@proposit/proposit-core 1.11.0 → 2.0.0

package/dist/extensions/openai/provider.js

@@ -30,9 +30,12 @@
 // readability.
 import { debugLlmFailure, debugLlmRequest, debugLlmResponse, } from "../../lib/pipelines/debug-log.js";
 import { typeboxToOpenAiSchema } from "./structured-output.js";
-import { NonRetryableLlmError, QuotaExhaustedLlmError, RateLimitLlmError, ResponseNotFoundError, SchemaValidationLlmError, ToolLoopExhaustedError, TransientLlmError, } from "./errors.js";
+import { deriveSchemaName, findFunctionHandler, translateTools, } from "./openai-tools.js";
+import { fetchResponseEnvelope } from "./openai-http.js";
+import { formatIncompleteMessage, NonRetryableLlmError, ToolLoopExhaustedError, TransientLlmError, } from "./errors.js";
+import { DEFAULT_BASE_URL, } from "./types.js";
+import { extractAssistantText, extractUsage, mergeUsage, pickFunctionCalls, safeParseJson, } from "./openai-parsing.js";
 const STAGE_ID_MARKER = /<!--\s*stage-id:\s*([^\s>]+)\s*-->/;
-const DEFAULT_BASE_URL = "https://api.openai.com/v1/responses";
 const DEFAULT_MAX_TOOL_ROUNDS = 6;
 export function createOpenAiResponsesProvider(options) {
     const baseUrl = options.baseUrl ?? DEFAULT_BASE_URL;
@@ -321,888 +324,4 @@ export function createOpenAiResponsesProvider(options) {
     };
     return { respond };
 }
-/**
- * Retrieve a stored OpenAI response by id. Surfaces the current
- * status, output text (when completed), and token usage.
- *
- * Throws {@link ResponseNotFoundError} when the response is not found
- * (HTTP 404), which typically means the ~10-minute retention window
- * has elapsed. Callers should clear the stored id, settle the
- * associated stage as failed, and surface a retry prompt.
- *
- * @param id - The OpenAI response id to retrieve.
- * @param options - Provider configuration (apiKey, optional baseUrl and fetch).
- */
-export async function retrieveResponse(id, options) {
-    const fetchImpl = resolveFetch(options.fetch, "retrieveResponse");
-    const envelope = await getResponseById({
-        url: options.baseUrl ?? DEFAULT_BASE_URL,
-        id,
-        apiKey: options.apiKey,
-        fetchImpl,
-        signal: options.signal,
-    });
-    return envelopeToRetrievedResponse(envelope, id);
-}
-/**
- * Reconnect to a stored, still-generating background response and
- * **stream it to completion**. This is what actually drives a dropped
- * background response forward: a passive `retrieveResponse` GET only
- * reads the current state and leaves a `queued` / `in_progress`
- * response sitting where it is, whereas reconnecting with `stream=true`
- * resumes consumption so the response reaches a terminal status.
- *
- * Issues `GET /responses/{id}?stream=true&starting_after=<cursor>` and
- * consumes the SSE stream to its terminal event, returning the same
- * {@link TRetrievedResponse} shape as {@link retrieveResponse}.
- *
- * Throws {@link ResponseNotFoundError} when the response is not found
- * (HTTP 404 — typically the ~10-minute retention window elapsed).
- * Honors `signal`: an abort propagates as an `AbortError` from the
- * underlying stream read.
- *
- * @param id - The OpenAI response id to reconnect to.
- * @param options - `apiKey`, optional `startingAfter` SSE cursor
- *   (defaults to 0 — replay from the start of the stored stream),
- *   optional `baseUrl`, `fetch`, and `signal`.
- */
-export async function reconnectStream(id, options) {
-    const fetchImpl = resolveFetch(options.fetch, "reconnectStream");
-    const baseUrl = options.baseUrl ?? DEFAULT_BASE_URL;
-    const startingAfter = options.startingAfter ?? 0;
-    const url = `${baseUrl}/${id}?stream=true&starting_after=${startingAfter.toString()}`;
-    let response;
-    try {
-        response = await fetchImpl(url, {
-            method: "GET",
-            headers: { Authorization: `Bearer ${options.apiKey}` },
-            signal: options.signal,
-        });
-    }
-    catch (err) {
-        if (isAbortError(err))
-            throw err;
-        throw new TransientLlmError({
-            message: `Network error reconnecting to OpenAI background response: ${err instanceof Error ? err.message : String(err)}`,
-        });
-    }
-    if (response.status === 404) {
-        throw new ResponseNotFoundError({ responseId: id });
-    }
-    if (!response.ok) {
-        const errorBody = await response.text().catch(() => "");
-        throw classifyHttpError(response.status, `OpenAI reconnect ${response.status.toString()}: ${errorBody || response.statusText}`);
-    }
-    const envelope = await readSseEnvelope(response);
-    return envelopeToRetrievedResponse(envelope, id);
-}
-/**
- * Cancel a stored, in-flight OpenAI response. Issues
- * `POST /responses/{id}/cancel` and returns the resulting
- * {@link TRetrievedResponse} (typically `status: "cancelled"`).
