import { z } from 'zod';
import { type Tool, type ToolCallError, type ToolCallWithCall } from './tools.js';
import type { ToolDialect } from './completion-stream.js';
export declare const completionStatsSchema: z.ZodObject<{
    timeToFirstToken: z.ZodOptional<z.ZodNumber>;
    tokensPerSecond: z.ZodOptional<z.ZodNumber>;
    promptTokensPerSecond: z.ZodOptional<z.ZodNumber>;
    cacheTokens: z.ZodOptional<z.ZodNumber>;
    promptTokens: z.ZodOptional<z.ZodNumber>;
    generatedTokens: z.ZodOptional<z.ZodNumber>;
    emittedTokens: z.ZodOptional<z.ZodNumber>;
    avgConcurrentSeq: z.ZodOptional<z.ZodNumber>;
    toolDefinitionsDropped: z.ZodOptional<z.ZodNumber>;
    backendDevice: z.ZodOptional<z.ZodEnum<{
        cpu: "cpu";
        gpu: "gpu";
    }>>;
}, z.core.$strip>;
export type CompletionStats = z.infer<typeof completionStatsSchema>;
export declare const seqSchema: z.ZodNumber;
export declare const contentDeltaEventSchema: z.ZodObject<{
    type: z.ZodLiteral<"contentDelta">;
    seq: z.ZodNumber;
    text: z.ZodString;
}, z.core.$strip>;
export declare const rawDeltaEventSchema: z.ZodObject<{
    type: z.ZodLiteral<"rawDelta">;
    seq: z.ZodNumber;
    text: z.ZodString;
}, z.core.$strip>;
export declare const thinkingDeltaEventSchema: z.ZodObject<{
    type: z.ZodLiteral<"thinkingDelta">;
    seq: z.ZodNumber;
    text: z.ZodString;
}, z.core.$strip>;
export declare const toolCallEventSchema: z.ZodObject<{
    type: z.ZodLiteral<"toolCall">;
    seq: z.ZodNumber;
    call: z.ZodObject<{
        id: z.ZodString;
        name: z.ZodString;
        arguments: z.ZodRecord<z.ZodString, z.ZodUnknown>;
        raw: z.ZodOptional<z.ZodString>;
    }, z.core.$strip>;
}, z.core.$strip>;
export declare const toolErrorEventSchema: z.ZodObject<{
    type: z.ZodLiteral<"toolError">;
    seq: z.ZodNumber;
    error: z.ZodObject<{
        code: z.ZodEnum<{
            PARSE_ERROR: "PARSE_ERROR";
            VALIDATION_ERROR: "VALIDATION_ERROR";
            UNKNOWN_TOOL: "UNKNOWN_TOOL";
        }>;
        message: z.ZodString;
        raw: z.ZodOptional<z.ZodString>;
    }, z.core.$strip>;
}, z.core.$strip>;
export declare const statsEventSchema: z.ZodObject<{
    type: z.ZodLiteral<"completionStats">;
    seq: z.ZodNumber;
    stats: z.ZodObject<{
        timeToFirstToken: z.ZodOptional<z.ZodNumber>;
        tokensPerSecond: z.ZodOptional<z.ZodNumber>;
        promptTokensPerSecond: z.ZodOptional<z.ZodNumber>;
        cacheTokens: z.ZodOptional<z.ZodNumber>;
        promptTokens: z.ZodOptional<z.ZodNumber>;
        generatedTokens: z.ZodOptional<z.ZodNumber>;
        emittedTokens: z.ZodOptional<z.ZodNumber>;
        avgConcurrentSeq: z.ZodOptional<z.ZodNumber>;
        toolDefinitionsDropped: z.ZodOptional<z.ZodNumber>;
        backendDevice: z.ZodOptional<z.ZodEnum<{
            cpu: "cpu";
            gpu: "gpu";
        }>>;
    }, z.core.$strip>;
}, z.core.$strip>;
export declare const completionErrorSchema: z.ZodObject<{
    message: z.ZodString;
}, z.core.$strip>;
declare const stopReasonEnum: z.ZodEnum<{
    cancelled: "cancelled";
    length: "length";
    eos: "eos";
    stopSequence: "stopSequence";
}>;
export declare const doneEventSchema: z.ZodUnion<readonly [z.ZodObject<{
    type: z.ZodLiteral<"completionDone">;
    seq: z.ZodNumber;
