@mlx-node/server 0.0.7 → 0.0.8

package/dist/types-anthropic.d.ts

@@ -37,7 +37,7 @@ export interface SystemBlock {
     };
 }
 export interface AnthropicMessage {
-    role: 'user' | 'assistant';
+    role: 'user' | 'assistant' | 'system';
     content: string | AnthropicContentBlock[];
 }
 export interface AnthropicToolDefinition {
@@ -65,6 +65,40 @@ export interface AnthropicMessagesRequest {
     metadata?: {
         user_id?: string;
     };
+    output_config?: {
+        format?: {
+            type?: string;
+            schema?: unknown;
+        };
+    };
+    /**
+     * MLX-Node extension carrier for non-Anthropic fields, namespaced
+     * under `extra_body` to mirror the OpenAI-side `/v1/responses`
+     * surface. Unknown keys are ignored (additive, forward-compat).
+     *
+     * Currently exposes:
+     *   * `generation_mode`: `"mtp"` forces W6 speculative-decode (sets
+     *     `enableMtp = true`), `"ar"` forces plain autoregressive
+     *     (`enableMtp = false`). Absent / null / unrecognized leaves
+     *     `enableMtp` untouched so the downstream `ChatSession` auto-
+     *     default (true when the model ships an MTP head) applies.
+     *   * `mtp_depth`: positive integer override for the per-call draft
+     *     depth. The server forwards any positive integer ≤ 64 (a sanity
+     *     ceiling that only blocks garbage) and the native per-family
+     *     `resolve_params` owns the real clamps: qwen3.5 native MTP
+     *     [1, 5], gemma4 DSpark capped at the draft block size, gemma4
+     *     assistant drafts [1, 8].
+     */
+    extra_body?: {
+        generation_mode?: string | null;
+        mtp_depth?: number | null;
+    };
+}
+export type AnthropicCountTokensRequest = Omit<AnthropicMessagesRequest, 'max_tokens'> & {
+    max_tokens?: number;
+};
+export interface AnthropicCountTokensResponse {
+    input_tokens: number;
 }
 export interface AnthropicResponseTextBlock {
     type: 'text';
@@ -81,9 +115,85 @@ export interface AnthropicResponseToolUseBlock {
     input: Record<string, unknown>;
 }
 export type AnthropicResponseContent = AnthropicResponseTextBlock | AnthropicResponseThinkingBlock | AnthropicResponseToolUseBlock;
+/**
+ * Anthropic Messages API `usage` block.
+ *
+ * Per the Anthropic spec the cache accounting fields are OPTIONAL and
+ * carry distinct semantics that drive client-side cost / billing
+ * displays (Claude Code reads `cache_read_input_tokens` directly):
+ *
+ *   * `input_tokens` — prompt tokens that were processed at full cost
+ *     this turn. When a cached prefix is reused, this MUST be reduced
+ *     to the unsuffixed remainder (`promptTokens - cachedTokens`) so
+ *     the client doesn't double-count the cached prefix.
+ *   * `cache_read_input_tokens` — prompt tokens served from a cached
+ *     prefix on this turn. Emitted only when reuse genuinely happened
+ *     (`cachedTokens > 0`); omitted on a cold turn so the wire matches
+ *     other Anthropic-compatible servers that elide the field on cache
+ *     misses.
+ *   * `cache_creation_input_tokens` — prompt tokens written to a NEW
+ *     cache entry. Currently always omitted: this server's KV reuse
+ *     is implicit and has no `cache_control` breakpoints, so a client
+ *     that did not request explicit caching should never see a
+ *     non-zero creation count.
+ *
+ * The timing fields below are NON-Anthropic extension fields surfaced
+ * for the launcher's verbose log (`requests.ndjson`) so per-turn
+ * decode-rate / TTFT / prefill-rate telemetry rides the same response
+ * envelope. They describe native/server inference timing, not the
+ * HTTP logger's outer `elapsedMs` envelope. They are emitted only when
+ * the underlying native dispatch produced a finite, positive value —
+ * missing or non-finite metrics are elided rather than surfaced as
+ * zero / null. Anthropic-compatible clients (Claude Code, official
+ * Anthropic SDKs) ignore unknown fields, so the extension is wire-safe
+ * — it parallels how `cache_read_input_tokens` is treated above.
+ *
+ * `prefill_input_tokens` and `cached_prefix_tokens` provide the
+ * denominator/context for `prefill_tokens_per_second`: on cached-prefix
+ * turns the prefill rate is for the uncached suffix, not the full
+ * logical prompt.
+ */
 export interface AnthropicUsage {
     input_tokens: number;
     output_tokens: number;
+    cache_read_input_tokens?: number;
+    cache_creation_input_tokens?: number;
+    /** Server-extension: time-to-first-token in milliseconds. */
+    time_to_first_token_ms?: number;
+    /** Server-extension: prompt-token throughput during prefill. */
+    prefill_tokens_per_second?: number;
+    /** Server-extension: generated-token throughput during decode. */
+    decode_tokens_per_second?: number;
+    /** Server-extension: native/server inference elapsed, excluding HTTP transport and logging overhead. */
+    server_inference_elapsed_ms?: number;
+    /** Server-extension alias for disambiguating native TTFT from request/HTTP elapsed time. */
+    server_time_to_first_token_ms?: number;
+    /** Server-extension: handler-start to first native token, including model resolve/load and queue wait. */
+    server_total_time_to_first_token_ms?: number;
+    /** Server-extension alias for native prefill throughput. */
+    server_prefill_tokens_per_second?: number;
+    /** Server-extension alias for native decode throughput. */
+    server_decode_tokens_per_second?: number;
+    /** Server-extension: prompt tokens actually prefetched this turn after cached-prefix reuse. */
+    prefill_input_tokens?: number;
+    /** Server-extension: prompt tokens skipped because a cached prefix was reused. */
+    cached_prefix_tokens?: number;
+    /** Server-extension: time spent resolving/loading/aliasing the requested model before registry lookup. */
+    server_model_resolve_ms?: number;
+    /** Server-extension: wall-clock time blocked on the process-wide model-load writer lock. */
+    server_load_wait_ms?: number;
+    /** Server-extension: true when this request acquired the model-load lock without contention. */
+    server_load_owner?: boolean;
+    /** Server-extension: time spent waiting behind the per-model execution mutex. */
+    server_queue_ms?: number;
+    /** Server-extension: handler time before native inference begins, including resolve and queue wait. */
+    server_pre_inference_ms?: number;
+    /** Server-extension: effective process-level paged-prefill chunk size. */
+    server_paged_prefill_chunk_size?: number;
+    /** Server-extension: effective process-level paged-prefill eval/clear cadence. */
+    server_paged_prefill_eval_interval?: number;
+    /** Server-extension: effective process-level paged-decode cache-clear cadence. */
+    server_paged_decode_cache_clear_interval?: number;
 }
 export interface AnthropicMessagesResponse {
     id: string;
@@ -132,9 +242,59 @@ export interface AnthropicMessageDeltaEvent {
         stop_reason: string | null;
         stop_sequence: string | null;
     };
+    /**
+     * Streaming `message_delta` carries the SAME cache-accounting
+     * semantics as the non-streaming response `usage` block (see
+     * `AnthropicUsage` above) — `input_tokens` MUST be net of any reused
+     * prefix, and `cache_read_input_tokens` is emitted only on a true
+     * reuse turn.