- *
- * Cancel is **idempotent** per the Responses API: cancelling twice, or
- * cancelling an already-terminal response, simply returns the final
- * `Response` object rather than erroring — so callers do not need to
- * guard against double-cancel.
- *
- * Throws {@link ResponseNotFoundError} when the response is not found
- * (HTTP 404 — typically the ~10-minute retention window elapsed).
- * Honors `signal` (an abort propagates as an `AbortError`).
- *
- * Use this to stop an in-flight background response when a stage is
- * abandoned (resync timeout) or an import is cancelled, so generation
- * does not keep running (and billing) server-side after the consumer
- * has given up on it.
- *
- * @param id - The OpenAI response id to cancel.
- * @param options - `apiKey`, optional `baseUrl`, `fetch`, and `signal`.
- */
-export async function cancelResponse(id, options) {
-    const fetchImpl = resolveFetch(options.fetch, "cancelResponse");
-    const baseUrl = options.baseUrl ?? DEFAULT_BASE_URL;
-    let response;
-    try {
-        response = await fetchImpl(`${baseUrl}/${id}/cancel`, {
-            method: "POST",
-            headers: { Authorization: `Bearer ${options.apiKey}` },
-            signal: options.signal,
-        });
-    }
-    catch (err) {
-        if (isAbortError(err))
-            throw err;
-        throw new TransientLlmError({
-            message: `Network error cancelling OpenAI background response: ${err instanceof Error ? err.message : String(err)}`,
-        });
-    }
-    if (response.status === 404) {
-        throw new ResponseNotFoundError({ responseId: id });
-    }
-    if (!response.ok) {
-        const errorBody = await response.text().catch(() => "");
-        throw classifyHttpError(response.status, `OpenAI cancel ${response.status.toString()}: ${errorBody || response.statusText}`);
-    }
-    const envelope = await parseJsonOrThrowTransient(response, "OpenAI cancel body was not valid JSON");
-    return envelopeToRetrievedResponse(envelope, id);
-}
-/**
- * Submit a background OpenAI response and return its `responseId` +
- * submit-time `status` **without polling or streaming to completion**.
- *
- * This is the submit-only half of the existing `backgroundMode`
- * (`runBackground`): it POSTs `{ background: true, store: true }`, parses
- * the submit envelope, and returns immediately — the caller drives the
- * response to completion later via {@link retrieveResponse} (typically
- * after a durable suspend keyed on the returned `responseId`). It is the
- * provider capability the pipeline's `launchStage` needs.
- *
- * **Terminal-on-submit fast-path:** a small/cached request can come back
- * already terminal (`completed`/`failed`/`incomplete`/`cancelled`) on the
- * submit POST. This function still returns `{ responseId, status }` for
- * that case (no throw, no poll); the caller proceeds to
- * `retrieveResponse(responseId)`, which sees the terminal state
- * immediately.
- *
- * **No-tools precondition:** background mode does not support function
- * tools in V1 — a tool-bearing request throws {@link NonRetryableLlmError},
- * matching `respond`'s background guard.
- *
- * @param req - The structured-output request (system/user prompts +
- *   `outputSchema` + model knobs). Tools are rejected.
- * @param options - `apiKey`, optional `baseUrl`, `fetch`, and `signal`.