    stopReason: z.ZodLiteral<"error">;
    error: z.ZodObject<{
        message: z.ZodString;
    }, z.core.$strip>;
    raw: z.ZodOptional<z.ZodObject<{
        fullText: z.ZodString;
    }, z.core.$strip>>;
}, z.core.$strict>, z.ZodObject<{
    type: z.ZodLiteral<"completionDone">;
    seq: z.ZodNumber;
    stopReason: z.ZodOptional<z.ZodEnum<{
        cancelled: "cancelled";
        length: "length";
        eos: "eos";
        stopSequence: "stopSequence";
    }>>;
    raw: z.ZodOptional<z.ZodObject<{
        fullText: z.ZodString;
    }, z.core.$strip>>;
}, z.core.$strict>]>;
export declare const completionEventSchema: z.ZodUnion<readonly [z.ZodObject<{
    type: z.ZodLiteral<"contentDelta">;
    seq: z.ZodNumber;
    text: z.ZodString;
}, z.core.$strip>, z.ZodObject<{
    type: z.ZodLiteral<"rawDelta">;
    seq: z.ZodNumber;
    text: z.ZodString;
}, z.core.$strip>, z.ZodObject<{
    type: z.ZodLiteral<"thinkingDelta">;
    seq: z.ZodNumber;
    text: z.ZodString;
}, z.core.$strip>, z.ZodObject<{
    type: z.ZodLiteral<"toolCall">;
    seq: z.ZodNumber;
    call: z.ZodObject<{
        id: z.ZodString;
        name: z.ZodString;
        arguments: z.ZodRecord<z.ZodString, z.ZodUnknown>;
        raw: z.ZodOptional<z.ZodString>;
    }, z.core.$strip>;
}, z.core.$strip>, z.ZodObject<{
    type: z.ZodLiteral<"toolError">;
    seq: z.ZodNumber;
    error: z.ZodObject<{
        code: z.ZodEnum<{
            PARSE_ERROR: "PARSE_ERROR";
            VALIDATION_ERROR: "VALIDATION_ERROR";
            UNKNOWN_TOOL: "UNKNOWN_TOOL";
        }>;
        message: z.ZodString;
        raw: z.ZodOptional<z.ZodString>;
    }, z.core.$strip>;
}, z.core.$strip>, z.ZodObject<{
    type: z.ZodLiteral<"completionStats">;
    seq: z.ZodNumber;
    stats: z.ZodObject<{
        timeToFirstToken: z.ZodOptional<z.ZodNumber>;
        tokensPerSecond: z.ZodOptional<z.ZodNumber>;
        promptTokensPerSecond: z.ZodOptional<z.ZodNumber>;
        cacheTokens: z.ZodOptional<z.ZodNumber>;
        promptTokens: z.ZodOptional<z.ZodNumber>;
        generatedTokens: z.ZodOptional<z.ZodNumber>;
        emittedTokens: z.ZodOptional<z.ZodNumber>;
        avgConcurrentSeq: z.ZodOptional<z.ZodNumber>;
        toolDefinitionsDropped: z.ZodOptional<z.ZodNumber>;
        backendDevice: z.ZodOptional<z.ZodEnum<{
            cpu: "cpu";
            gpu: "gpu";
        }>>;
    }, z.core.$strip>;
}, z.core.$strip>, z.ZodObject<{
    type: z.ZodLiteral<"completionDone">;
    seq: z.ZodNumber;
    stopReason: z.ZodLiteral<"error">;
    error: z.ZodObject<{
        message: z.ZodString;
    }, z.core.$strip>;
    raw: z.ZodOptional<z.ZodObject<{
        fullText: z.ZodString;
    }, z.core.$strip>>;
}, z.core.$strict>, z.ZodObject<{
    type: z.ZodLiteral<"completionDone">;
    seq: z.ZodNumber;
    stopReason: z.ZodOptional<z.ZodEnum<{
        cancelled: "cancelled";
        length: "length";
        eos: "eos";
        stopSequence: "stopSequence";
    }>>;
    raw: z.ZodOptional<z.ZodObject<{
        fullText: z.ZodString;
    }, z.core.$strip>>;
}, z.core.$strict>]>;
export type ContentDeltaEvent = z.infer<typeof contentDeltaEventSchema>;
export type RawDeltaEvent = z.infer<typeof rawDeltaEventSchema>;
export type ThinkingDeltaEvent = z.infer<typeof thinkingDeltaEventSchema>;
