@srouterhq/server 0.0.0-stage → 0.9.0

package/dist/providers/openai-compat.js

@@ -0,0 +1,537 @@
+import { BaseProvider, providerHttpError } from './base.js';
+import { extendedBodyParams, resolveMaxTokens } from '../lib/sampling-params.js';
+import { rescueInlineToolCalls } from '../lib/tool-call-rescue.js';
+import { extractThinkFromMessage } from '../lib/think-tags.js';
+import { repairToolArguments, toolSchemaMap } from '../lib/tool-args.js';
+import { invalidToolCallReasons, isToolArgumentValidationEnabled } from '../lib/tool-validate.js';
+import { recordQuotaObservationsFromResponse } from '../services/provider-quota.js';
+import { providerTimeoutMs } from '../lib/provider-timeout.js';
+import { isAbortLikeError } from '../lib/error-classify.js';
+import { contentToString } from '../lib/content.js';
+import { bearerAuthHeader } from '../lib/credential.js';
+// Shared with the media/embeddings custom-endpoint paths; re-exported so
+// existing imports keep working (#1331).
+export { isAnonymousCredential } from '../lib/credential.js';
+/** Hosts that ARE Moonshot's OpenAI-compatible API (api.moonshot.ai,
+ * api.moonshot.cn, api.kimi.com and their subdomains). */
+const MOONSHOT_HOST_SUFFIXES = ['moonshot.ai', 'moonshot.cn', 'kimi.com'];
+/**
+ * Whether a base URL points at Moonshot's own API. Moonshot documents an
+ * assistant-message `partial: true` prefill flag that no other OpenAI-compatible
+ * upstream understands, and there is no built-in moonshot platform — Kimi
+ * models are otherwise served by Groq, Cloudflare, OpenRouter, Hugging Face,
+ * Ollama, ... — so the flag is gated on the endpoint host, never on the model
+ * id. Returns false for anything that does not parse as a URL. (#1038)
+ */
+export function isMoonshotEndpoint(baseUrl) {
+    let host;
+    try {
+        host = new URL(baseUrl).hostname.toLowerCase();
+    }
+    catch {
+        return false;
+    }
+    return MOONSHOT_HOST_SUFFIXES.some((d) => host === d || host.endsWith(`.${d}`));
+}
+/**
+ * Some free-tier upstreams (notably Pollinations) answer HTTP 200 but put an
+ * out-of-credits / top-up notice in the assistant message instead of returning
+ * a 402. That reads as a successful completion, so the fallback loop never
+ * rotates off the dead key (#pollinations-inband). Detect the notice so callers
+ * can throw a payment-required error and fail over. Kept deliberately tight —
+ * requires the credits phrase AND a top-up/quest/Pollinations marker — so a
+ * genuine reply that merely discusses credits does not trip it.
+ */
+export function inBandCreditsError(text) {
+    if (!text)
+        return null;
+    const t = text.toLowerCase();
+    const mentionsCredits = t.includes('enough credits') || t.includes('insufficient credit');
+    if (!mentionsCredits)
+        return null;
+    const topUpMarker = t.includes('top up') || t.includes('top-up') ||
+        t.includes('complete a quest') || t.includes('pollinations');
+    if (!topUpMarker)
+        return null;
+    return text.trim().slice(0, 200);
+}
+/**
+ * Generic provider for platforms that use an OpenAI-compatible API.
+ * Covers: Groq, Cerebras, NVIDIA NIM, Mistral, OpenRouter,
+ * GitHub Models, Fireworks AI.
+ */
+export class OpenAICompatProvider extends BaseProvider {
+    platform;
+    name;
+    baseUrl;
+    extraHeaders;
+    validateUrl;
+    /** Per-provider HTTP timeout override. OpenAI-compatible gateways often buffer
+     * non-streaming responses until generation completes, and reasoning models can
+     * take >15s before first byte. Default 60000. */
+    timeoutMs;
+    /** NVIDIA NIM models reject any request that permits parallel tool calls with
+     * `400 This model only supports single tool-calls at once!`. When set, pin
+     * parallel_tool_calls to false whenever tools are in play. See issue #255. */
+    forceSingleToolCall;
+    /** True only for a custom endpoint whose host is Moonshot's own API; see
+     * isMoonshotEndpoint(). Gates the assistant `partial` prefill flag (#1038). */
+    forwardsPartial;
+    constructor(opts) {
+        super();
+        this.platform = opts.platform;
+        this.name = opts.name;
+        this.baseUrl = opts.baseUrl;
+        this.extraHeaders = opts.extraHeaders ?? {};
+        this.validateUrl = opts.validateUrl;
+        // PROVIDER_TIMEOUT_<PLATFORM> wins over the registration default (#547).
+        this.timeoutMs = providerTimeoutMs(opts.platform, opts.timeoutMs ?? 60_000);
+        this.keyless = opts.keyless ?? false;
+        this.forceSingleToolCall = opts.forceSingleToolCall ?? false;
+        this.forwardsPartial = opts.platform === 'custom' && isMoonshotEndpoint(opts.baseUrl);
+    }
+    /** Resolve the parallel_tool_calls flag to send upstream. For providers that
+     * only accept single tool calls (NVIDIA NIM), force `false` whenever tools are
+     * present so the model never tries to emit two at once and 400s; otherwise pass
+     * the caller's value through unchanged. See issue #255. */
+    resolveParallelToolCalls(options) {
+        if (this.forceSingleToolCall && options?.tools && options.tools.length > 0)
+            return false;
+        return options?.parallel_tool_calls;
+    }
+    /** Some providers (Groq especially) reject a model's tool call with a 400
+     * `tool_use_failed` when the model emitted it as inline DIALECT TEXT
+     * (`<function=NAME{...}</function>`, Hermes/Qwen XML, etc.) that the provider's
+     * own parser couldn't convert — but they hand back the raw text in
+     * `error.failed_generation`. Weaker tool models (e.g. groq llama-3.3-70b) hit
+     * this constantly, dead-ending an agent's whole turn even though the call is
+     * perfectly recoverable. Reuse the same inline-dialect rescue the proxy already
+     * applies to streamed text: parse `failed_generation` into structured
+     * tool_calls so the turn succeeds instead of failing over (or exhausting the
+     * chain when every enabled tool model behaves the same way). See issue #264. */
+    rescueFailedGeneration(errBody, options) {
+        const failed = errBody?.error?.failed_generation;
+        if (typeof failed !== 'string' || failed.length === 0)
+            return null;
+        const toolNames = new Set((options?.tools ?? []).map(t => t.function.name));
+        if (toolNames.size === 0)
+            return null;
+        const rescue = rescueInlineToolCalls(failed, toolNames);
+        if (!rescue.detected || !rescue.calls?.length)
+            return null;
+        const schemas = toolSchemaMap(options?.tools);
+        const rescued = rescue.calls.map((c, i) => ({
+            id: `call_rescued_${i + 1}`,
+            type: 'function',
+            function: { name: c.name, arguments: repairToolArguments(c.arguments, schemas.get(c.name)) },
+        }));
+        // Opt-in schema verdict. A rescue that produces schema-invalid arguments
+        // has turned the provider's 400 into a "success" the client cannot use, so
+        // decline it instead: the original upstream error propagates and the loop
+        // fails over exactly as it did before the rescue existed. Declining is the
+        // right shape here rather than throwing our own error — we are inside the
+        // provider's error path already.