- */
-export async function submitBackgroundResponse(req, options) {
-    if (req.tools && req.tools.length > 0) {
-        throw new NonRetryableLlmError({
-            message: "OpenAI background mode does not support function tools in V1. Disable backgroundMode / backgroundStreamMode for tool-using requests, or run the tools synchronously.",
-        });
-    }
-    const fetchImpl = resolveFetch(options.fetch, "submitBackgroundResponse");
-    const baseUrl = options.baseUrl ?? DEFAULT_BASE_URL;
-    const schemaName = deriveSchemaName(req.outputSchema);
-    const convertedSchema = typeboxToOpenAiSchema(req.outputSchema);
-    const body = {
-        model: req.model,
-        input: [
-            { role: "system", content: req.systemPrompt },
-            { role: "user", content: req.userMessage },
-        ],
-        text: {
-            format: {
-                type: "json_schema",
-                name: schemaName,
-                strict: true,
-                schema: convertedSchema,
-            },
-        },
-        background: true,
-        store: true,
-    };
-    if (req.maxOutputTokens !== undefined) {
-        body.max_output_tokens = req.maxOutputTokens;
-    }
-    if (req.reasoningEffort) {
-        body.reasoning = { effort: req.reasoningEffort };
-    }
-    const submit = await callOnce({
-        url: baseUrl,
-        apiKey: options.apiKey,
-        body,
-        fetchImpl,
-        signal: req.signal ?? options.signal,
-    });
-    const submitEnvelope = await parseJsonOrThrowTransient(submit, "OpenAI background submit body was not valid JSON");
-    const id = submitEnvelope.id;
-    if (!id) {
-        throw new TransientLlmError({
-            message: "OpenAI background submit returned no response id.",
-        });
-    }
-    // Return at submit — including the terminal-on-submit fast-path. Unlike
-    // `runBackground`, a `cancelled` submit envelope is NOT thrown here: the
-    // submit-only contract returns every terminal status, and the caller's
-    // completion step classifies it (a cancelled response settles as a skip).
-    return {
-        responseId: id,
-        status: (submitEnvelope.status ?? "queued"),
-    };
-}
-function resolveFetch(injected, fnName) {
-    const fetchImpl = injected ?? globalThis.fetch;
-    if (!fetchImpl) {
-        throw new Error(`${fnName}: no fetch implementation available. Pass \`fetch\` explicitly or run in an environment that provides \`globalThis.fetch\` (Node ≥18, modern browsers, Expo).`);
-    }
-    return fetchImpl;
-}
-function envelopeToRetrievedResponse(envelope, id) {
-    const output = extractAssistantText(envelope.output);
-    const usage = extractUsage(envelope);
-    const result = {
-        status: (envelope.status ?? "in_progress"),
-        rawResponseId: envelope.id ?? id,
-    };
-    if (output !== undefined) {
-        result.output = output;
-    }
-    if (usage.input > 0 || usage.output > 0) {
-        result.tokenUsage = usage;
-    }
-    // Surface the incomplete reason / error message so a completion-side
-    // consumer (e.g. `completeStage`) can classify the outcome.
-    const incompleteReason = envelope.incomplete_details?.reason;
-    if (incompleteReason !== undefined) {
-        result.incompleteReason = incompleteReason;
-    }
-    const errorMessage = envelope.error?.message;
-    if (errorMessage !== undefined) {
-        result.errorMessage = errorMessage;
-    }
-    return result;
-}
-// -- HTTP --
-async function fetchResponseEnvelope(args) {
-    if (args.backgroundStream) {
-        return runBackgroundStream({
-            url: args.url,
-            apiKey: args.apiKey,
-            body: args.body,
-            fetchImpl: args.fetchImpl,
-            signal: args.signal,
-            onResponseId: args.onResponseId,
-        });
-    }
-    if (args.background) {
-        return runBackground({
-            url: args.url,
-            apiKey: args.apiKey,
-            body: args.body,
-            fetchImpl: args.fetchImpl,
-            signal: args.signal,
-            pollIntervalMs: args.pollIntervalMs,
-        });
-    }
-    if (args.stream) {
-        const response = await callOnce({
-            url: args.url,
-            apiKey: args.apiKey,
-            body: { ...args.body, stream: true },
-            fetchImpl: args.fetchImpl,
-            signal: args.signal,
-        });
-        return readSseEnvelope(response);
-    }
-    const response = await callOnce({
-        url: args.url,
-        apiKey: args.apiKey,
-        body: args.body,
-        fetchImpl: args.fetchImpl,
-        signal: args.signal,
-    });
-    return parseJsonOrThrowTransient(response, "OpenAI response body was not valid JSON");
-}
-async function parseJsonOrThrowTransient(response, context) {
-    return response
-        .json()
-        .then((j) => j)
-        .catch((err) => {
-        throw new TransientLlmError({
-            message: `${context}: ${err instanceof Error ? err.message : String(err)}`,
-        });
-    });
-}
-function abortError() {
-    const e = new Error("The OpenAI background request was aborted.");
-    e.name = "AbortError";
-    return e;
-}
-// Resolves (never rejects) on abort by design: the poll loop owns the
-// abort→cancel→throw decision at exactly two checkpoints (top-of-loop and
-// the in-flight-GET catch), so this helper just needs to wake the loop
-// promptly instead of waiting out the full interval. Keeping it
-// non-throwing avoids a second, competing abort surface.