export type ToolCallEvent = z.infer<typeof toolCallEventSchema>;
export type ToolErrorEvent = z.infer<typeof toolErrorEventSchema>;
export type StatsEvent = z.infer<typeof statsEventSchema>;
export type CompletionError = z.infer<typeof completionErrorSchema>;
export type DoneEvent = z.infer<typeof doneEventSchema>;
export type CompletionEvent = z.infer<typeof completionEventSchema>;
export type StopReason = z.infer<typeof stopReasonEnum>;
export type CompletionFinal = {
    contentText: string;
    thinkingText?: string;
    toolCalls: ToolCallWithCall[];
    /**
     * Tool-call regions the model emitted that could not be turned into a
     * `toolCalls` entry — malformed markup, arguments failing the tool's
     * schema, or a call to an undeclared tool. Omitted when there were none.
     * Lets a caller tell "the model made a mistake" from "no tool was called".
     */
    toolErrors?: ToolCallError[];
    stats?: CompletionStats;
    stopReason?: StopReason;
    raw: {
        fullText: string;
    };
    /**
     * Canonical assistant text to push back into `history` for the next turn
     * when using `kvCache: true` (auto-cache). Equals the assistant content
     * we persisted to the cache key on this turn, so re-using it
     * verbatim guarantees a cache hit on the next call.
     *
     * Derived from `raw.fullText` (or `contentText` if the addon didn't
     * emit raw text) by stripping `<think>` reasoning blocks and trimming
     * surrounding whitespace — see `normalizeAssistantCacheContent`.
     *
     * Tool-call turns currently can't be auto-cached, so this
     * field is omitted when `toolCalls.length > 0`.
     */
    cacheableAssistantContent?: string;
};
export type CompletionRun = {
    /**
     * Stable identifier for this run, generated by the caller at call time
     * (UUIDv4 when `crypto.randomUUID` is available, otherwise an opaque
     * random hex token) and available synchronously the moment
     * `completion(...)` returns. Pass it to `cancel({ requestId })` to
     * target this specific request without affecting any other inference
     * running on the same model. The cancel only takes effect once the
     * request has begun, so a cancel issued in the same tick
     * as `completion(...)` may race the begin and is logged as a no-match.
     */
    requestId: string;
    /** Ordered stream of typed completion events — the canonical consumption API. */
    events: AsyncIterable<CompletionEvent>;
    /** Resolves when the stream ends with aggregated content, thinking, tool calls, stats, and raw output. */
    final: Promise<CompletionFinal>;
    tokenStream: AsyncGenerator<string>;
    toolCallStream: AsyncGenerator<ToolCallEvent>;
    text: Promise<string>;
    toolCalls: Promise<ToolCallWithCall[]>;
    stats: Promise<CompletionStats | undefined>;
};
export type ToolCallingCapability = 'textParse' | 'none';
export type ThinkingFramingCapability = 'thinkTags' | 'none';
export type PluginCapabilities = {
    toolCalling: ToolCallingCapability;
    thinkingFraming: ThinkingFramingCapability;
};
export declare const DEFAULT_PLUGIN_CAPABILITIES: PluginCapabilities;
export type NormalizerConfig = {
    capabilities: PluginCapabilities;
    tools: Tool[];
    captureThinking: boolean;
    emitRawDeltas: boolean;
    toolDialect?: ToolDialect;
};
export {};