+     *
+     * Timing/accounting fields are server extensions (not in
+     * Anthropic's spec) surfaced for the launcher's verbose log; see
+     * the docstring on `AnthropicUsage`. Clients that do not recognize
+     * them ignore them.
+     */
     usage: {
         input_tokens?: number;
         output_tokens: number;
+        cache_read_input_tokens?: number;
+        cache_creation_input_tokens?: number;
+        /** Server-extension: time-to-first-token in milliseconds. */
+        time_to_first_token_ms?: number;
+        /** Server-extension: prompt-token throughput during prefill. */
+        prefill_tokens_per_second?: number;
+        /** Server-extension: generated-token throughput during decode. */
+        decode_tokens_per_second?: number;
+        /** Server-extension: native/server inference elapsed, excluding HTTP transport and logging overhead. */
+        server_inference_elapsed_ms?: number;
+        /** Server-extension alias for disambiguating native TTFT from request/HTTP elapsed time. */
+        server_time_to_first_token_ms?: number;
+        /** Server-extension: handler-start to first native token, including model resolve/load and queue wait. */
+        server_total_time_to_first_token_ms?: number;
+        /** Server-extension alias for native prefill throughput. */
+        server_prefill_tokens_per_second?: number;
+        /** Server-extension alias for native decode throughput. */
+        server_decode_tokens_per_second?: number;
+        /** Server-extension: prompt tokens actually prefetched this turn after cached-prefix reuse. */
+        prefill_input_tokens?: number;
+        /** Server-extension: prompt tokens skipped because a cached prefix was reused. */
+        cached_prefix_tokens?: number;
+        /** Server-extension: time spent resolving/loading/aliasing the requested model before registry lookup. */
+        server_model_resolve_ms?: number;
+        /** Server-extension: wall-clock time blocked on the process-wide model-load writer lock. */
+        server_load_wait_ms?: number;
+        /** Server-extension: true when this request acquired the model-load lock without contention. */
+        server_load_owner?: boolean;
+        /** Server-extension: time spent waiting behind the per-model execution mutex. */
+        server_queue_ms?: number;
+        /** Server-extension: handler time before native inference begins, including resolve and queue wait. */
+        server_pre_inference_ms?: number;
+        /** Server-extension: effective process-level paged-prefill chunk size. */
+        server_paged_prefill_chunk_size?: number;
+        /** Server-extension: effective process-level paged-prefill eval/clear cadence. */
+        server_paged_prefill_eval_interval?: number;
+        /** Server-extension: effective process-level paged-decode cache-clear cadence. */
+        server_paged_decode_cache_clear_interval?: number;
     };
 }
 export interface AnthropicMessageStopEvent {

package/dist/types-anthropic.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"types-anthropic.d.ts","sourceRoot":"","sources":["../src/types-anthropic.ts"],"names":[],"mappings":"AAAA,4GAA4G;AAE5G,MAAM,WAAW,yBAAyB;IACxC,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;CACd;AAED,MAAM,WAAW,0BAA0B;IACzC,IAAI,EAAE,OAAO,CAAC;IACd,MAAM,EAAE;QACN,IAAI,EAAE,QAAQ,CAAC;QACf,UAAU,EAAE,MAAM,CAAC;QACnB,IAAI,EAAE,MAAM,CAAC;KACd,CAAC;CACH;AAED,MAAM,WAAW,+BAA+B;IAC9C,IAAI,EAAE,aAAa,CAAC;IACpB,WAAW,EAAE,MAAM,CAAC;IACpB,yJAAyJ;IACzJ,OAAO,CAAC,EAAE,MAAM,GAAG,CAAC,yBAAyB,GAAG,0BAA0B,CAAC,EAAE,CAAC;IAC9E,QAAQ,CAAC,EAAE,OAAO,CAAC;CACpB;AAED,MAAM,WAAW,4BAA4B;IAC3C,IAAI,EAAE,UAAU,CAAC;IACjB,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CAChC;AAED,MAAM,WAAW,6BAA6B;IAC5C,IAAI,EAAE,UAAU,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,MAAM,qBAAqB,GAC7B,yBAAyB,GACzB,0BAA0B,GAC1B,+BAA+B,GAC/B,4BAA4B,GAC5B,6BAA6B,CAAC;AAMlC,MAAM,WAAW,WAAW;IAC1B,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,aAAa,CAAC,EAAE;QAAE,IAAI,EAAE,MAAM,CAAA;KAAE,CAAC;CAClC;AAMD,MAAM,WAAW,gBAAgB;IAC/B,IAAI,EAAE,MAAM,GAAG,WAAW,CAAC;IAC3B,OAAO,EAAE,MAAM,GAAG,qBAAqB,EAAE,CAAC;CAC3C;AAMD,MAAM,WAAW,uBAAuB;IACtC,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,YAAY,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACvC;AAED,MAAM,WAAW,mBAAmB;IAClC,IAAI,EAAE,MAAM,GAAG,KAAK,GAAG,MAAM,CAAC;IAC9B,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,yBAAyB,CAAC,EAAE,OAAO,CAAC;CACrC;AAMD,MAAM,WAAW,wBAAwB;IACvC,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,gBAAgB,EAAE,CAAC;IAC7B,UAAU,EAAE,MAAM,CAAC;IACnB,MAAM,CAAC,EAAE,MAAM,GAAG,WAAW,EAAE,CAAC;IAChC,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,KAAK,CAAC,EAAE,uBAAuB,EAAE,CAAC;IAClC,WAAW,CAAC,EAAE,mBAAmB,CAAC;IAClC,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