-function abortableDelay(ms, signal) {
-    return new Promise((resolve) => {
-        if (signal?.aborted) {
-            resolve();
-            return;
-        }
-        const onAbort = () => {
-            cleanup();
-            resolve();
-        };
-        const cleanup = () => {
-            clearTimeout(timer);
-            signal?.removeEventListener("abort", onAbort);
-        };
-        const timer = setTimeout(() => {
-            cleanup();
-            resolve();
-        }, ms);
-        signal?.addEventListener("abort", onAbort, { once: true });
-    });
-}
-async function getResponseById(args) {
-    let response;
-    try {
-        response = await args.fetchImpl(`${args.url}/${args.id}`, {
-            method: "GET",
-            headers: { Authorization: `Bearer ${args.apiKey}` },
-            signal: args.signal,
-        });
-    }
-    catch (err) {
-        if (isAbortError(err))
-            throw err;
-        throw new TransientLlmError({
-            message: `Network error polling OpenAI background response: ${err instanceof Error ? err.message : String(err)}`,
-        });
-    }
-    if (response.status === 404) {
-        // Response has aged out of the ~10-minute retention window or
-        // was never stored. Surface as a typed error so callers can
-        // clear the stored id and settle the stage as failed.
-        throw new ResponseNotFoundError({ responseId: args.id });
-    }
-    if (!response.ok) {
-        const errorBody = await response.text().catch(() => "");
-        throw classifyHttpError(response.status, `OpenAI poll ${response.status.toString()}: ${errorBody || response.statusText}`);
-    }
-    return parseJsonOrThrowTransient(response, "OpenAI poll body was not valid JSON");
-}
-// Intentionally takes no AbortSignal — cancel must fire even though the
-// caller's signal has already aborted; passing the fired signal would
-// abort the cancel itself.
-async function cancelBackground(args) {
-    // Best-effort + idempotent — swallow errors; the abort is surfaced
-    // regardless of whether cancel succeeds.
-    try {
-        await args.fetchImpl(`${args.url}/${args.id}/cancel`, {
-            method: "POST",
-            headers: { Authorization: `Bearer ${args.apiKey}` },
-        });
-    }
-    catch {
-        // ignore
-    }
-}
-function isTerminalBackgroundStatus(status) {
-    return (status === "completed" ||
-        status === "failed" ||
-        status === "incomplete" ||
-        status === "cancelled");
-}
-/**
- * Submit a single request with `{ background: true, stream: true,
- * store: true }` and consume the resulting SSE stream live, returning
- * the terminal envelope.
- *
- * A background response can only be streamed if it was *created* with
- * `stream: true` (a background-without-stream response is poll-only and
- * cannot later be streamed), so this mode uses one streaming create
- * call rather than a separate non-streaming submit POST. The response
- * id is therefore not in a JSON POST body — it arrives in the first
- * `response.created` SSE event. `onResponseId` fires the moment that
- * event is parsed (before the terminal event), so the caller can
- * persist the id while the call is still in flight.
- *
- * The response keeps generating server-side even if the connection
- * drops during stream consumption, and can be recovered via
- * `retrieveResponse` within the ~10-minute retention window. A
- * connection drop mid-stream (no terminal event before stream end) is
- * classified as a `TransientLlmError` so the framework's retry policy
- * applies — but because the id was already surfaced mid-flight, a
- * crashed in-flight call can be recovered rather than blindly re-run.