+        if (isToolArgumentValidationEnabled() && invalidToolCallReasons(rescued, schemas).length > 0) {
+            return null;
+        }
+        return rescued;
+    }
+    /** Extract the useful text from an upstream error body. Most providers put it
+     * at error.message, but NVIDIA NIM answers RFC7807-style ({"title": ...,
+     * "detail": "Function id '...': DEGRADED function cannot be invoked"}) — the
+     * old error.message-only read collapsed that to "Bad Request", so neither the
+     * logs nor the error classifier could ever see the DEGRADED marker (#522). */
+    upstreamErrorText(errBody, res) {
+        const e = errBody;
+        if (typeof e?.error?.message === 'string' && e.error.message)
+            return e.error.message;
+        if (typeof e?.detail === 'string' && e.detail)
+            return e.detail;
+        if (typeof e?.title === 'string' && e.title)
+            return e.title;
+        return res.statusText;
+    }
+    /** Anonymous access sends NO Authorization header: keyless providers
+     * (Kilo's anonymous free tier) holding the stored `no-key` sentinel, and
+     * custom endpoints whose stored credential is the same sentinel with auth
+     * off. Sending `Bearer no-key` upstream is never right — upstreams read it
+     * as an invalid key. A REAL key saved on a keyless platform is used instead
+     * of the anonymous path, and a real key on a custom endpoint still gets its
+     * bearer (#1331): the presence of a credential decides at request time —
+     * Kilo, OVH and AI Horde all accept both modes per their docs. */
+    authHeader(apiKey) {
+        return bearerAuthHeader(apiKey);
+    }
+    /** Requesty's Leanstral route rejects greedy sampling when temperature=0.
+     * Omitting that value and supplying a neutral top_p keeps the caller's intent
+     * deterministic enough while using the provider's supported sampling path. */
+    samplingForModel(modelId, options) {
+        if (this.platform === 'requesty' &&
+            modelId === 'mistral/leanstral-1-5' &&
+            options?.temperature === 0) {
+            return { temperature: undefined, topP: options.top_p ?? 1 };
+        }
+        return { temperature: options?.temperature, topP: options?.top_p };
+    }
+    /**
+     * OpenAI-compatible endpoints that are strict about unknown nested fields:
+     * Mistral returns 422 for provider-private replay fields, Groq rejects
+     * assistant `reasoning_content` with 400 (verified in production, #1070),
+     * and Cerebras rejects it too — `property 'messages.N.assistant.
+     * reasoning_content' is unsupported` (confirmed via vercel/ai#15042 and
+     * opencode#26762, and by Cerebras' own docs, which use a `reasoning` field
+     * instead). Other gateways ignore the fields, so keep the OpenAI wire
+     * shape but strip our internal reasoning / thought-signature extensions
+     * before sending to these platforms.
+     */
+    static STRICT_PLATFORMS = new Set(['mistral', 'groq', 'cerebras']);
+    messagesForPlatform(messages, modelId) {
+        if (OpenAICompatProvider.STRICT_PLATFORMS.has(this.platform)) {
+            // Rebuild every message from a whitelist of keys, so `partial` and the
+            // reasoning extensions never reach these platforms.
+            return messages.map((m) => {
+                if (m.role === 'assistant') {
+                    return {
+                        role: m.role,
+                        content: m.content,
+                        ...(m.name ? { name: m.name } : {}),
+                        ...(m.tool_calls && m.tool_calls.length > 0 ? {
+                            tool_calls: m.tool_calls.map((tc) => ({
+                                id: tc.id,
+                                type: tc.type,
+                                function: {
+                                    name: tc.function.name,
+                                    arguments: tc.function.arguments,
+                                },
+                            })),
+                        } : {}),
+                    };
+                }
+                if (m.role === 'tool') {
+                    return {
+                        role: m.role,
+                        content: m.content,
+                        ...(m.tool_call_id ? { tool_call_id: m.tool_call_id } : {}),
+                        ...(m.name ? { name: m.name } : {}),
+                    };
+                }
+                return {
+                    role: m.role,
+                    content: m.content,
+                    ...(m.name ? { name: m.name } : {}),
+                };
+            });
+        }
+        // Moonshot's assistant `partial` prefill flag only survives to the wire
+        // when this provider IS Moonshot's own API (a custom endpoint on a
+        // Moonshot/Kimi host). Every other OpenAI-compatible gateway — including
+        // the ones that serve Kimi models, like Groq, Cloudflare or OpenRouter —
+        // gets it stripped, since strict upstreams 400/422 on unknown message keys
+        // and the rest would ignore it anyway. (#1038)
+        const sanitized = this.forwardsPartial ? messages : messages.map((m) => {
+            if (m.role === 'assistant' && m.partial !== undefined) {
+                const { partial: _partial, ...rest } = m;
+                return rest;
+            }
+            return m;
+        });
+        // Qwen3.8-Flash-Next accepts exactly one system message, and only at index
+        // zero. Profiles can prepend a gateway system prompt to a client's own
+        // system message, so coalesce every system instruction before dispatch
+        // instead of letting an otherwise valid request fail upstream.
+        if (this.platform === 'radeon' && modelId === 'Qwen3.8-Flash-Next') {
+            const systems = sanitized.filter(m => m.role === 'system');
+            if (systems.length === 1 && sanitized[0]?.role === 'system')
+                return sanitized;
+            const systemText = systems.map(m => contentToString(m.content)).filter(Boolean).join('\n\n');
+            const rest = sanitized.filter(m => m.role !== 'system');
+            return systemText ? [{ role: 'system', content: systemText }, ...rest] : rest;
+        }
+        return sanitized;
+    }
+    async chatCompletion(apiKey, messages, modelId, options, quotaContext) {
+        const sampling = this.samplingForModel(modelId, options);
+        const res = await this.fetchWithTimeout(`${this.baseUrl}/chat/completions`, {
+            method: 'POST',
+            headers: {
+                ...this.authHeader(apiKey),
+                'Content-Type': 'application/json',
+                ...this.extraHeaders,
+            },
+            body: JSON.stringify({
+                model: modelId,
+                messages: this.messagesForPlatform(messages, modelId),
+                temperature: sampling.temperature,
+                max_tokens: resolveMaxTokens(this.platform, options?.max_tokens, options?.contextBudget),
+                top_p: sampling.topP,
+                stop: options?.stop,
+                tools: options?.tools,
+                tool_choice: options?.tool_choice,
+                parallel_tool_calls: this.resolveParallelToolCalls(options),
+                ...extendedBodyParams(this.platform, options),
+            }),
+            // 'request' bounds: the deadline covers the body read too, so a 200
+            // whose body hangs aborts instead of stalling res.json() forever.
+        }, options?.timeoutMs ?? this.timeoutMs, { signal: options?.signal, timeoutBounds: 'request' });
+        recordQuotaObservationsFromResponse(res, {
+            platform: this.platform,
+            keyId: quotaContext?.keyId,
+            providerAccountId: quotaContext?.providerAccountId,
+            modelId,
+            quotaPoolKey: quotaContext?.quotaPoolKey,
+            endpoint: 'chat/completions',
+        });
+        if (!res.ok) {
+            const err = await res.json().catch(() => ({}));
+            const rescued = this.rescueFailedGeneration(err, options);
+            if (rescued) {
+                console.log(`[${this.name}] Rescued ${rescued.length} inline tool call(s) from a ${res.status} tool_use_failed (#264)`);
+                const out = {
+                    id: `chatcmpl-rescued-${Date.now()}`,
+                    object: 'chat.completion',
+                    created: Math.floor(Date.now() / 1000),
+                    model: modelId,
+                    choices: [{ index: 0, message: { role: 'assistant', content: null, tool_calls: rescued }, finish_reason: 'tool_calls' }],
+                    usage: { prompt_tokens: 0, completion_tokens: 0, total_tokens: 0 },
+                };
+                out._routed_via = { platform: this.platform, model: modelId };
+                return out;
+            }
+            throw providerHttpError(res, `${this.name} API error ${res.status}: ${this.upstreamErrorText(err, res)}`, err);
+        }
+        let data;
+        let parseErr;
+        try {
+            data = await res.json();
+        }
+        catch (err) {
+            // An aborted body read (per-attempt deadline, client disconnect) is not
+            // a malformed body — rethrow so it keeps its abort classification
+            // instead of reading as a "non-OpenAI-compatible endpoint" below.
+            if (isAbortLikeError(err))
+                throw err;
+            parseErr = err;
+            data = undefined;
+        }
+        if (!data) {
+            // A 200 whose body isn't a single JSON document. Two distinct causes:
+            //   (1) base URL points at a non-OpenAI-compatible API (Ollama's native
+            //       NDJSON /api endpoints, llama.cpp's non-/v1 server, etc., #189).