,cAAc,CAAC,EAAE,MAAM,EAAE,CAAC;IAC1B,QAAQ,CAAC,EAAE;QAAE,OAAO,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;CACjC;AAMD,MAAM,WAAW,0BAA0B;IACzC,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;CACd;AAED,MAAM,WAAW,8BAA8B;IAC7C,IAAI,EAAE,UAAU,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,WAAW,6BAA6B;IAC5C,IAAI,EAAE,UAAU,CAAC;IACjB,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CAChC;AAED,MAAM,MAAM,wBAAwB,GAChC,0BAA0B,GAC1B,8BAA8B,GAC9B,6BAA6B,CAAC;AAMlC,MAAM,WAAW,cAAc;IAC7B,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;CACvB;AAMD,MAAM,WAAW,yBAAyB;IACxC,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,SAAS,CAAC;IAChB,IAAI,EAAE,WAAW,CAAC;IAClB,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,wBAAwB,EAAE,CAAC;IACpC,WAAW,EAAE,UAAU,GAAG,YAAY,GAAG,eAAe,GAAG,UAAU,GAAG,IAAI,CAAC;IAC7E,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,KAAK,EAAE,cAAc,CAAC;CACvB;AAMD,MAAM,WAAW,kBAAkB;IACjC,IAAI,EAAE,YAAY,CAAC;IACnB,IAAI,EAAE,MAAM,CAAC;CACd;AAED,MAAM,WAAW,sBAAsB;IACrC,IAAI,EAAE,gBAAgB,CAAC;IACvB,QAAQ,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,WAAW,uBAAuB;IACtC,IAAI,EAAE,kBAAkB,CAAC;IACzB,YAAY,EAAE,MAAM,CAAC;CACtB;AAED,MAAM,MAAM,cAAc,GAAG,kBAAkB,GAAG,sBAAsB,GAAG,uBAAuB,CAAC;AAMnG,MAAM,WAAW,0BAA0B;IACzC,IAAI,EAAE,eAAe,CAAC;IACtB,OAAO,EAAE,yBAAyB,CAAC;CACpC;AAED,MAAM,WAAW,+BAA+B;IAC9C,IAAI,EAAE,qBAAqB,CAAC;IAC5B,KAAK,EAAE,MAAM,CAAC;IACd,aAAa,EAAE,wBAAwB,CAAC;CACzC;AAED,MAAM,WAAW,+BAA+B;IAC9C,IAAI,EAAE,qBAAqB,CAAC;IAC5B,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,cAAc,CAAC;CACvB;AAED,MAAM,WAAW,8BAA8B;IAC7C,IAAI,EAAE,oBAAoB,CAAC;IAC3B,KAAK,EAAE,MAAM,CAAC;CACf;AAED,MAAM,WAAW,0BAA0B;IACzC,IAAI,EAAE,eAAe,CAAC;IACtB,KAAK,EAAE;QACL,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;QAC3B,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;KAC9B,CAAC;IACF,KAAK,EAAE;QAAE,YAAY,CAAC,EAAE,MAAM,CAAC;QAAC,aAAa,EAAE,MAAM,CAAA;KAAE,CAAC;CACzD;AAED,MAAM,WAAW,yBAAyB;IACxC,IAAI,EAAE,cAAc,CAAC;CACtB;AAED,MAAM,MAAM,oBAAoB,GAC5B,0BAA0B,GAC1B,+BAA+B,GAC/B,+BAA+B,GAC/B,8BAA8B,GAC9B,0BAA0B,GAC1B,yBAAyB,CAAC"}
+{"version":3,"file":"types-anthropic.d.ts","sourceRoot":"","sources":["../src/types-anthropic.ts"],"names":[],"mappings":"AAAA,4GAA4G;AAE5G,MAAM,WAAW,yBAAyB;IACxC,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;CACd;AAED,MAAM,WAAW,0BAA0B;IACzC,IAAI,EAAE,OAAO,CAAC;IACd,MAAM,EAAE;QACN,IAAI,EAAE,QAAQ,CAAC;QACf,UAAU,EAAE,MAAM,CAAC;QACnB,IAAI,EAAE,MAAM,CAAC;KACd,CAAC;CACH;AAED,MAAM,WAAW,+BAA+B;IAC9C,IAAI,EAAE,aAAa,CAAC;IACpB,WAAW,EAAE,MAAM,CAAC;IACpB,yJAAyJ;IACzJ,OAAO,CAAC,EAAE,MAAM,GAAG,CAAC,yBAAyB,GAAG,0BAA0B,CAAC,EAAE,CAAC;IAC9E,QAAQ,CAAC,EAAE,OAAO,CAAC;CACpB;AAED,MAAM,WAAW,4BAA4B;IAC3C,IAAI,EAAE,UAAU,CAAC;IACjB,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CAChC;AAED,MAAM,WAAW,6BAA6B;IAC5C,IAAI,EAAE,UAAU,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,MAAM,qBAAqB,GAC7B,yBAAyB,GACzB,0BAA0B,GAC1B,+BAA+B,GAC/B,4BAA4B,GAC5B,6BAA6B,CAAC;AAMlC,MAAM,WAAW,WAAW;IAC1B,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,aAAa,CAAC,EAAE;QAAE,IAAI,EAAE,MAAM,CAAA;KAAE,CAAC;CAClC;AAMD,MAAM,WAAW,gBAAgB;IAM/B,IAAI,EAAE,MAAM,GAAG,WAAW,GAAG,QAAQ,CAAC;IACtC,OAAO,EAAE,MAAM,GAAG,qBAAqB,EAAE,CAAC;CAC3C;AAMD,MAAM,WAAW,uBAAuB;IACtC,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,YAAY,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACvC;AAED,MAAM,WAAW,mBAAmB;IAClC,IAAI,EAAE,MAAM,GAAG,KAAK,GAAG,MAAM,CAAC;IAC9B,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,yBAAyB,CAAC,EAAE,OAAO,CAAC;CACrC;AAMD,MAAM,WAAW,wBAAwB;IACvC,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,gBAAgB,EAAE,CAAC;IAC7B,UAAU,EAAE,MAAM,CAAC;IACnB,MAAM,CAAC,EAAE,MAAM,GAAG,WAAW,EAAE,CAAC;IAChC,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,KAAK,CAAC,EAAE,uBAAuB,EAAE,CAAC;IAClC,WAAW,CAAC,EAAE,mBAAmB,CAAC;IAClC,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,cAAc,CAAC,EAAE,MAAM,EAAE,CAAC;IAC1B,QAAQ,CAAC,EAAE;QAAE,OAAO,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;IAChC,aAAa,CAAC,EAAE;QACd,MAAM,CAAC,EAAE;YACP,IAAI,CAAC,EAAE,MAAM,CAAC;YACd,MAAM,CAAC,EAAE,OAAO,CAAC;SAClB,CAAC;KACH,CAAC;IAcF;;;;;;;;;;;;;;;;;OAiBG;IACH,UAAU,CAAC,EAAE;QAMX,eAAe,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;QAChC,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;KAC3B,CAAC;CACH;AAED,MAAM,MAAM,2BAA2B,GAAG,IAAI,CAAC,wBAAwB,EAAE,YAAY,CAAC,GAAG;IACvF,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB,CAAC;AAEF,MAAM,WAAW,4BAA4B;IAC3C,YAAY,EAAE,MAAM,CAAC;CACtB;AAMD,MAAM,WAAW,0BAA0B;IACzC,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;CACd;AAED,MAAM,WAAW,8BAA8B;IAC7C,IAAI,EAAE,UAAU,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,WAAW,6BAA6B;IAC5C,IAAI,EAAE,UAAU,CAAC;IACjB,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CAChC;AAED,MAAM,MAAM,wBAAwB,GAChC,0BAA0B,GAC1B,8BAA8B,GAC9B,6BAA6B,CAAC;AAMlC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAqCG;AACH,MAAM,WAAW,cAAc;IAC7B,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,uBAAuB,CAAC,EAAE,MAAM,CAAC;IACjC,2BAA2B,CAAC,EAAE,MAAM,CAAC;IACrC,6DAA6D;IAC7D,sBAAsB,CAAC,EAAE,MAAM,CAAC;IAChC,gEAAgE;IAChE,yBAAyB,CAAC,EAAE,MAAM,CAAC;IACnC,kEAAkE;IAClE,wBAAwB,CAAC,EAAE,MAAM,CAAC;IAClC,wGAAwG;IACxG,2BAA2B,CAAC,EAAE,MAAM,CAAC;IACrC,4FAA4F;IAC5F,6BAA6B,CAAC,EAAE,MAAM,CAAC;IACvC,0GAA0G;IAC1G,mCAAmC,CAAC,EAAE,MAAM,CAAC;IAC7C,4DAA4D;IAC5D,gCAAgC,CAAC,EAAE,MAAM,CAAC;IAC1C,2DAA2D;IAC3D,+BAA+B,CAAC,EAAE,MAAM,CAAC;IACzC,+FAA+F;IAC/F,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAC9B,kFAAkF;IAClF,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAC9B,0GAA0G;IAC1G,uBAAuB,CAAC,EAAE,MAAM,CAAC;IACjC,4FAA4F;IAC5F,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,gGAAgG;IAChG,iBAAiB,CAAC,EAAE,OAAO,CAAC;IAC5B,iFAAiF;IACjF,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,uGAAuG;IACvG,uBAAuB,CAAC,EAAE,MAAM,CAAC;IACjC,0EAA0E;IAC1E,+BAA+B,CAAC,EAAE,MAAM,CAAC;IACzC,kFAAkF;IAClF,kCAAkC,CAAC,EAAE,MAAM,CAAC;IAC5C,kFAAkF;IAClF,wCAAwC,CAAC,EAAE,MAAM,CAAC;CACnD;AAMD,MAAM,WAAW,yBAAyB;IACxC,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,SAAS,CAAC;IAChB,IAAI,EAAE,WAAW,CAAC;IAClB,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,wBAAwB,EAAE,CAAC;IACpC,WAAW,EAAE,UAAU,GAAG,YAAY,GAAG,eAAe,GAAG,UAAU,GAAG,IAAI,CAAC;IAC7E,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,KAAK,EAAE,cAAc,CAAC;CACvB;AAMD,MAAM,WAAW,kBAAkB;IACjC,IAAI,EAAE,YAAY,CAAC;IACnB,IAAI,EAAE,MAAM,CAAC;CACd;AAED,MAAM,WAAW,sBAAsB;IACrC,IAAI,EAAE,gBAAgB,CAAC;IACvB,QAAQ,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,WAAW,uBAAuB;IACtC,IAAI,EAAE,kBAAkB,CAAC;IACzB,YAAY,EAAE,MAAM,CAAC;CACtB;AAED,MAAM,MAAM,cAAc,GAAG,kBAAkB,GAAG,sBAAsB,GAAG,uBAAuB,CAAC;AAMnG,MAAM,WAAW,0BAA0B;IACzC,IAAI,EAAE,eAAe,CAAC;IACtB,OAAO,EAAE,yBAAyB,CAAC;CACpC;AAED,MAAM,WAAW,+BAA+B;IAC9C,IAAI,EAAE,qBAAqB,CAAC;IAC5B,KAAK,EAAE,MAAM,CAAC;IACd,aAAa,EAAE,wBAAwB,CAAC;CACzC;AAED,MAAM,WAAW,+BAA+B;IAC9C,IAAI,EAAE,qBAAqB,CAAC;IAC5B,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,cAAc,CAAC;CACvB;AAED,MAAM,WAAW,8BAA8B;IAC7C,IAAI,EAAE,oBAAoB,CAAC;IAC3B,KAAK,EAAE,MAAM,CAAC;CACf;AAED,MAAM,WAAW,0BAA0B;IACzC,IAAI,EAAE,eAAe,CAAC;IACtB,KAAK,EAAE;QACL,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;QAC3B,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;KAC9B,CAAC;IACF;;;;;;;;;;;OAWG;IACH,KAAK,EAAE;QACL,YAAY,CAAC,EAAE,MAAM,CAAC;QACtB,aAAa,EAAE,MAAM,CAAC;QACtB,uBAAuB,CAAC,EAAE,MAAM,CAAC;QACjC,2BAA2B,CAAC,EAAE,MAAM,CAAC;QACrC,6DAA6D;QAC7D,sBAAsB,CAAC,EAAE,MAAM,CAAC;QAChC,gEAAgE;QAChE,yBAAyB,CAAC,EAAE,MAAM,CAAC;QACnC,kEAAkE;QAClE,wBAAwB,CAAC,EAAE,MAAM,CAAC;QAClC,wGAAwG;QACxG,2BAA2B,CAAC,EAAE,MAAM,CAAC;QACrC,4FAA4F;QAC5F,6BAA6B,CAAC,EAAE,MAAM,CAAC;QACvC,0GAA0G;QAC1G,mCAAmC,CAAC,EAAE,MAAM,CAAC;QAC7C,4DAA4D;QAC5D,gCAAgC,CAAC,EAAE,MAAM,CAAC;QAC1C,2DAA2D;QAC3D,+BAA+B,CAAC,EAAE,MAAM,CAAC;QACzC,+FAA+F;QAC/F,oBAAoB,CAAC,EAAE,MAAM,CAAC;QAC9B,kFAAkF;QAClF,oBAAoB,CAAC,EAAE,MAAM,CAAC;QAC9B,0GAA0G;QAC1G,uBAAuB,CAAC,EAAE,MAAM,CAAC;QACjC,4FAA4F;QAC5F,mBAAmB,CAAC,EAAE,MAAM,CAAC;QAC7B,gGAAgG;QAChG,iBAAiB,CAAC,EAAE,OAAO,CAAC;QAC5B,iFAAiF;QACjF,eAAe,CAAC,EAAE,MAAM,CAAC;QACzB,uGAAuG;QACvG,uBAAuB,CAAC,EAAE,MAAM,CAAC;QACjC,0EAA0E;QAC1E,+BAA+B,CAAC,EAAE,MAAM,CAAC;QACzC,kFAAkF;QAClF,kCAAkC,CAAC,EAAE,MAAM,CAAC;QAC5C,kFAAkF;QAClF,wCAAwC,CAAC,EAAE,MAAM,CAAC;KACnD,CAAC;CACH;AAED,MAAM,WAAW,yBAAyB;IACxC,IAAI,EAAE,cAAc,CAAC;CACtB;AAED,MAAM,MAAM,oBAAoB,GAC5B,0BAA0B,GAC1B,+BAA+B,GAC/B,+BAA+B,GAC/B,8BAA8B,GAC9B,0BAA0B,GAC1B,yBAAyB,CAAC"}

package/dist/types.d.ts

@@ -3,7 +3,41 @@ export interface InputTextPart {
     type: 'input_text';
     text: string;
 }
-export type ContentPart = InputTextPart;
+/**
+ * An image attached to a user message. `image_url` may be an `http(s)://` URL
+ * (not fetched by the mapper) or a `data:<mime>;base64,<payload>` URL
+ * (decoded and forwarded to the model as raw bytes).
+ */
+export interface InputImagePart {
+    type: 'input_image';
+    image_url?: string;
+    file_id?: string;
+    detail?: 'auto' | 'low' | 'high';
+}
+/**
+ * Echoed back by clients (Codex, pi-ai, etc.) when they replay prior
+ * assistant turns in `input[]` instead of using `previous_response_id`.
+ * Equivalent to `InputTextPart` for mapping purposes — we only need the text.
+ */
+export interface InputAssistantTextPart {
+    type: 'output_text';
+    text: string;
+    annotations?: never[];
+}
+/** Echoed back by clients when replaying a refusal block from a prior turn. */
+export interface InputRefusalPart {
+    type: 'refusal';
+    refusal: string;
+}
+/**
+ * Some clients inline a reasoning summary as a content part on an assistant
+ * message instead of as a top-level `reasoning` input item.