- */
-async function runBackgroundStream(args) {
-    if (args.signal?.aborted)
-        throw abortError();
-    const httpResponse = await callOnce({
-        url: args.url,
-        apiKey: args.apiKey,
-        body: { ...args.body, background: true, stream: true, store: true },
-        fetchImpl: args.fetchImpl,
-        signal: args.signal,
-    });
-    return readSseEnvelope(httpResponse, args.onResponseId);
-}
-async function runBackground(args) {
-    if (args.signal?.aborted)
-        throw abortError();
-    const submit = await callOnce({
-        url: args.url,
-        apiKey: args.apiKey,
-        body: { ...args.body, background: true, store: true },
-        fetchImpl: args.fetchImpl,
-        signal: args.signal,
-    });
-    const submitEnvelope = await parseJsonOrThrowTransient(submit, "OpenAI background submit body was not valid JSON");
-    const id = submitEnvelope.id;
-    if (!id) {
-        throw new TransientLlmError({
-            message: "OpenAI background submit returned no response id.",
-        });
-    }
-    // Fast-path: a small/cached request can come back already terminal on
-    // submit — return it directly rather than issuing a redundant poll GET
-    // (and avoid a `store`-expiry window between submit and first poll).
-    if (isTerminalBackgroundStatus(submitEnvelope.status)) {
-        if (submitEnvelope.status === "cancelled")
-            throw abortError();
-        return submitEnvelope;
-    }
-    for (;;) {
-        if (args.signal?.aborted) {
-            await cancelBackground({
-                url: args.url,
-                id,
-                apiKey: args.apiKey,
-                fetchImpl: args.fetchImpl,
-            });
-            throw abortError();
-        }
-        let env;
-        try {
-            env = await getResponseById({
-                url: args.url,
-                id,
-                apiKey: args.apiKey,
-                fetchImpl: args.fetchImpl,
-                signal: args.signal,
-            });
-        }
-        catch (err) {
-            // Abort landing DURING an in-flight poll GET surfaces as an
-            // AbortError from getResponseById (real fetch rejects the
-            // in-flight request). The top-of-loop check alone would miss
-            // it, so cancel here before re-throwing — this is what makes
-            // the "AbortSignal → cancel POST" guarantee hold mid-poll.
-            if (isAbortError(err)) {
-                await cancelBackground({
-                    url: args.url,
-                    id,
-                    apiKey: args.apiKey,
-                    fetchImpl: args.fetchImpl,
-                });
-                throw abortError();
-            }
-            throw err;
-        }
-        const status = env.status;
-        // Deliberately inlined rather than calling
-        // `isTerminalBackgroundStatus`: `cancelled` is terminal too but
-        // needs the opposite disposition (throw `abortError()` vs. return
-        // the envelope), so the two terminal cases are split here.
-        if (status === "completed" ||
-            status === "failed" ||
-            status === "incomplete") {
-            return env;
-        }
-        if (status === "cancelled") {
-            throw abortError();
-        }
-        await abortableDelay(args.pollIntervalMs, args.signal);
-    }
-}
-async function callOnce(args) {
-    let response;
-    try {
-        response = await args.fetchImpl(args.url, {
-            method: "POST",
-            headers: {
-                "Content-Type": "application/json",
-                Authorization: `Bearer ${args.apiKey}`,
-            },
-            body: JSON.stringify(args.body),
-            signal: args.signal,
-        });
-    }
-    catch (err) {
-        // Aborts come through as a thrown error with `name === "AbortError"`
-        // (DOM spec). Surface as-is so `llmStage`'s mid-flight-abort
-        // detector turns it into `StageAbortedError`. Other low-level
-        // fetch failures are transient.
-        if (isAbortError(err)) {
-            throw err;
-        }
-        throw new TransientLlmError({
-            message: `Network error calling OpenAI: ${err instanceof Error ? err.message : String(err)}`,
-        });
-    }
-    if (response.ok) {
-        return response;
-    }
-    const errorBody = await response.text().catch(() => "");
-    const message = `OpenAI Responses API ${response.status.toString()}: ${errorBody || response.statusText}`;
-    // Best-effort structured extraction of the provider error
-    // code/type so a 429 can be split into persistent quota exhaustion
-    // vs. transient throttling. Never throws on malformed bodies — an
-    // unparseable body leaves `providerErrorCode` undefined, which
-    // `classifyHttpError` treats as the safe transient default.
-    let providerErrorCode;
-    try {
-        const parsed = JSON.parse(errorBody);
-        providerErrorCode = parsed.error?.code ?? parsed.error?.type;
-    }
-    catch {
-        providerErrorCode = undefined;
-    }
-    throw classifyHttpError(response.status, message, providerErrorCode);
-}
-function classifyHttpError(status, message, providerErrorCode) {
-    if (status >= 500) {
-        return new TransientLlmError({ message, status });
-    }
-    if (status === 429) {
-        // 429 splits on the body's structured error code: persistent
-        // budget exhaustion (`insufficient_quota`) is fail-fast, every
-        // other (and every unparseable) 429 stays the transient
-        // throttle. The safe default is always "transient + retryable"
-        // — never a false quota trip.