+            //       Typical signals: Content-Type is application/x-ndjson or
+            //       text/event-stream; parser sees a SECOND JSON object after the
+            //       first one ends ("Unexpected non-whitespace character after JSON
+            //       at position <n> (line <n> column <n>)") where <n> sits inside
+            //       the whitespace between two valid JSON documents.
+            //   (2) the upstream connection was cut short mid-response — most often
+            //       Cloudflare's 600-second edge idle keepalive dropping a slow
+            //       free-tier queue (Kilo provider, NVIDIA nemotron / Poolside
+            //       Laguna models on Cloudflare-fronted upstreams). Signals: body
+            //       ends inside a string or mid-token, parser sees
+            //       "Unexpected end of JSON input"; Content-Type is application/json
+            //       (CF proxies it transparently); latency_ms ≈ 600000.
+            // Without this split every Cloudflare-truncated request was logged as
+            // "endpoint is not OpenAI-compatible", which sent operators chasing a
+            // base-URL config bug that doesn't exist (#430).
+            const msg = parseErr instanceof Error ? parseErr.message : String(parseErr);
+            const contentType = (res.headers?.get?.('content-type') ?? '').toLowerCase();
+            const looksLikeNdjson = /ndjson|text\/event-stream|x-ndjson/.test(contentType);
+            const looksLikeJson = /application\/json/.test(contentType);
+            // Truncation = body parsed cleanly until mid-stream EOF/mid-token garbage.
+            // Only attribute it to a CDN keepalive when the upstream claims to be
+            // sending JSON (Content-Type: application/json). Without that hint,
+            // the parser is more likely choking on NDJSON, native API output, or
+            // HTML — all "wrong endpoint" cases. This is the safe default.
+            const looksTruncated = /Unexpected end of JSON input/.test(msg) ||
+                (/Unexpected non-whitespace character after JSON at position/.test(msg) && looksLikeJson && !looksLikeNdjson);
+            if (looksTruncated) {
+                throw new Error(`${this.name} returned 200 but the response body was truncated mid-stream ` +
+                    `(likely an idle-keepalive timeout at an upstream proxy or CDN, e.g. Cloudflare's ` +
+                    `600s edge limit). Retry, or switch to a faster model/upstream.`);
+            }
+            throw new Error(`${this.name} returned 200 with a non-JSON body — the endpoint is not OpenAI-compatible. ` +
+                `Check the base URL (for Ollama use http://host:11434/v1, for llama.cpp/vLLM/LM Studio the /v1 path).`);
+        }
+        normalizeChoices(data);
+        // #pollinations-inband: a 200 whose body is really an out-of-credits notice
+        // must fail over, not be returned as the answer. The "402 ... insufficient
+        // credit" wording makes isPaymentRequiredError classify it as retryable and
+        // bench the key with the day-long payment-required cooldown.
+        const creditsNotice = inBandCreditsError((data.choices ?? []).map(c => contentToString(c.message?.content)).join('\n'));
+        if (creditsNotice) {
+            throw new Error(`${this.name} API error 402: insufficient credit (upstream returned 200 with an out-of-credits notice): ${creditsNotice}`);
+        }
+        data._routed_via = { platform: this.platform, model: modelId };
+        return data;
+    }
+    async *streamChatCompletion(apiKey, messages, modelId, options, quotaContext) {
+        const sampling = this.samplingForModel(modelId, options);
+        const res = await this.fetchWithTimeout(`${this.baseUrl}/chat/completions`, {
+            method: 'POST',
+            headers: {
+                ...this.authHeader(apiKey),
+                'Content-Type': 'application/json',
+                ...this.extraHeaders,
+            },
+            body: JSON.stringify({
+                model: modelId,
+                messages: this.messagesForPlatform(messages, modelId),
+                temperature: sampling.temperature,
+                max_tokens: resolveMaxTokens(this.platform, options?.max_tokens, options?.contextBudget),
+                top_p: sampling.topP,
+                stop: options?.stop,
+                tools: options?.tools,
+                tool_choice: options?.tool_choice,
+                parallel_tool_calls: this.resolveParallelToolCalls(options),
+                ...extendedBodyParams(this.platform, options),
+                stream: true,
+                stream_options: options?.stream_options,
+            }),
+            // Default 'headers' bounds: the deadline dies at response headers, and
+            // the client signal + stall watchdog own the stream from there.
+        }, options?.timeoutMs ?? this.timeoutMs, { signal: options?.signal });
+        recordQuotaObservationsFromResponse(res, {
+            platform: this.platform,
+            keyId: quotaContext?.keyId,
+            providerAccountId: quotaContext?.providerAccountId,
+            modelId,
+            quotaPoolKey: quotaContext?.quotaPoolKey,
+            endpoint: 'chat/completions',
+        });
+        if (!res.ok) {
+            const err = await res.json().catch(() => ({}));
+            const rescued = this.rescueFailedGeneration(err, options);
+            if (rescued) {
+                console.log(`[${this.name}] Rescued ${rescued.length} inline tool call(s) from a ${res.status} tool_use_failed (stream, #264)`);
+                const base = { id: `chatcmpl-rescued-${Date.now()}`, object: 'chat.completion.chunk', created: Math.floor(Date.now() / 1000), model: modelId };
+                yield { ...base, choices: [{ index: 0, delta: { role: 'assistant' }, finish_reason: null }] };
+                yield { ...base, choices: [{ index: 0, delta: { tool_calls: rescued.map((c, i) => ({ index: i, ...c })) }, finish_reason: null }] };
+                yield { ...base, choices: [{ index: 0, delta: {}, finish_reason: 'tool_calls' }] };
+                return;
+            }
+            throw providerHttpError(res, `${this.name} API error ${res.status}: ${this.upstreamErrorText(err, res)}`, err);
+        }
+        // First-byte grace (#584): the same chat timeout that bounded the headers
+        // also budgets the first stream read — NIM-style providers send SSE
+        // headers instantly, then prefill long prompts for minutes.
+        yield* this.guardInBandCreditsError(this.readSseStream(res, { firstByteTimeoutMs: options?.timeoutMs ?? this.timeoutMs }));
+    }
+    /**
+     * #pollinations-inband: wrap a chat stream so an in-band out-of-credits notice
+     * (a 200 that streams the top-up message as content) fails over instead of
+     * being shown as the answer. Only the FIRST content-bearing chunk is inspected
+     * — a Pollinations credits notice arrives as a single canned message — so the
+     * stream is otherwise byte-for-byte unchanged: reasoning/role chunks pass
+     * straight through (first-token ttfb intact) and every chunk after the first
+     * content one is untouched. Throwing on that first content chunk happens before
+     * it reaches the proxy's commit point, so the fallback loop can still rotate.
+     */
+    async *guardInBandCreditsError(src) {
+        let sawContent = false;
+        for await (const chunk of src) {
+            if (!sawContent) {
+                const content = (chunk.choices ?? [])
+                    .map(c => c.delta?.content)
+                    .filter((c) => typeof c === 'string')
+                    .join('');
+                if (content) {
+                    sawContent = true;
+                    const notice = inBandCreditsError(content);
+                    if (notice) {
+                        throw new Error(`${this.name} API error 402: insufficient credit (upstream streamed an out-of-credits notice): ${notice}`);
+                    }
+                }
+            }
+            yield chunk;
+        }
+    }
+    /** This provider's OpenAI-style model catalog URL. */
+    get modelsUrl() {
+        return `${this.baseUrl}/models`;
+    }
+    /**
+     * GET a catalog-style endpoint with this provider's auth header, extra
+     * headers, proxy routing, timeout policy and quota bookkeeping. Shared by
+     * validateKey (which only reads the status) and custom-endpoint model
+     * discovery (#488), which also reads the body — one place owns how we talk
+     * to a provider's /models route.
+     *
+     * Note: transport errors (DNS / timeout / TLS) propagate to the caller.
+     * health.ts catches them and marks status='error' WITHOUT incrementing the
+     * consecutive-failure counter — only confirmed 401/403 disables a key.