+ */
+export interface InputSummaryTextPart {
+    type: 'summary_text';
+    text: string;
+}
+export type ContentPart = InputTextPart | InputImagePart | InputAssistantTextPart | InputRefusalPart | InputSummaryTextPart;
 export interface InputMessage {
     type?: 'message';
     role: 'user' | 'assistant' | 'system' | 'developer';
@@ -21,7 +55,21 @@ export interface InputFunctionCallOutput {
     call_id: string;
     output: string;
 }
-export type InputItem = InputMessage | InputFunctionCall | InputFunctionCallOutput;
+/**
+ * Top-level reasoning item replayed by clients that echo prior assistant
+ * reasoning summaries instead of using `previous_response_id`. The summary
+ * is coalesced onto the next assistant `ChatMessage` as `reasoningContent`.
+ */
+export interface InputReasoningItem {
+    type: 'reasoning';
+    id?: string;
+    summary?: {
+        type?: 'summary_text';
+        text: string;
+    }[];
+    encrypted_content?: string;
+}
+export type InputItem = InputMessage | InputFunctionCall | InputFunctionCallOutput | InputReasoningItem;
 export interface ResponsesToolDefinition {
     type: 'function';
     name: string;
@@ -48,6 +96,58 @@ export interface ResponsesAPIRequest {
     };
     previous_response_id?: string;
     store?: boolean;
+    /**
+     * Stable caller-supplied key identifying the logical conversation for
+     * warm-session reuse across stateless turns. Stateless agent clients
+     * (pi-mono, Aider, Codex CLI, etc.) own the conversation history
+     * client-side and resend the full transcript on every turn — they do
+     * NOT use `previous_response_id` — so the server-side
+     * `SessionRegistry` can only reuse a warm `ChatSession` across those
+     * turns if the client threads a stable key through every request.
+     *
+     * When present (and `previous_response_id` is absent), the registry's
+     * tier-2 lookup scans for a still-live entry whose `promptCacheKey`
+     * matches and whose `instructions` are byte-equal to this request's
+     * — on a match it leases the warm session out, so the native prefix-
+     * cache verifier (`verify_cache_prefix_direct`) can skip the re-
+     * prefill of the conversation history and only prefill the newly
+     * appended user turn. When absent, each stateless turn cold-starts.
+     *
+     * `previous_response_id` unconditionally wins when both are set —
+     * the two keys may identify different conversation branches, so
+     * picking the prompt_cache_key branch on a prev-id miss would risk
+     * routing the request through the wrong warm state. Fall through to
+     * fresh on prev-id miss instead.
+     *
+     * **Enabled by default.** Opt out with
+     * `MLX_DISABLE_PROMPT_CACHE_KEY=1` in multi-tenant deployments,
+     * where the tier-2 lookup becomes unsafe — the key is caller-
+     * controlled, so two clients picking the same raw key would lease
+     * each other's warm sessions. HMAC-scoping with a boot-time nonce
+     * hides the raw value from memory dumps but does not protect
+     * against that shared-key hijack. Operators who need multi-tenant
+     * isolation should either disable the feature or front the server
+     * with an auth proxy that rewrites `prompt_cache_key` per tenant.
+     *
+     * **Prerequisites (ALL must hold, else the field is a silent no-op):**
+     *
+     *   1. `MLX_DISABLE_PROMPT_CACHE_KEY` must NOT be set to `"1"` in the
+     *      server environment (default behavior is enabled).
+     *   2. The key must be at least 8 characters. Shorter values
+     *      (including the empty string) are silently treated as if no
+     *      key were supplied — trivial guessing collisions on short
+     *      keys would be a real risk even in single-tenant use.
+     *   3. `previous_response_id` must NOT be set on the same request.
+     *      Prev-id takes precedence; tier-2 never runs when both are
+     *      present.
+     *
+     * When any prerequisite fails the server FALLS BACK silently to a
+     * cold-start for this turn — no error, no 4xx. Integrators who
+     * depend on warm reuse should verify via the `X-Session-Cache`
+     * response header: `prefix_hit` means tier-2 engaged AND the
+     * native prefix verifier reused tokens; `fresh` means no reuse.
+     */
+    prompt_cache_key?: string;
     /**
      * OpenAI-reserved `metadata` slot, repurposed here to carry MLX-Node
      * extensions. Today this only exposes `retention_seconds` as a
@@ -63,6 +163,28 @@ export interface ResponsesAPIRequest {
          */
         retention_seconds?: number;
     };
+    /**
+     * MLX-Node extension carrier for non-OpenAI fields, namespaced under
+     * `extra_body` to mirror the OpenAI SDK convention for vendor
+     * passthrough. Unknown keys are ignored (additive, forward-compat).
+     *
+     * Currently exposes:
+     *   * `generation_mode`: `"mtp"` forces W6 speculative-decode (sets
+     *     `enableMtp = true`), `"ar"` forces plain autoregressive
+     *     (`enableMtp = false`). Absent / null / unrecognized leaves
+     *     `enableMtp` untouched so the downstream `ChatSession` auto-
+     *     default (true when the model ships an MTP head) applies.
+     *   * `mtp_depth`: positive integer override for the per-call draft
+     *     depth. The server forwards any positive integer ≤ 64 (a sanity
+     *     ceiling that only blocks garbage) and the native per-family
+     *     `resolve_params` owns the real clamps: qwen3.5 native MTP
+     *     [1, 5], gemma4 DSpark capped at the draft block size, gemma4
+     *     assistant drafts [1, 8].
+     */
+    extra_body?: {
+        generation_mode?: string | null;
+        mtp_depth?: number | null;
+    };
 }
 export interface OutputTextPart {
     type: 'output_text';
@@ -101,6 +223,54 @@ export interface ResponseUsage {
         reasoning_tokens: number;
     };
     total_tokens: number;
+    /**
+     * Mirrors the upstream OpenAI Responses API `usage.input_tokens_details`
+     * object. Populated only when the native dispatch reports
+     * `cachedTokens > 0` — a non-zero value means that many prompt tokens
+     * were served from the reused KV-cache prefix on this turn. The
+     * `X-Cached-Tokens` response header carries the same number on
+     * non-streaming responses, but SSE flushes its headers before the
+     * native prefix verifier runs, so streaming clients have to read the
+     * value out of the terminal `response.completed` event's
+     * `usage.input_tokens_details.cached_tokens` field to verify
+     * cache-reuse (see Round 5 Fix #3: streaming `X-Session-Cache` is
+     * documented as non-authoritative on the streaming path — this
+     * in-band number is the authoritative signal).
+     */
+    input_tokens_details?: {
+        cached_tokens: number;
+    };
+    /**
+     * Server-extension timing fields surfaced for verbose logs. These
+     * are native/server inference measurements, distinct from the HTTP
+     * logger's outer `elapsedMs` envelope. Unknown fields are ignored by
+     * OpenAI-compatible clients.
+     */
+    time_to_first_token_ms?: number;
+    prefill_tokens_per_second?: number;
+    decode_tokens_per_second?: number;
+    server_inference_elapsed_ms?: number;
+    server_time_to_first_token_ms?: number;
+    server_total_time_to_first_token_ms?: number;
+    server_prefill_tokens_per_second?: number;
+    server_decode_tokens_per_second?: number;
+    server_model_resolve_ms?: number;
+    server_load_wait_ms?: number;
+    server_load_owner?: boolean;
+    server_queue_ms?: number;
+    server_pre_inference_ms?: number;
+    server_paged_prefill_chunk_size?: number;
+    server_paged_prefill_eval_interval?: number;
+    server_paged_decode_cache_clear_interval?: number;
+    /**
+     * Server-extension cache context for `prefill_tokens_per_second`.