-        if (providerErrorCode === "insufficient_quota") {
-            return new QuotaExhaustedLlmError({ message, status });
-        }
-        return new RateLimitLlmError({ message, status });
-    }
-    // 400 vs 422 split:
-    //
-    // OpenAI returns 400 for malformed requests — typically a
-    // converter bug, an unsupported parameter, or a request shape
-    // the API doesn't accept. Retrying a 400 just burns the second
-    // attempt; classify as non-retryable so the framework surfaces
-    // the error immediately.
-    //
-    // OpenAI returns 422 when the model's structured-output reply
-    // failed server-side strict-mode validation. A re-roll can
-    // sometimes succeed, so we route 422 through the
-    // schema-validation class (which carries `retryReason:
-    // "transient"` as the V1 retry workaround until the framework
-    // grows a dedicated `schema_validation` retry tag).
-    if (status === 400) {
-        return new NonRetryableLlmError({ message, status });
-    }
-    if (status === 422) {
-        return new SchemaValidationLlmError({ message, status });
-    }
-    return new NonRetryableLlmError({ message, status });
-}
-function isAbortError(err) {
-    return (typeof err === "object" &&
-        err !== null &&
-        err.name === "AbortError");
-}
-// -- incomplete-reason → user-facing message ----------------------------
-//
-// Per-reason error messages for the `status: "incomplete"` branch.
-// The message is the dev's first read when a stage fails — keep it
-// actionable, name the cap reason verbatim, and (for the truncation
-// case specifically) point at the override knob that fixes it.
-function formatIncompleteMessage(reason) {
-    if (reason === "max_output_tokens") {
-        return `OpenAI Responses API returned status: "incomplete" (reason: max_output_tokens). The model's output exceeded the per-call \`max_output_tokens\` cap (either an explicit value on the request or the model's default). Pass a larger \`maxOutputTokens\` to TLlmRequest, or set the stage-level \`maxOutputTokens\` on the llmStage factory (e.g., the \`createIngestionV2Pipeline({ llm: { overrides: { ... } } })\` surface).`;
-    }
-    if (reason === "content_filter") {
-        return `OpenAI Responses API returned status: "incomplete" (reason: content_filter). OpenAI's content policy refused to complete this output; the input or generated content was flagged. Retrying will not succeed — review the input text or the stage's prompt and re-request.`;
-    }
-    return `OpenAI Responses API returned status: "incomplete" (reason: ${reason}). The model stopped before completing the response. See OpenAI Responses API documentation for the complete \`incomplete_details.reason\` enumeration.`;
-}
-// -- response parsing --
-const SSE_TERMINAL_EVENTS = new Set([
-    "response.completed",
-    "response.incomplete",
-    "response.failed",
-]);
-// The lifecycle event the Responses API emits first, before any output
-// chunks. It carries the response object — including `.id` — so a
-// streaming consumer learns the id while the call is still in flight.
-const SSE_CREATED_EVENT = "response.created";
-/**
- * Parse one SSE event block. Returns:
- *
- *   * `{ kind: "terminal", envelope }` for a terminal Responses-API
- *     event (`response.completed` / `.incomplete` / `.failed`),
- *     carrying the embedded full `response` envelope;
- *   * `{ kind: "created", responseId }` for the lifecycle
- *     `response.created` event, surfacing the response id the moment
- *     it is known (before any output);
- *   * `undefined` for every intermediate / unrecognized event.
- *
- * The events carry a `type` field inside the data JSON, so we key off
- * that and fall back to the SSE `event:` line.
- */
-function parseSseEvent(raw) {
-    let eventType;
-    const dataLines = [];
-    for (const line of raw.split("\n")) {
-        if (line.startsWith(":"))
-            continue; // SSE comment line — ignore.
-        if (line.startsWith("event:")) {
-            eventType = line.slice(6).trim();
-        }
-        else if (line.startsWith("data:")) {
-            dataLines.push(line.slice(5).replace(/^ /, ""));
-        }
-    }
-    if (dataLines.length === 0)
-        return undefined;
-    let parsed;
-    try {
-        parsed = JSON.parse(dataLines.join("\n"));
-    }
-    catch {
-        return undefined;
-    }
-    // Prefer the discriminator inside the data JSON (Responses-API
-    // events carry `type` there); fall back to the SSE `event:` line so
-    // the parser is robust if the API ever sends the type only there.