+     */
+    fetchCatalogEndpoint(url, apiKey, quotaContext) {
+        // 30s (not 10s): some upstreams return a large /v1/models catalog that
+        // takes >10s from high-latency regions (e.g. NVIDIA NIM measured ~11.2s
+        // from India). A 10s cap aborted those calls and health.ts marked a
+        // perfectly good key status='error'. 30s aligns with chatCompletion's
+        // own slow-upstream allowance and costs nothing for fast providers.
+        return this.fetchWithTimeout(url, {
+            method: 'GET',
+            headers: {
+                ...this.authHeader(apiKey),
+                ...this.extraHeaders,
+            },
+            // 'request' bounds: a catalog body that hangs mid-transfer must not
+            // stall the health cycle past the deadline.
+        }, 30000, { timeoutBounds: 'request' }).then(res => {
+            recordQuotaObservationsFromResponse(res, {
+                platform: this.platform,
+                keyId: quotaContext?.keyId,
+                providerAccountId: quotaContext?.providerAccountId,
+                quotaPoolKey: quotaContext?.quotaPoolKey,
+                endpoint: 'models',
+            });
+            return res;
+        });
+    }
+    /** The raw `${baseUrl}/models` response, body unread. Used by custom-endpoint
+     *  model discovery (#488); unlike validateKey it always hits /models, never a
+     *  provider-specific validateUrl, because the caller wants the catalog. */
+    fetchModelCatalog(apiKey, quotaContext) {
+        return this.fetchCatalogEndpoint(this.modelsUrl, apiKey, quotaContext);
+    }
+    async validateKey(apiKey, quotaContext) {
+        const res = await this.fetchCatalogEndpoint(this.validateUrl ?? this.modelsUrl, apiKey, quotaContext);
+        return this.validationResult(res);
+    }
+}
+/**
+ * Some providers (Z.ai glm-4.5-flash, Cloudflare DeepSeek-R1-distill, others)
+ * return reasoning models' actual answer in `message.reasoning_content` with
+ * `message.content === ""`. Fold reasoning_content into content so OpenAI-
+ * compatible clients see a non-empty assistant message.
+ *
+ * Other providers (Mistral magistral-medium) return `message.content` as an
+ * array of text segments instead of a string. Flatten to string.
+ */
+function normalizeChoices(data) {
+    for (const choice of data.choices ?? []) {
+        const msg = choice.message;
+        // Flatten array content (Mistral magistral) → join text segments.
+        if (Array.isArray(msg.content)) {
+            msg.content = msg.content
+                .map(seg => (typeof seg === 'string' ? seg : (seg.text ?? '')))
+                .join('');
+        }
+        // Inline `<think>…</think>` extraction (DeepSeek-style) BEFORE the fold
+        // below: a leading think block moves out of content into
+        // reasoning_content. Runs before the fold so a think-only message (no
+        // answer after the block) still folds back into content and is never
+        // returned as an empty assistant message.
+        extractThinkFromMessage(msg);
+        // Fold reasoning into content if content is empty AND there are no
+        // tool_calls. With tool_calls present, content=null is the correct OpenAI
+        // shape; folding reasoning would confuse clients that branch on content.
+        // Field naming varies by provider: Z.ai uses `reasoning_content`, Ollama
+        // uses `reasoning`. Prefer `reasoning_content` when both are set.
+        const hasToolCalls = Array.isArray(msg.tool_calls) && msg.tool_calls.length > 0;
+        if (!hasToolCalls && (msg.content === '' || msg.content == null)) {
+            const fold = (typeof msg.reasoning_content === 'string' && msg.reasoning_content.length > 0)
+                ? msg.reasoning_content
+                : (typeof msg.reasoning === 'string' && msg.reasoning.length > 0 ? msg.reasoning : null);
+            if (fold !== null)
+                msg.content = fold;
+        }
+    }
+}
+//# sourceMappingURL=openai-compat.js.map

package/dist/providers/openai-compat.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"openai-compat.js","sourceRoot":"","sources":["../../src/providers/openai-compat.ts"],"names":[],"mappings":"AAOA,OAAO,EAAE,YAAY,EAAE,iBAAiB,EAAoD,MAAM,WAAW,CAAC;AAC9G,OAAO,EAAE,kBAAkB,EAAE,gBAAgB,EAAE,MAAM,2BAA2B,CAAC;AACjF,OAAO,EAAE,qBAAqB,EAAE,MAAM,4BAA4B,CAAC;AACnE,OAAO,EAAE,uBAAuB,EAAE,MAAM,sBAAsB,CAAC;AAC/D,OAAO,EAAE,mBAAmB,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AACzE,OAAO,EAAE,sBAAsB,EAAE,+BAA+B,EAAE,MAAM,yBAAyB,CAAC;AAClG,OAAO,EAAE,mCAAmC,EAAgC,MAAM,+BAA+B,CAAC;AAClH,OAAO,EAAE,iBAAiB,EAAE,MAAM,4BAA4B,CAAC;AAC/D,OAAO,EAAE,gBAAgB,EAAE,MAAM,0BAA0B,CAAC;AAC5D,OAAO,EAAE,eAAe,EAAE,MAAM,mBAAmB,CAAC;AACpD,OAAO,EAAE,gBAAgB,EAAE,MAAM,sBAAsB,CAAC;AAExD,yEAAyE;AACzE,yCAAyC;AACzC,OAAO,EAAE,qBAAqB,EAAE,MAAM,sBAAsB,CAAC;AAE7D;0