+     * On cached-prefix turns, `prefill_input_tokens` is the uncached
+     * suffix that was actually prefetched and `cached_prefix_tokens`
+     * is the skipped prefix, so verbose logs do not mistake suffix-only
+     * prefill throughput for full-prompt throughput.
+     */
+    prefill_input_tokens?: number;
+    cached_prefix_tokens?: number;
 }
 export interface ResponseError {
     type: string;

package/dist/types.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA,2GAA2G;AAE3G,MAAM,WAAW,aAAa;IAC5B,IAAI,EAAE,YAAY,CAAC;IACnB,IAAI,EAAE,MAAM,CAAC;CACd;AAED,MAAM,MAAM,WAAW,GAAG,aAAa,CAAC;AAMxC,MAAM,WAAW,YAAY;IAC3B,IAAI,CAAC,EAAE,SAAS,CAAC;IACjB,IAAI,EAAE,MAAM,GAAG,WAAW,GAAG,QAAQ,GAAG,WAAW,CAAC;IACpD,OAAO,EAAE,MAAM,GAAG,WAAW,EAAE,CAAC;CACjC;AAED,MAAM,WAAW,iBAAiB;IAChC,IAAI,EAAE,eAAe,CAAC;IACtB,EAAE,EAAE,MAAM,CAAC;IACX,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,uBAAuB;IACtC,IAAI,EAAE,sBAAsB,CAAC;IAC7B,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,MAAM,SAAS,GAAG,YAAY,GAAG,iBAAiB,GAAG,uBAAuB,CAAC;AAMnF,MAAM,WAAW,uBAAuB;IACtC,IAAI,EAAE,UAAU,CAAC;IACjB,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,UAAU,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACrC,MAAM,CAAC,EAAE,OAAO,CAAC;CAClB;AAMD,MAAM,WAAW,mBAAmB;IAClC,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,GAAG,SAAS,EAAE,CAAC;IAC5B,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,KAAK,CAAC,EAAE,uBAAuB,EAAE,CAAC;IAClC,WAAW,CAAC,EAAE,MAAM,GAAG,UAAU,GAAG,MAAM,GAAG;QAAE,IAAI,EAAE,UAAU,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAC;IAChF,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,SAAS,CAAC,EAAE;QAAE,MAAM,CAAC,EAAE,MAAM,CAAC;QAAC,OAAO,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;IAClD,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAC9B,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB;;;;;OAKG;IACH,QAAQ,CAAC,EAAE;QACT;;;;;WAKG;QACH,iBAAiB,CAAC,EAAE,MAAM,CAAC;KAC5B,CAAC;CACH;AAMD,MAAM,WAAW,cAAc;IAC7B,IAAI,EAAE,aAAa,CAAC;IACpB,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,KAAK,EAAE,CAAC;CACtB;AAED,MAAM,WAAW,eAAe;IAC9B,IAAI,EAAE,cAAc,CAAC;IACrB,IAAI,EAAE,MAAM,CAAC;CACd;AAED,MAAM,WAAW,iBAAiB;IAChC,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,SAAS,CAAC;IAChB,IAAI,EAAE,WAAW,CAAC;IAClB,MAAM,EAAE,WAAW,GAAG,YAAY,GAAG,aAAa,CAAC;IACnD,OAAO,EAAE,cAAc,EAAE,CAAC;CAC3B;AAED,MAAM,WAAW,mBAAmB;IAClC,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,WAAW,CAAC;IAClB,OAAO,EAAE,eAAe,EAAE,CAAC;CAC5B;AAED,MAAM,WAAW,sBAAsB;IACrC,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,eAAe,CAAC;IACtB,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,WAAW,GAAG,YAAY,CAAC;CACpC;AAED,MAAM,MAAM,UAAU,GAAG,iBAAiB,GAAG,mBAAmB,GAAG,sBAAsB,CAAC;AAM1F,MAAM,WAAW,aAAa;IAC5B,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,qBAAqB,EAAE;QAAE,gBAAgB,EAAE,MAAM,CAAA;KAAE,CAAC;IACpD,YAAY,EAAE,MAAM,CAAC;CACtB;AAMD,MAAM,WAAW,aAAa;IAC5B,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,GAAG,IAAI,CAAC;IACpB,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;CACtB;AAMD,MAAM,WAAW,cAAc;IAC7B,EAAE,EAAE,MAAM,CAAC;IACX,MAAM,EAAE,UAAU,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,MAAM,EAAE,WAAW,GAAG,YAAY,GAAG,aAAa,GAAG,QAAQ,CAAC;IAC9D,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,UAAU,EAAE,CAAC;IACrB,WAAW,EAAE,MAAM,CAAC;IACpB,KAAK,EAAE,aAAa,GAAG,IAAI,CAAC;IAC5B,kBAAkB,EAAE;QAAE,MAAM,EAAE,MAAM,CAAA;KAAE,GAAG,IAAI,CAAC;IAC9C,KAAK,EAAE,aAAa,CAAC;IACrB,YAAY,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,iBAAiB,EAAE,MAAM,GAAG,IAAI,CAAC;IACjC,KAAK,EAAE,uBAAuB,EAAE,CAAC;IACjC,WAAW,EAAE,MAAM,GAAG,UAAU,GAAG,MAAM,GAAG;QAAE,IAAI,EAAE,UAAU,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,GAAG,IAAI,CAAC;IACtF,SAAS,EAAE;QAAE,MAAM,CAAC,EAAE,MAAM,CAAC;QAAC,OAAO,CAAC,EAAE,MAAM,CAAA;KAAE,GAAG,IAAI,CAAC;IACxD,oBAAoB,EAAE,MAAM,GAAG,IAAI,CAAC;CACrC;AAMD,MAAM,WAAW,oBAAoB;IACnC,IAAI,EAAE,kBAAkB,CAAC;IACzB,QAAQ,EAAE,cAAc,CAAC;CAC1B;AAED,MAAM,WAAW,uBAAuB;IACtC,IAAI,EAAE,sBAAsB,CAAC;IAC7B,QAAQ,EAAE,cAAc,CAAC;CAC1B;AAED,MAAM,WAAW,sBAAsB;IACrC,IAAI,EAAE,oBAAoB,CAAC;IAC3B,QAAQ,EAAE,cAAc,CAAC;CAC1B;AAED,MAAM,WAAW,mBAAmB;IAClC,IAAI,EAAE,iBAAiB,CAAC;IACxB,QAAQ,EAAE,cAAc,CAAC;CAC1B;AAED,MAAM,WAAW,oBAAoB;IACnC,IAAI,EAAE,4BAA4B,CAAC;IACnC,YAAY,EAAE,MAAM,CAAC;IACrB,IAAI,EAAE,UAAU,CAAC;CAClB;AAED,MAAM,WAAW,mBAAmB;IAClC,IAAI,EAAE,2BAA2B,CAAC;IAClC,YAAY,EAAE,MAAM,CAAC;IACrB,IAAI,EAAE,UAAU,CAAC;CAClB;AAED,MAAM,WAAW,qBAAqB;IACpC,IAAI,EAAE,6BAA