-    // (No `[DONE]` sentinel handling — that is a Chat-Completions frame
-    // the Responses API does not emit.)
-    const type = parsed.type ?? eventType;
-    if (type && SSE_TERMINAL_EVENTS.has(type) && parsed.response) {
-        return { kind: "terminal", envelope: parsed.response };
-    }
-    if (type === SSE_CREATED_EVENT && parsed.response?.id) {
-        return { kind: "created", responseId: parsed.response.id };
-    }
-    return undefined;
-}
-/**
- * Read an SSE `text/event-stream` body and return the envelope carried
- * by the terminal event. A stream that ends with no terminal event
- * (connection drop) throws `TransientLlmError` so the framework retries.
- * `AbortError` from the underlying reader propagates verbatim so
- * `llmStage` marks the stage `skipped`.
- *
- * `onResponseId`, when supplied, fires the moment the `response.created`
- * lifecycle event is parsed — i.e. while the call is still streaming,
- * before the terminal event arrives. This is the load-bearing seam for
- * background-stream mode: it lets a caller persist the response id
- * mid-flight so an in-flight call interrupted before completion can be
- * recovered from the upstream's stored copy. Invoked at most once.
- *
- * Note: the event-separator scan assumes LF (`\n\n`) framing, which the
- * OpenAI Responses API emits. Reuse against a strict CRLF-only SSE
- * server would need the separator scan adjusted to `\r\n\r\n`.
- */
-async function readSseEnvelope(response, onResponseId) {
-    const body = response.body;
-    if (!body) {
-        throw new TransientLlmError({
-            message: "OpenAI streaming response carried no body.",
-        });
-    }
-    const reader = body.getReader();
-    const decoder = new TextDecoder();
-    let buffer = "";
-    let terminal;
-    let idSurfaced = false;
-    const handleEvent = (rawEvent) => {
-        const parsedEvent = parseSseEvent(rawEvent);
-        if (!parsedEvent)
-            return;
-        if (parsedEvent.kind === "terminal") {
-            terminal = parsedEvent.envelope;
-            return;
-        }
-        // kind === "created": surface the id once, the moment it's known.
-        if (!idSurfaced) {
-            idSurfaced = true;
-            onResponseId?.(parsedEvent.responseId);
-        }
-    };
-    try {
-        for (;;) {
-            const chunk = await reader.read();
-            if (chunk.done)
-                break;
-            buffer += decoder.decode(chunk.value, { stream: true });
-            let sep = buffer.indexOf("\n\n");
-            while (sep !== -1) {
-                const rawEvent = buffer.slice(0, sep);
-                buffer = buffer.slice(sep + 2);
-                handleEvent(rawEvent);
-                sep = buffer.indexOf("\n\n");
-            }
-        }
-    }
-    catch (err) {
-        if (isAbortError(err)) {
-            throw err;
-        }
-        throw new TransientLlmError({
-            message: `OpenAI streaming read failed: ${err instanceof Error ? err.message : String(err)}`,
-        });
-    }
-    // Flush any bytes the streaming decoder is still holding (an
-    // incomplete multi-byte UTF-8 sequence at the final chunk boundary),
-    // then scan for any remaining terminal frame.