DAC0D;AAC1D,MAAM,sBAAsB,GAAG,CAAC,aAAa,EAAE,aAAa,EAAE,UAAU,CAAC,CAAC;AAE1E;;;;;;;GAOG;AACH,MAAM,UAAU,kBAAkB,CAAC,OAAe;IAChD,IAAI,IAAY,CAAC;IACjB,IAAI,CAAC;QACH,IAAI,GAAG,IAAI,GAAG,CAAC,OAAO,CAAC,CAAC,QAAQ,CAAC,WAAW,EAAE,CAAC;IACjD,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,KAAK,CAAC;IACf,CAAC;IACD,OAAO,sBAAsB,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,KAAK,CAAC,IAAI,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC;AAClF,CAAC;AAED;;;;;;;;GAQG;AACH,MAAM,UAAU,kBAAkB,CAAC,IAA+B;IAChE,IAAI,CAAC,IAAI;QAAE,OAAO,IAAI,CAAC;IACvB,MAAM,CAAC,GAAG,IAAI,CAAC,WAAW,EAAE,CAAC;IAC7B,MAAM,eAAe,GAAG,CAAC,CAAC,QAAQ,CAAC,gBAAgB,CAAC,IAAI,CAAC,CAAC,QAAQ,CAAC,qBAAqB,CAAC,CAAC;IAC1F,IAAI,CAAC,eAAe;QAAE,OAAO,IAAI,CAAC;IAClC,MAAM,WAAW,GACf,CAAC,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,QAAQ,CAAC,QAAQ,CAAC;QAC5C,CAAC,CAAC,QAAQ,CAAC,kBAAkB,CAAC,IAAI,CAAC,CAAC,QAAQ,CAAC,cAAc,CAAC,CAAC;IAC/D,IAAI,CAAC,WAAW;QAAE,OAAO,IAAI,CAAC;IAC9B,OAAO,IAAI,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;AACnC,CAAC;AAED;;;;GAIG;AACH,MAAM,OAAO,oBAAqB,SAAQ,YAAY;IAC3C,QAAQ,CAAW;IACnB,IAAI,CAAS;IACL,OAAO,CAAS;IAChB,YAAY,CAAyB;IACrC,WAAW,CAAU;IACtC;;qDAEiD;IAChC,SAAS,CAAS;IACnC;;kFAE8E;IAC7D,mBAAmB,CAAU;IAC9C;mFAC+E;IAC9D,eAAe,CAAU;IAE1C,YAAY,IASX;QACC,KAAK,EAAE,CAAC;QACR,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC,QAAQ,CAAC;QAC9B,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC;QACtB,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,OAAO,CAAC;QAC5B,IAAI,CAAC,YAAY,GAAG,IAAI,CAAC,YAAY,IAAI,EAAE,CAAC;QAC5C,IAAI,CAAC,WAAW,GAAG,IAAI,CAAC,WAAW,CAAC;QACpC,yEAAyE;QACzE,IAAI,CAAC,SAAS,GAAG,iBAAiB,CAAC,IAAI,CAAC,QAAQ,EAAE,IAAI,CAAC,SAAS,IAAI,MAAM,CAAC,CAAC;QAC5E,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,OAAO,IAAI,KAAK,CAAC;QACrC,IAAI,CAAC,mBAAmB,GAAG,IAAI,CAAC,mBAAmB,IAAI,KAAK,CAAC;QAC7D,IAAI,CAAC,eAAe,GAAG,IAAI,CAAC,QAAQ,KAAK,QAAQ,IAAI,kBAAkB,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;IACxF,CAAC;IAED;;;+DAG2D;IACnD,wBAAwB,CAAC,OAA2B;QAC1D,IAAI,IAAI,CAAC,mBAAmB,IAAI,OAAO,EAAE,KAAK,IAAI,OAAO,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC;YAAE,OAAO,KAAK,CAAC;QACzF,OAAO,OAAO,EAAE,mBAAmB,CAAC;IACtC,CAAC;IAED;;;;;;;;;oFASgF;IACxE,sBAAsB,CAAC,OAAgB,EAAE,OAA2B;QAC1E,MAAM,MAAM,GAAI,OAAuD,EAAE,KAAK,EAAE,iBAAiB,CAAC;QAClG,IAAI,OAAO,MAAM,KAAK,QAAQ,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO,IAAI,CAAC;QACnE,MAAM,SAAS,GAAG,IAAI,GAAG,CAAC,CAAC,OAAO,EAAE,KAAK,IAAI,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC;QAC5E,IAAI,SAAS,CAAC,IAAI,KAAK,CAAC;YAAE,OAAO,IAAI,CAAC;QACtC,MAAM,MAAM,GAAG,qBAAqB,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;QACxD,IAAI,CAAC,MAAM,CAAC,QAAQ,IAAI,CAAC,MAAM,CAAC,KAAK,EAAE,MAAM;YAAE,OAAO,IAAI,CAAC;QAC3D,MAAM,OAAO,GAAG,aAAa,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC;QAC9C,MAAM,OAAO,GAAG,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC;YAC1C,EAAE,EAAE,gBAAgB,CAAC,GAAG,CAAC,EAAE;YAC3B,IAAI,EAAE,UAAmB;YACzB,QAAQ,EAAE,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,SAAS,EAAE,mBAAmB,CAAC,CAAC,CAAC,SAAS,EAAE,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,EAAE;SAC7F,CAAC,CAAC,CAAC;QACJ,yEAAyE;QACzE,2EAA2E;QAC3E,0EAA0E;QAC1E,2EAA2E;QAC3E,0EAA0E;QAC1E,iCAAiC;QACjC,IAAI,+BAA+B,EAAE,IAAI,sBAAsB,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC7F,OAAO,IAAI,CAAC;QACd,CAAC;QACD,OAAO,OAAO,CAAC;IACjB,CAAC;IAED;;;;kFAI8E;IACtE,iBAAiB,CAAC,OAAgB,EAAE,GAAa;QACvD,MAAM,CAAC,GAAG,OAA+E,CAAC;QAC1F,IAAI,OAAO,CAAC,EAAE,KAAK,EAAE,OAAO,KAAK,QAAQ,IAAI,CAAC,CAAC,KAAK,CAAC,OAAO;YAAE,OAAO,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC;QACrF,IAAI,OAAO,CAAC,EAAE,MAAM,KAAK,QAAQ,IAAI,CAAC,CAAC,MAAM;YAAE,OAAO,CAAC,CAAC,MAAM,CAAC;QAC/D,IAAI,OAAO,CAAC,EAAE,KAAK,KAAK,QAAQ,IAAI,CAAC,CAAC,KAAK;YAAE,OAAO,CAAC,CAAC,KAAK,CAAC;QAC5D,OAAO,GAAG,CAAC,UAAU,CAAC;IACxB,CAAC;IAED;;;;;;;sEAOkE;IAC1D,UAAU,CAAC,MAAc;QAC/B,OAAO,gBAAgB,CAAC,MAAM,CAAC,CAAC;IAClC,CAAC;IAED;;kFAE8E;IACpE,gBAAgB,CAAC,OAAe,EAAE,OAA2B;QAIrE,IACE,IAAI,CAAC,QAAQ,KAAK,UAAU;YAC5B,OAAO,KAAK,uBAAuB;YACnC,OAAO,EAAE,WAAW,KAAK,CAAC,EAC1B,CAAC;YACD,OAAO,EAAE,WAAW,EAAE,SAAS,EAAE,IAAI,EAAE,OAAO,CAAC,KAAK,IAAI,CAAC,EAAE,CAAC;QAC9D,CAAC;QACD,OAAO,EAAE,WAAW,EAAE,OAAO,EAAE,WAAW,EAAE,IAAI,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;IACrE,CAAC;IAED;;;;;;;;;;OAUG;IACK,MAAM,CAAU,gBAAgB,GAAG,IAAI,GAAG,CAAC,CAAC,SAAS,EAAE,MAAM,EAAE,UAAU,CAAC,CAAC,CAAC;IAC5E,mBAAmB,CAAC,QAAuB,EAAE,OAAe;QAClE,IAAI,oBAAoB,CAAC,gBAAgB,CAAC,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,EAAE,CAAC;YAC7D,uEAAuE;YACvE,oDAAoD;YACpD,OAAO,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE;gBACxB,IAAI,CAAC,CAAC,IAAI,KAAK,WAAW,EAAE,CAAC;oBAC3B,OAAO;wBACL,IAAI,EAAE,CAAC,CAAC,IAAI;wBACZ,OAAO,EAAE,CAAC,CAAC,OAAO;wBAClB,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;wBACnC,GAAG,CAAC,CAAC,CAAC,UAAU,IAAI,CAAC,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC;4BAC5C,UAAU,EAAE,CAAC,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC;gCACpC,EAAE,EAAE,EAAE,CAAC,EAAE;gCACT,IAAI,EAAE,EAAE,CAAC,IAAI;gCACb,QAAQ,EAAE;oCACR,IAAI,EAAE,EAAE,CAAC,QAAQ,CAAC,IAAI;oCACtB,SAAS,EAAE,EAAE,CAAC,QAAQ,CAAC,SAAS;iCACjC;6BACF,CAAC,CAAC;yBACJ,CAAC,CAAC,CAAC,EAAE,CAAC;qBACR,CAAC;gBACJ,CAAC;gBACD,IAAI,CAAC,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;oBACtB,OAAO;wBACL,IAAI,EAAE,CAAC,CAAC,IAAI;wBACZ,OAAO,EAAE,CAAC,CAAC,OAAO;wBAClB,GAAG,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,YAAY,EAAE,CAAC,CAAC,YAAY,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;wBAC3D,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;qBACpC,CAAC;gBACJ,CAAC;gBACD,OAAO;oBACL,IAAI,EAAE,CAAC,CAAC,IAAI;oBACZ,OAAO,EAAE,CAAC,CAAC,OAAO;oBAClB,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;iBACpC,CAAC;YACJ,CAAC,CAAC,CAAC;QACL,CAAC;QAED,wEAAwE;QACxE,mEAAmE;QACnE,yEAAyE;QACzE,yEAAyE;QACzE,2EAA2E;QAC3E,+CAA+C;QAC/C,MAAM,SAAS,GAAG,IAAI,CAAC,eAAe,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE;YACrE,IAAI,CAAC,CAAC,IAAI,KAAK,WAAW,IAAI,CAAC,CAAC,OAAO,KAAK,SAAS,EAAE,CAAC;gBACtD,MAAM,EAAE,OAAO,EAAE,QAAQ,EAAE,GAAG,IAAI,EAAE,GAAG,CAAC,CAAC;gBACzC,OAAO,IAAI,CAAC;YACd,CAAC;YACD,OAAO,CAAC,CAAC;QACX,CAAC,CAAC,CAAC;QAEH,2EAA2E;QAC3E,uEAAuE;QACvE,uEAAuE;QACvE,+DAA+D;QAC/D,IAAI,IAAI,CAAC,QAAQ,KAAK,QAAQ,IAAI,OAAO,KAAK,oBAAoB,EAAE,CAAC;YACnE,MAAM,OAAO,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC;YAC3D,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,IAAI,SAAS,CAAC,CAAC,CAAC,EAAE,IAAI,KAAK,QAAQ;gBAAE,OAAO,SAAS,CAAC;YAC9E,MAAM,UAAU,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,eAAe,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;YAC7F,MAAM,IAAI,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC;YACxD,OAAO,UAAU,CAAC,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,UAAU,EAAE,EAAE,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;QAChF,CAAC;QAED,OAAO,SAAS,CAAC;IACnB,CAAC;IAED,KAAK,CAAC,cAAc,CAClB,MAAc,EACd,QAAuB,EACvB,OAAe,EACf,OAA2B,EAC3B,YAAsC;QAEtC,MAAM,QAAQ,GAAG,IAAI,CAAC,gBAAgB,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;QACzD,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,gBAAgB,CAAC,GAAG,IAAI,CAAC,OAAO,mBAAmB,EAAE;YAC1E,MAAM,EAAE,MAAM;YACd,OAAO,EAAE;gBACP,GAAG,IAAI,CAAC,UAAU,CAAC,MAAM,CAAC;gBAC1B,cAAc,EAAE,kBAAkB;gBAClC,GAAG,IAAI,CAAC,YAAY;aACrB;YACD,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC;gBACnB,KAAK,EAAE,OAAO;gBACd,QAAQ,EAAE,IAAI,CAAC,mBAAmB,CAAC,QAAQ,EAAE,OAAO,CAAC;gBACrD,WAAW,EAAE,QAAQ,CAAC,WAAW;gBACjC,UAAU,EAAE,gBAAgB,CAAC,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,UAAU,EAAE,OAAO,EAAE,aAAa,CAAC;gBACxF,KAAK,EAAE,QAAQ,CAAC,IAAI;gBACpB,IAAI,EAAE,OAAO,EAAE,IAAI;gBACnB,KAAK,EAAE,OAAO,EAAE,KAAK;gBACrB,WAAW,EAAE,OAAO,EAAE,WAAW;gBACjC,mBAAmB,EAAE,IAAI,CAAC,wBAAwB,CAAC,OAAO,CAAC;gBAC3D,GAAG,kBAAkB,CAAC,IAAI,CAAC,QAAQ,EAAE,OAAO,CAAC;aAC9C,CAAC;YACF,oEAAoE;YACpE,kEAAkE;SACnE,EAAE,OAAO,EAAE,SAAS,IAAI,IAAI,CAAC,SAAS,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,aAAa,EAAE,SAAS,EAAE,CAAC,CAAC;QAEhG,mCAAmC,CAAC,GAAG,EAAE;YACvC,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,KAAK,EAAE,YAAY,EAAE,KAAK;YAC1B,iBAAiB,EAAE,YAAY,EAAE,iBAAiB;YAClD,OAAO;YACP,YAAY,EAAE,YAAY,EAAE,YAAY;YACxC,QAAQ,EAAE,kBAAkB;SAC7B,CAAC,CAAC;QAEH,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC;YACZ,MAAM,GAAG,GAAG,MAAM,GAAG,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;YAC/C,MAAM,OAAO,GAAG,IAAI,CAAC,sBAAsB,CAAC,GAAG,EAAE,OAAO,CAAC,CAAC;YAC1D,IAAI,OAAO,EAAE,CAAC;gBACZ,OAAO,CAAC,GAAG,CAAC,IAAI,IAAI,CAAC,IAAI,aAAa,OAAO,CAAC,MAAM,+BAA+B,GAAG,CAAC,MAAM,yBAAyB,CAAC,CAAC;gBACxH,MAAM,GAAG,GAA2B;oBAClC,EAAE,EAAE,oBAAoB,IAAI,CAAC,GAAG,EAAE,EAAE;oBACpC,MAAM,EAAE,iBAAiB;oBACzB,OAAO,EAAE,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,EAAE,GAAG,IAAI,CAAC;oBACtC,KAAK,EAAE,OAAO;oBACd,OAAO,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,OAAO,EAAE,EAAE,IAAI,EAAE,WAAW,EAAE,OAAO,EAAE,IAAyB,EAAE,UAAU,EAAE,OAAO,EAAE,EAAE,aAAa,EAAE,YAAY,EAAE,CAAC;oBAC7I,KAAK,EAAE,EAAE,aAAa,EAAE,CAAC,EAAE,iBAAiB,EAAE,CAAC,EAAE,YAAY,EAAE,CAAC,EAAE;iBACnE,CAAC;gBACF,GAAG,CAAC,WAAW,GAAG,EAAE,QAAQ,EAAE,IAAI,CAAC,QAAQ,EAAE,KAAK,EAAE,OAAO,EAAE,CAAC;gBAC9D,OAAO,GAAG,CAAC;YACb,CAAC;YACD,MAAM,iBAAiB,CAAC,GAAG,EAAE,GAAG,IAAI,CAAC,IAAI,cAAc,GAAG,CAAC,MAAM,KAAK,IAAI,CAAC,iBAAiB,CAAC,GAAG,EAAE,GAAG,CAAC,EAAE,EAAE,GAAG,CAAC,CAAC;QACjH,CAAC;QAED,IAAI,IAA4B,CAAC;QACjC,IAAI,QAAiB,CAAC;QACtB,IAAI,CAAC;YACH,IAAI,GAAG,MAAM,GAAG,CAAC,IAAI,EAA4B,CAAC;QACpD,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,wEAAwE;YACxE,kEAAkE;YAClE,kEAAkE;YAClE,IAAI,gBAAgB,CAAC,GAAG,CAAC;gBAAE,MAAM,GAAG,CAAC;YACrC,QAAQ,GAAG,GAAG,CAAC;YACf,IAAI,GAAG,SAA8C,CAAC;QACxD,CAAC;QACD,IAAI,CAAC,IAAI,EAAE,CAAC;YACV,sEAAsE;YACtE,wEAAwE;YACxE,wEAAwE;YACxE,iEAAiE;YACjE,sEAAsE;YACtE,wEAAwE;YACxE,sEAAsE;YACtE,yDAAyD;YACzD,wEAAwE;YACxE,oEAAoE;YACpE,mEAAmE;YACnE,sEAAsE;YACtE,uDAAuD;YACvD,yEAAyE;YACzE,4DAA4D;YAC5D,sEAAsE;YACtE,sEAAsE;YACtE,iDAAiD;YACjD,MAAM,GAAG,GAAG,QAAQ,YAAY,KAAK,CAAC,CAAC,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;YAC5E,MAAM,WAAW,GAAG,CAAC,GAAG,CAAC,OAAO,EAAE,GAAG,EAAE,CAAC,cAAc,CAAC,IAAI,EAAE,CAAC,CAAC,WAAW,EAAE,CAAC;YAC7E,MAAM,eAAe,GAAG,oCAAoC,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;YAC/E,MAAM,aAAa,GAAG,mBAAmB,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;YAC5D,2EAA2E;YAC3E,sEAAsE;YACtE,oEAAoE;YACpE,qEAAqE;YACrE,+DAA+D;YAC/D,MAAM,cAAc,GAClB,8BAA8B,CAAC,IAAI,CAAC,GAAG,CAAC;gBACxC,CAAC,4DAA4D,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,aAAa,IAAI,CAAC,eAAe,CAAC,CAAC;YAChH,IAAI,cAAc,EAAE,CAAC;gBACnB,MAAM,IAAI,KAAK,CACb,GAAG,IAAI,CAAC,IAAI,+DAA+D;oBAC3E,mFAAmF;oBACnF,gEAAgE,CACjE,CAAC;YACJ,CAAC;YACD,MAAM,IAAI,KAAK,CACb,GAAG,IAAI,CAAC,IAAI,8EAA8E;gBAC1F,sGAAsG,CACvG,CAAC;QACJ,CAAC;QACD,gBAAgB,CAAC,IAAI,CAAC,CAAC;QACvB,4EAA4E;QAC5E,2EAA2E;QAC3E,4EAA4E;QAC5E,6DAA6D;QAC7D,MAAM,aAAa,GAAG,kBAAkB,CACtC,CAAC,IAAI,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,eAAe,CAAE,CAAC,CAAC,OAAuB,EAAE,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAC/F,CAAC;QACF,IAAI,aAAa,EAAE,CAAC;YAClB,MAAM,IAAI,KAAK,CAAC,GAAG,IAAI,CAAC,IAAI,8FAA8F,aAAa,EAAE,CAAC,CAAC;QAC7I,CAAC;QACD,IAAI,CAAC,WAAW,GAAG,EAAE,QAAQ,EAAE,IAAI,CAAC,QAAQ,EAAE,KAAK,EAAE,OAAO,EAAE,CAAC;QAC/D,OAAO,IAAI,CAAC;IACd,CAAC;IAED,KAAK,CAAC,CAAC,oBAAoB,CACzB,MAAc,EACd,QAAuB,EACvB,OAAe,EACf,OAA2B,EAC3B,YAAsC;QAEtC,MAAM,QAAQ,GAAG,IAAI,CAAC,gBAAgB,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;QACzD,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,gBAAgB,CAAC,GAAG,IAAI,CAAC,OAAO,mBAAmB,EAAE;YAC1E,MAAM,EAAE,MAAM;YACd,OAAO,EAAE;gBACP,GAAG,IAAI,CAAC,UAAU,CAAC,MAAM,CAAC;gBAC1B,cAAc,EAAE,kBAAkB;gBAClC,GAAG,IAAI,CAAC,YAAY;aACrB;YACD,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC;gBACnB,KAAK,EAAE,OAAO;gBACd,QAAQ,EAAE,IAAI,CAAC,mBAAmB,CAAC,QAAQ,EAAE,OAAO,CAAC;gBACrD,WAAW,EAAE,QAAQ,CAAC,WAAW;gBACjC,UAAU,EAAE,gBAAgB,CAAC,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,UAAU,EAAE,OAAO,EAAE,aAAa,CAAC;gBACxF,KAAK,EAAE,QAAQ,CAAC,IAAI;gBACpB,IAAI,EAAE,OAAO,EAAE,IAAI;gBACnB,KAAK,EAAE,OAAO,EAAE,KAAK;gBACrB,WAAW,EAAE,OAAO,EAAE,WAAW;gBACjC,mBAAmB,EAAE,IAAI,CAAC,wBAAwB,CAAC,OAAO,CAAC;gBAC3D,GAAG,kBAAkB,CAAC,IAAI,CAAC,QAAQ,EAAE,OAAO,CAAC;gBAC7C,MAAM,EAAE,IAAI;gBACZ,cAAc,EAAE,OAAO,EAAE,cAAc;aACxC,CAAC;YACF,uEAAuE;YACvE,gEAAgE;SACjE,EAAE,OAAO,EAAE,SAAS,IAAI,IAAI,CAAC,SAAS,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;QAEtE,mCAAmC,CAAC,GAAG,EAAE;YACvC,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,KAAK,EAAE,YAAY,EAAE,KAAK;YAC1B,iBAAiB,EAAE,YAAY,EAAE,iBAAiB;YAClD,OAAO;YACP,YAAY,EAAE,YAAY,EAAE,YAAY;YACxC,QAAQ,EAAE,kBAAkB;SAC7B,CAAC,CAAC;QAEH,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC;YACZ,MAAM,GAAG,GAAG,MAAM,GAAG,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;YAC/C,MAAM,OAAO,GAAG,IAAI,CAAC,sBAAsB,CAAC,GAAG,EAAE,OAAO,CAAC,CAAC;YAC1D,IAAI,OAAO,EAAE,CAAC;gBACZ,OAAO,CAAC,GAAG,CAAC,IAAI,IAAI,CAAC,IAAI,aAAa,OAAO,CAAC,MAAM,+BAA+B,GAAG,CAAC,MAAM,iCAAiC,CAAC,CAAC;gBAChI,MAAM,IAAI,GAAG,EAAE,EAAE,EAAE,oBAAoB,IAAI,CAAC,GAAG,EAAE,EAAE,EAAE,MAAM,EAAE,uBAAgC,EAAE,OAAO,EAAE,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,EAAE,GAAG,IAAI,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,CAAC;gBACxJ,MAAM,EAAE,GAAG,IAAI,EAAE,OAAO,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,WAAW,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,EAAE,CAAC;gBAC9F,MAAM,EAAE,GAAG,IAAI,EAAE,OAAO,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,EAAE,UAAU,EAAE,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,CAA8B,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,EAAE,CAAC;gBACjK,MAAM,EAAE,GAAG,IAAI,EAAE,OAAO,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,EAAE,EAAE,aAAa,EAAE,YAAY,EAAE,CAAC,EAAE,CAAC;gBACnF,OAAO;YACT,CAAC;YACD,MAAM,iBAAiB,CAAC,GAAG,EAAE,GAAG,IAAI,CAAC,IAAI,cAAc,GAAG,CAAC,MAAM,KAAK,IAAI,CAAC,iBAAiB,CAAC,GAAG,EAAE,GAAG,CAAC,EAAE,EAAE,GAAG,CAAC,CAAC;QACjH,CAAC;QAED,0EAA0E;QAC1E,oEAAoE;QACpE,4DAA4D;QAC5D,KAAK,CAAC,CAAC,IAAI,CAAC,uBAAuB,CACjC,IAAI,CAAC,aAAa,CAAC,GAAG,EAAE,EAAE,kBAAkB,EAAE,OAAO,EAAE,SAAS,IAAI,IAAI,CAAC,SAAS,EAAE,CAAC,CACtF,CAAC;IACJ,CAAC;IAED;;;;;;;;;OASG;IACK,KAAK,CAAC,CAAC,uBAAuB,CACpC,GAAwC;QAExC,IAAI,UAAU,GAAG,KAAK,CAAC;QACvB,IAAI,KAAK,EAAE,MAAM,KAAK,IAAI,GAAG,EAAE,CAAC;YAC9B,IAAI,CAAC,UAAU,EAAE,CAAC;gBAChB,MAAM,OAAO,GAAG,CAAC,KAAK,CAAC,OAAO,IAAI,EAAE,CAAC;qBAClC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAE,CAAC,CAAC,KAA2C,EAAE,OAAO,CAAC;qBACjE,MAAM,CAAC,CAAC,CAAC,EAAe,EAAE,CAAC,OAAO,CAAC,KAAK,QAAQ,CAAC;qBACjD,IAAI,CAAC,EAAE,CAAC,CAAC;gBACZ,IAAI,OAAO,EAAE,CAAC;oBACZ,UAAU,GAAG,IAAI,CAAC;oBAClB,MAAM,MAAM,GAAG,kBAAkB,CAAC,OAAO,CAAC,CAAC;oBAC3C,IAAI,MAAM,EAAE,CAAC;wBACX,MAAM,IAAI,KAAK,CAAC,GAAG,IAAI,CAAC,IAAI,qFAAqF,MAAM,EAAE,CAAC,CAAC;oBAC7H,CAAC;gBACH,CAAC;YACH,CAAC;YACD,MAAM,KAAK,CAAC;QACd,CAAC;IACH,CAAC;IAED,sDAAsD;IACtD,IAAI,SAAS;QACX,OAAO,GAAG,IAAI,CAAC,OAAO,SAAS,CAAC;IAClC,CAAC;IAED;;;;;;;;;;OAUG;IACO,oBAAoB,CAAC,GAAW,EAAE,MAAc,EAAE,YAAsC;QAChG,uEAAuE;QACvE,wEAAwE;QACxE,oEAAoE;QACpE,sEAAsE;QACtE,oEAAoE;QACpE,OAAO,IAAI,CAAC,gBAAgB,CAAC,GAAG,EAAE;YAChC,MAAM,EAAE,KAAK;YACb,OAAO,EAAE;gBACP,GAAG,IAAI,CAAC,UAAU,CAAC,MAAM,CAAC;gBAC1B,GAAG,IAAI,CAAC,YAAY;aACrB;YACD,oEAAoE;YACpE,4CAA4C;SAC7C,EAAE,KAAK,EAAE,EAAE,aAAa,EAAE,SAAS,EAAE,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE;YACjD,mCAAmC,CAAC,GAAG,EAAE;gBACvC,QAAQ,EAAE,IAAI,CAAC,QAAQ;gBACvB,KAAK,EAAE,YAAY,EAAE,KAAK;gBAC1B,iBAAiB,EAAE,YAAY,EAAE,iBAAiB;gBAClD,YAAY,EAAE,YAAY,EAAE,YAAY;gBACxC,QAAQ,EAAE,QAAQ;aACnB,CAAC,CAAC;YACH,OAAO,GAAG,CAAC;QACb,CAAC,CAAC,CAAC;IACL,CAAC;IAED;;+EAE2E;IAC3E,iBAAiB,CAAC,MAAc,EAAE,YAAsC;QACtE,OAAO,IAAI,CAAC,oBAAoB,CAAC,IAAI,CAAC,SAAS,EAAE,MAAM,EAAE,YAAY,CAAC,CAAC;IACzE,CAAC;IAED,KAAK,CAAC,WAAW,CAAC,MAAc,EAAE,YAAsC;QACtE,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,oBAAoB,CAAC,IAAI,CAAC,WAAW,IAAI,IAAI,CAAC,SAAS,EAAE,MAAM,EAAE,YAAY,CAAC,CAAC;QACtG,OAAO,IAAI,CAAC,gBAAgB,CAAC,GAAG,CAAC,CAAC;IACpC,CAAC;;AAGH;;;;;;;;GAQG;AACH,SAAS,gBAAgB,CAAC,IAA4B;IACpD,KAAK,MAAM,MAAM,IAAI,IAAI,CAAC,OAAO,IAAI,EAAE,EAAE,CAAC;QACxC,MAAM,GAAG,GAAG,MAAM,CAAC,OAIlB,CAAC;QACF,kEAAkE;QAClE,IAAI,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,CAAC;YAC/B,GAAG,CAAC,OAAO,GAAI,GAAG,CAAC,OAAmD;iBACnE,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,OAAO,GAAG,KAAK,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC,CAAC;iBAC9D,IAAI,CAAC,EAAE,CAAC,CAAC;QACd,CAAC;QACD,wEAAwE;QACxE,yDAAyD;QACzD,sEAAsE;QACtE,qEAAqE;QACrE,0CAA0C;QAC1C,uBAAuB,CAAC,GAAG,CAAC,CAAC;QAC7B,mEAAmE;QACnE,0EAA0E;QAC1E,yEAAyE;QACzE,yEAAyE;QACzE,kEAAkE;QAClE,MAAM,YAAY,GAAG,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,UAAU,CAAC,IAAI,GAAG,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC,CAAC;QAChF,IAAI,CAAC,YAAY,IAAI,CAAC,GAAG,CAAC,OAAO,KAAK,EAAE,IAAI,GAAG,CAAC,OAAO,IAAI,IAAI,CAAC,EAAE,CAAC;YACjE,MAAM,IAAI,GAAG,CAAC,OAAO,GAAG,CAAC,iBAAiB,KAAK,QAAQ,IAAI,GAAG,CAAC,iBAAiB,CAAC,MAAM,GAAG,CAAC,CAAC;gBAC1F,CAAC,CAAC,GAAG,CAAC,iBAAiB;gBACvB,CAAC,CAAC,CAAC,OAAO,GAAG,CAAC,SAAS,KAAK,QAAQ,IAAI,GAAG,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;YAC3F,IAAI,IAAI,KAAK,IAAI;gBAAE,GAAG,CAAC,OAAO,GAAG,IAAI,CAAC;QACxC,CAAC;IACH,CAAC;AACH,CAAC"}

package/dist/providers/opencode-free.d.ts

@@ -0,0 +1,24 @@
+import type { ChatCompletionChunk, ChatCompletionResponse, ChatMessage } from '@srouter/shared/types.js';
+import { BaseProvider, type CompletionOptions, type KeyValidationResult } from './base.js';
+import { type QuotaObservationContext } from '../services/provider-quota.js';
+/**
+ * OpenCode Free � keyless provider that spoofs the official OpenCode CLI
+ * identity to access the zen free tier without a real API key.
+ *
+ * The upstream ALWAYS requires stream: true; stream: false or omitted
+ * results in 403 FreeTierError.  Non-streaming callers get a synthetic
+ * single-chunk response assembled from the SSE stream.
+ */
+export declare class OpenCodeFreeProvider extends BaseProvider {
+    readonly platform: "opencode-free";
+    readonly name = "OpenCode Free";
+    keyless: boolean;
+    private identityHeaders;
+    validateKey(_apiKey: string, _quotaContext?: QuotaObservationContext): Promise<KeyValidationResult>;
+    private buildBody;
+    private fetchChat;
+    chatCompletion(_apiKey: string, messages: ChatMessage[], modelId: string, options?: CompletionOptions, quotaContext?: QuotaObservationContext): Promise<ChatCompletionResponse>;
+    streamChatCompletion(_apiKey: string, messages: ChatMessage[], modelId: string, options?: CompletionOptions, quotaContext?: QuotaObservationContext): AsyncGenerator<ChatCompletionChunk>;
+    private assembleResponse;
+}
+//# sourceMappingURL=opencode-free.d.ts.map

package/dist/providers/opencode-free.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"opencode-free.d.ts","sourceRoot":"","sources":["../../src/providers/opencode-free.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACV,mBAAmB,EACnB,sBAAsB,EACtB,WAAW,EAEZ,MAAM,0BAA0B,CAAC;AAClC,OAAO,EACL,YAAY,EAEZ,KAAK,iBAAiB,EACtB,KAAK,mBAAmB,EACzB,MAAM,WAAW,CAAC;AAGnB,OAAO,EAAuC,KAAK,uBAAuB,EAAE,MAAM,+BAA+B,CAAC;AA+FlH;;;;;;;GAOG;AACH,qBAAa,oBAAqB,SAAQ,YAAY;IACpD,QAAQ,CAAC,QAAQ,EAAG,eAAe,CAAU;IAC7C,QAAQ,CAAC,IAAI,mBAAmB;IACvB,OAAO,UAAQ;IAIxB,OAAO,CAAC,eAAe;IAgBjB,WAAW,CACf,OAAO,EAAE,MAAM,EACf,aAAa,CAAC,EAAE,uBAAuB,GACtC,OAAO,CAAC,mBAAmB,CAAC;IAiB/B,OAAO,CAAC,SAAS;YA2BH,SAAS;IAkCjB,cAAc,CAClB,OAAO,EAAE,MAAM,EACf,QAAQ,EAAE,WAAW,EAAE,EACvB,OAAO,EAAE,MAAM,EACf,OAAO,CAAC,EAAE,iBAAiB,EAC3B,YAAY,CAAC,EAAE,uBAAuB,GACrC,OAAO,CAAC,sBAAsB,CAAC;IAsB3B,oBAAoB,CACzB,OAAO,EAAE,MAAM,EACf,QAAQ,EAAE,WAAW,EAAE,EACvB,OAAO,EAAE,MAAM,EACf,OAAO,CAAC,EAAE,iBAAiB,EAC3B,YAAY,CAAC,EAAE,uBAAuB,GACrC,cAAc,CAAC,mBAAmB,CAAC;IAkBtC,OAAO,CAAC,gBAAgB;CAoEzB"}