6B,CAAC;IACpC,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,IAAI,EAAE,cAAc,CAAC;CACtB;AAED,MAAM,WAAW,oBAAoB;IACnC,IAAI,EAAE,4BAA4B,CAAC;IACnC,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,IAAI,EAAE,cAAc,CAAC;CACtB;AAED,MAAM,WAAW,oBAAoB;IACnC,IAAI,EAAE,4BAA4B,CAAC;IACnC,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;CACf;AAED,MAAM,WAAW,mBAAmB;IAClC,IAAI,EAAE,2BAA2B,CAAC;IAClC,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,IAAI,EAAE,MAAM,CAAC;CACd;AAED,MAAM,WAAW,8BAA8B;IAC7C,IAAI,EAAE,uCAAuC,CAAC;IAC9C,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;CACf;AAED,MAAM,WAAW,6BAA6B;IAC5C,IAAI,EAAE,sCAAsC,CAAC;IAC7C,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,IAAI,EAAE,MAAM,CAAC;CACd;AAED,MAAM,WAAW,+BAA+B;IAC9C,IAAI,EAAE,wCAAwC,CAAC;IAC/C,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,MAAM,CAAC;IACrB,KAAK,EAAE,MAAM,CAAC;CACf;AAED,MAAM,WAAW,8BAA8B;IAC7C,IAAI,EAAE,uCAAuC,CAAC;IAC9C,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,MAAM,CAAC;IACrB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,MAAM,WAAW,GACnB,oBAAoB,GACpB,uBAAuB,GACvB,sBAAsB,GACtB,mBAAmB,GACnB,oBAAoB,GACpB,mBAAmB,GACnB,qBAAqB,GACrB,oBAAoB,GACpB,oBAAoB,GACpB,mBAAmB,GACnB,8BAA8B,GAC9B,6BAA6B,GAC7B,+BAA+B,GAC/B,8BAA8B,CAAC"}
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA,2GAA2G;AAE3G,MAAM,WAAW,aAAa;IAC5B,IAAI,EAAE,YAAY,CAAC;IACnB,IAAI,EAAE,MAAM,CAAC;CACd;AAED;;;;GAIG;AACH,MAAM,WAAW,cAAc;IAC7B,IAAI,EAAE,aAAa,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,MAAM,CAAC,EAAE,MAAM,GAAG,KAAK,GAAG,MAAM,CAAC;CAClC;AAED;;;;GAIG;AACH,MAAM,WAAW,sBAAsB;IACrC,IAAI,EAAE,aAAa,CAAC;IACpB,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,CAAC,EAAE,KAAK,EAAE,CAAC;CACvB;AAED,+EAA+E;AAC/E,MAAM,WAAW,gBAAgB;IAC/B,IAAI,EAAE,SAAS,CAAC;IAChB,OAAO,EAAE,MAAM,CAAC;CACjB;AAED;;;GAGG;AACH,MAAM,WAAW,oBAAoB;IACnC,IAAI,EAAE,cAAc,CAAC;IACrB,IAAI,EAAE,MAAM,CAAC;CACd;AAED,MAAM,MAAM,WAAW,GACnB,aAAa,GACb,cAAc,GACd,sBAAsB,GACtB,gBAAgB,GAChB,oBAAoB,CAAC;AAMzB,MAAM,WAAW,YAAY;IAC3B,IAAI,CAAC,EAAE,SAAS,CAAC;IACjB,IAAI,EAAE,MAAM,GAAG,WAAW,GAAG,QAAQ,GAAG,WAAW,CAAC;IACpD,OAAO,EAAE,MAAM,GAAG,WAAW,EAAE,CAAC;CACjC;AAED,MAAM,WAAW,iBAAiB;IAChC,IAAI,EAAE,eAAe,CAAC;IACtB,EAAE,EAAE,MAAM,CAAC;IACX,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,uBAAuB;IACtC,IAAI,EAAE,sBAAsB,CAAC;IAC7B,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;CAChB;AAED;;;;GAIG;AACH,MAAM,WAAW,kBAAkB;IACjC,IAAI,EAAE,WAAW,CAAC;IAClB,EAAE,CAAC,EAAE,MAAM,CAAC;IACZ,OAAO,CAAC,EAAE;QAAE,IAAI,CAAC,EAAE,cAAc,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,EAAE,CAAC;IACpD,iBAAiB,CAAC,EAAE,MAAM,CAAC;CAC5B;AAED,MAAM,MAAM,SAAS,GAAG,YAAY,GAAG,iBAAiB,GAAG,uBAAuB,GAAG,kBAAkB,CAAC;AAMxG,MAAM,WAAW,uBAAuB;IACtC,IAAI,EAAE,UAAU,CAAC;IACjB,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,UAAU,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACrC,MAAM,CAAC,EAAE,OAAO,CAAC;CAClB;AAMD,MAAM,WAAW,mBAAmB;IAClC,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,GAAG,SAAS,EAAE,CAAC;IAC5B,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,KAAK,CAAC,EAAE,uBAAuB,EAAE,CAAC;IAClC,WAAW,CAAC,EAAE,MAAM,GAAG,UAAU,GAAG,MAAM,GAAG;QAAE,IAAI,EAAE,UAAU,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAC;IAChF,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,SAAS,CAAC,EAAE;QAAE,MAAM,CAAC,EAAE,MAAM,CAAC;QAAC,OAAO,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;IAClD,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAC9B,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;OAkDG;IACH,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B;;;;;OAKG;IACH,QAAQ,CAAC,EAAE;QACT;;;;;WAKG;QACH,iBAAiB,CAAC,EAAE,MAAM,CAAC;KAC5B,CAAC;IACF;;;;;;;;;;;;;;;;;OAiBG;IACH,UAAU,CAAC,EAAE;QAMX,eAAe,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;QAChC,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;KAC3B,CAAC;CACH;AAMD,MAAM,WAAW,cAAc;IAC7B,IAAI,EAAE,aAAa,CAAC;IACpB,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,KAAK,EAAE,CAAC;CACtB;AAED,MAAM,WAAW,eAAe;IAC9B,IAAI,EAAE,cAAc,CAAC;IACrB,IAAI,EAAE,MAAM,CAAC;CACd;AAED,MAAM,WAAW,iBAAiB;IAChC,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,SAAS,CAAC;IAChB,IAAI,EAAE,WAAW,CAAC;IAClB,MAAM,EAAE,WAAW,GAAG,YAAY,GAAG,aAAa,CAAC;IACnD,OAAO,EAAE,cAAc,EAAE,CAAC;CAC3B;AAED,MAAM,WAAW,mBAAmB;IAClC,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,WAAW,CAAC;IAClB,OAAO,EAAE,eAAe,EAAE,CAAC;CAC5B;AAED,MAAM,WAAW,sBAAsB;IACrC,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