDir: string;
    ingested: true;
    chunkCount: number;
    timestamp: string;
} | {
    filePath: string;
    baseDir: string;
    ingested: false;
};
/**
 * list_files tool output — entry for content ingested via ingest_data,
 * or an orphaned DB entry whose file no longer exists on disk
 */
export type SourceEntry = {
    source: string;
    chunkCount: number;
    timestamp: string;
} | {
    filePath: string;
    chunkCount: number;
    timestamp: string;
};
/**
 * list_files output.
 *
 * `baseDir` remains as `baseDirs[0]` for clients written against the
 * single-root shape. Duplicate paths across roots collapse to the first
 * occurrence in root order. `sources` holds raw-data and orphaned DB entries,
 * which no root produced and so carry no `baseDir`.
 */
export interface ListFilesResult {
    baseDir: string;
    baseDirs: string[];
    files: FileEntry[];
    sources: SourceEntry[];
}
/**
 * query_documents tool output
 */
export interface QueryResult {
    sourceContext?: SourceContext;
    /** File path */
    filePath: string;
    /** Chunk index */
    chunkIndex: number;
    /** Text */
    text: string;
    /** Similarity score */
    score: number;
    /** Original source (only for raw-data files, e.g., URLs ingested via ingest_data) */
    source?: string;
    /** Document title extracted from file content (display-only, not used for scoring) */
    fileTitle: string | null;
}
/**
 * read_chunk_neighbors tool input.
 * Exactly one of filePath / source must be provided (XOR).
 */
export type ReadChunkNeighborsInput = DeleteFileInput & {
    /** Target chunk index (zero-based, required, non-negative integer). */
    chunkIndex: number;
    /** Number of chunks before the target to include (default 2, non-negative integer). */
    before?: number;
    /** Number of chunks after the target to include (default 2, non-negative integer). */
    after?: number;
};
/**
 * read_chunk_neighbors output item. `isTarget` is true only for the requested
 * target when it exists; `source` appears only on raw-data rows. `fileTitle`
 * mirrors `QueryResult` so results are drop-in with query_documents.
 */
export interface ReadChunkNeighborsResultItem {
    sourceContext?: SourceContext;
    /** File path */
    filePath: string;
    /** Chunk index */
    chunkIndex: number;
    /** Text */
    text: string;
    /** True iff this chunk's chunkIndex matches the requested target. */
    isTarget: boolean;
    /** Original source (only for raw-data files, e.g., URLs ingested via ingest_data). */
    source?: string;
    /** Document title extracted from file content (display-only, not used for scoring) */
    fileTitle: string | null;
}
/**
 * sync_start tool input
 */
export interface SyncStartInput {
    /** Absolute path inside a configured root. Omitted means every configured root. */
    path?: string;
}
/**
 * sync_status tool input
 */
export interface SyncStatusInput {
    /** Identifier returned by sync_start. */
    jobId: string;
}
/**
 * Lifecycle of the one process-local sync job the server retains. There is no
 * cancelled, queued, or recovered state: the record lives only for the server
 * process and a new job replaces a terminal one.
 */
export type SyncJobState = 'running' | 'succeeded' | 'failed';
/**
 * Per-file outcome counters of a sync job. Structurally identical to the sync
 * core's counters, restated here because `server/` does not depend on
 * `features/`.
 */
export interface SyncSummary {
    /** Files re-ingested because they were new, changed, or stored inconsistently. */
    upserted: number;
    /** Files whose stored content identity already matched the disk bytes. */
    skipped: number;
    /** Present files that produced zero chunks; their prior rows and hash are kept. */
    empty: number;
    /** Indexed files removed from the index. Counts files, not rows, and is outside `completed`. */
    pruned: number;
}
/**
 * sync_status output — the whole pollable record for one job.
 *
 * `total` stays `null` until the scan has counted the disk files, and
 * `completed` never exceeds a non-null `total`. A job succeeds only when
 * `error` is `null`; a failure carries one message, naming the file when
 * the failure was per-file.
 */
export interface SyncStatusResult {
    jobId: string;
    state: SyncJobState;
    total: number | null;
    completed: number;
    summary: SyncSummary;
    /**
     * Scanner coverage warnings, carried as JSON rather than as the content
     * blocks `list_files` uses, because status is a single pollable record.
     */
    warnings: string[];
    error: string | null;
}
export {};
//# sourceMappingURL=types.d.ts.map