-    buffer += decoder.decode();
-    let tailSep = buffer.indexOf("\n\n");
-    while (tailSep !== -1) {
-        const rawEvent = buffer.slice(0, tailSep);
-        buffer = buffer.slice(tailSep + 2);
-        handleEvent(rawEvent);
-        tailSep = buffer.indexOf("\n\n");
-    }
-    if (!terminal) {
-        throw new TransientLlmError({
-            message: "OpenAI streaming ended without a terminal response event (connection drop?).",
-        });
-    }
-    return terminal;
-}
-function pickFunctionCalls(output) {
-    if (!output)
-        return [];
-    const calls = [];
-    for (const item of output) {
-        if (item.type === "function_call") {
-            /* eslint-disable @typescript-eslint/naming-convention */
-            const fc = item;
-            /* eslint-enable @typescript-eslint/naming-convention */
-            calls.push({
-                callId: fc.call_id,
-                name: fc.name,
-                arguments: fc.arguments,
-            });
-        }
-    }
-    return calls;
-}
-function extractAssistantText(output) {
-    if (!output)
-        return undefined;
-    for (const item of output) {
-        if (item.type === "message") {
-            const msg = item;
-            const blocks = msg.content ?? [];
-            const textBlock = blocks.find((b) => b.type === "output_text" || b.type === "text");
-            if (textBlock?.text !== undefined) {
-                return textBlock.text;
-            }
-        }
-    }
-    return undefined;
-}
-function safeParseJson(raw) {
-    try {
-        return JSON.parse(raw);
-    }
-    catch (err) {
-        throw new SchemaValidationLlmError({
-            message: `OpenAI returned malformed JSON in structured-output text: ${err instanceof Error ? err.message : String(err)}`,
-        });
-    }
-}
-function extractUsage(envelope) {
-    const usage = envelope.usage;
-    if (!usage)
-        return { input: 0, output: 0 };
-    const result = {
-        input: usage.input_tokens ?? 0,
-        output: usage.output_tokens ?? 0,
-    };
-    const reasoning = usage.output_tokens_details?.reasoning_tokens;
-    if (reasoning !== undefined) {
-        result.reasoning = reasoning;
-    }
-    return result;
-}
-function mergeUsage(accumulated, next) {
-    const merged = {
-        input: accumulated.input + next.input,
-        output: accumulated.output + next.output,
-    };
-    if (accumulated.reasoning !== undefined || next.reasoning !== undefined) {
-        merged.reasoning = (accumulated.reasoning ?? 0) + (next.reasoning ?? 0);
-    }
-    return merged;
-}
-// -- tool translation --
-function translateTools(tools) {
-    return tools.map((tool) => {
-        switch (tool.kind) {
-            case "web_search":
-                return { type: "web_search" };
-            case "file_search":
-                return {
-                    type: "file_search",
-                    vector_store_ids: [tool.vectorStoreId],
-                };
-            case "mcp": {
-                const out = {
-                    type: "mcp",
-                    server_url: tool.serverUrl,
-                };
-                if (tool.toolName) {
-                    return {
-                        ...out,
-                        allowed_tools: [tool.toolName],
-                    };
-                }
-                return out;
-            }
-            case "function":
-                return {
-                    type: "function",
-                    name: tool.name,
-                    description: tool.description,
-                    parameters: typeboxToOpenAiSchema(tool.parameters),
-                    strict: true,
-                };
-        }
-    });
-}
-function findFunctionHandler(tools, name) {
-    if (!tools)
-        return undefined;
-    for (const tool of tools) {
-        if (tool.kind === "function" && tool.name === name) {
-            return tool;
-        }
-    }
-    return undefined;
-}
-// -- schema name derivation --
-function deriveSchemaName(schema) {
-    const id = schema.$id;
-    if (typeof id === "string" && id.length > 0) {
-        return sanitizeName(id);
-    }
-    const serialized = canonicalJson(schema);
-    return `schema_${shortHash(serialized)}`;
-}
-function sanitizeName(raw) {
-    // OpenAI requires schema names match `^[a-zA-Z0-9_-]{1,64}$`.
-    const cleaned = raw.replace(/[^a-zA-Z0-9_-]/g, "_").slice(0, 64);
-    return cleaned.length > 0 ? cleaned : "schema";
-}
-function canonicalJson(value) {
-    return JSON.stringify(value, (_key, v) => {
-        if (v !== null && typeof v === "object" && !Array.isArray(v)) {
-            const obj = v;
-            const sorted = {};
-            for (const key of Object.keys(obj).sort()) {
-                sorted[key] = obj[key];
-            }
-            return sorted;
-        }
-        return v;
-    });
-}
-function shortHash(input) {
-    // Stable 12-hex-char hash. We avoid pulling `crypto.subtle` because
-    // it's async and would force `deriveSchemaName` to be async too;
-    // FNV-1a is sufficient for naming uniqueness within a process.
-    let h1 = 0xcbf29ce4;
-    let h2 = 0x84222325;
-    for (let i = 0; i < input.length; i += 1) {
-        const c = input.charCodeAt(i);
-        h1 = Math.imul(h1 ^ c, 0x01000193) >>> 0;
-        h2 = Math.imul(h2 ^ c, 0x01000193) >>> 0;
-    }
-    const hex1 = h1.toString(16).padStart(8, "0");
-    const hex2 = h2.toString(16).padStart(8, "0");
-    return (hex1 + hex2).slice(0, 12);
-}
 //# sourceMappingURL=provider.js.map