,eAAe,CAAC;IACtB,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,WAAW,GAAG,YAAY,CAAC;CACpC;AAED,MAAM,MAAM,UAAU,GAAG,iBAAiB,GAAG,mBAAmB,GAAG,sBAAsB,CAAC;AAM1F,MAAM,WAAW,aAAa;IAC5B,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,qBAAqB,EAAE;QAAE,gBAAgB,EAAE,MAAM,CAAA;KAAE,CAAC;IACpD,YAAY,EAAE,MAAM,CAAC;IACrB;;;;;;;;;;;;;OAaG;IACH,oBAAoB,CAAC,EAAE;QAAE,aAAa,EAAE,MAAM,CAAA;KAAE,CAAC;IACjD;;;;;OAKG;IACH,sBAAsB,CAAC,EAAE,MAAM,CAAC;IAChC,yBAAyB,CAAC,EAAE,MAAM,CAAC;IACnC,wBAAwB,CAAC,EAAE,MAAM,CAAC;IAClC,2BAA2B,CAAC,EAAE,MAAM,CAAC;IACrC,6BAA6B,CAAC,EAAE,MAAM,CAAC;IACvC,mCAAmC,CAAC,EAAE,MAAM,CAAC;IAC7C,gCAAgC,CAAC,EAAE,MAAM,CAAC;IAC1C,+BAA+B,CAAC,EAAE,MAAM,CAAC;IACzC,uBAAuB,CAAC,EAAE,MAAM,CAAC;IACjC,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,iBAAiB,CAAC,EAAE,OAAO,CAAC;IAC5B,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,uBAAuB,CAAC,EAAE,MAAM,CAAC;IACjC,+BAA+B,CAAC,EAAE,MAAM,CAAC;IACzC,kCAAkC,CAAC,EAAE,MAAM,CAAC;IAC5C,wCAAwC,CAAC,EAAE,MAAM,CAAC;IAClD;;;;;;OAMG;IACH,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAC9B,oBAAoB,CAAC,EAAE,MAAM,CAAC;CAC/B;AAMD,MAAM,WAAW,aAAa;IAC5B,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,GAAG,IAAI,CAAC;IACpB,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;CACtB;AAMD,MAAM,WAAW,cAAc;IAC7B,EAAE,EAAE,MAAM,CAAC;IACX,MAAM,EAAE,UAAU,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,MAAM,EAAE,WAAW,GAAG,YAAY,GAAG,aAAa,GAAG,QAAQ,CAAC;IAC9D,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,UAAU,EAAE,CAAC;IACrB,WAAW,EAAE,MAAM,CAAC;IACpB,KAAK,EAAE,aAAa,GAAG,IAAI,CAAC;IAC5B,kBAAkB,EAAE;QAAE,MAAM,EAAE,MAAM,CAAA;KAAE,GAAG,IAAI,CAAC;IAC9C,KAAK,EAAE,aAAa,CAAC;IACrB,YAAY,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,iBAAiB,EAAE,MAAM,GAAG,IAAI,CAAC;IACjC,KAAK,EAAE,uBAAuB,EAAE,CAAC;IACjC,WAAW,EAAE,MAAM,GAAG,UAAU,GAAG,MAAM,GAAG;QAAE,IAAI,EAAE,UAAU,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,GAAG,IAAI,CAAC;IACtF,SAAS,EAAE;QAAE,MAAM,CAAC,EAAE,MAAM,CAAC;QAAC,OAAO,CAAC,EAAE,MAAM,CAAA;KAAE,GAAG,IAAI,CAAC;IACxD,oBAAoB,EAAE,MAAM,GAAG,IAAI,CAAC;CACrC;AAMD,MAAM,WAAW,oBAAoB;IACnC,IAAI,EAAE,kBAAkB,CAAC;IACzB,QAAQ,EAAE,cAAc,CAAC;CAC1B;AAED,MAAM,WAAW,uBAAuB;IACtC,IAAI,EAAE,sBAAsB,CAAC;IAC7B,QAAQ,EAAE,cAAc,CAAC;CAC1B;AAED,MAAM,WAAW,sBAAsB;IACrC,IAAI,EAAE,oBAAoB,CAAC;IAC3B,QAAQ,EAAE,cAAc,CAAC;CAC1B;AAED,MAAM,WAAW,mBAAmB;IAClC,IAAI,EAAE,iBAAiB,CAAC;IACxB,QAAQ,EAAE,cAAc,CAAC;CAC1B;AAED,MAAM,WAAW,oBAAoB;IACnC,IAAI,EAAE,4BAA4B,CAAC;IACnC,YAAY,EAAE,MAAM,CAAC;IACrB,IAAI,EAAE,UAAU,CAAC;CAClB;AAED,MAAM,WAAW,mBAAmB;IAClC,IAAI,EAAE,2BAA2B,CAAC;IAClC,YAAY,EAAE,MAAM,CAAC;IACrB,IAAI,EAAE,UAAU,CAAC;CAClB;AAED,MAAM,WAAW,qBAAqB;IACpC,IAAI,EAAE,6BAA6B,CAAC;IACpC,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,IAAI,EAAE,cAAc,CAAC;CACtB;AAED,MAAM,WAAW,oBAAoB;IACnC,IAAI,EAAE,4BAA4B,CAAC;IACnC,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,IAAI,EAAE,cAAc,CAAC;CACtB;AAED,MAAM,WAAW,oBAAoB;IACnC,IAAI,EAAE,4BAA4B,CAAC;IACnC,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;CACf;AAED,MAAM,WAAW,mBAAmB;IAClC,IAAI,EAAE,2BAA2B,CAAC;IAClC,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,IAAI,EAAE,MAAM,CAAC;CACd;AAED,MAAM,WAAW,8BAA8B;IAC7C,IAAI,EAAE,uCAAuC,CAAC;IAC9C,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;CACf;AAED,MAAM,WAAW,6BAA6B;IAC5C,IAAI,EAAE,sCAAsC,CAAC;IAC7C,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,IAAI,EAAE,MAAM,CAAC;CACd;AAED,MAAM,WAAW,+BAA+B;IAC9C,IAAI,EAAE,wCAAwC,CAAC;IAC/C,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,MAAM,CAAC;IACrB,KAAK,EAAE,MAAM,CAAC;CACf;AAED,MAAM,WAAW,8BAA8B;IAC7C,IAAI,EAAE,uCAAuC,CAAC;IAC9C,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,MAAM,CAAC;IACrB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,MAAM,WAAW,GACnB,oBAAoB,GACpB,uBAAuB,GACvB,sBAAsB,GACtB,mBAAmB,GACnB,oBAAoB,GACpB,mBAAmB,GACnB,qBAAqB,GACrB,oBAAoB,GACpB,oBAAoB,GACpB,mBAAmB,GACnB,8BAA8B,GAC9B,6BAA6B,GAC7B,+BAA+B,GAC/B,8BAA8B,CAAC"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mlx-node/server",
-  "version": "0.0.7",
+  "version": "0.0.8",
   "homepage": "https://github.com/mlx-node/mlx-node",
   "bugs": {
     "url": "https://github.com/mlx-node/mlx-node/issues"
@@ -27,8 +27,8 @@
     "build": "tsc -b"
   },
   "dependencies": {
-    "@mlx-node/core": "0.0.7",
-    "@mlx-node/lm": "0.0.7"
+    "@mlx-node/core": "0.0.8",
+    "@mlx-node/lm": "0.0.8"
   },
   "devDependencies": {
     "@types/node": "@types/node@24.12.2"