@agnt-sdk/studio 0.0.34 → 0.0.36

package/dist/providers/openaiCompatible.js

@@ -0,0 +1,241 @@
+/**
+ * OpenAICompatibleExecutor — shared adapter for OpenAI-compatible providers
+ *
+ * One adapter for every provider that speaks the OpenAI Chat Completions wire
+ * format: Together AI (Kimi / Qwen), Fireworks, DeepInfra, DeepSeek, and any
+ * future open-model host. Per-provider differences — base URL, credentials,
+ * model IDs, pricing, cache behavior — live in CONFIG, never in code:
+ *
+ *   - baseURL:  resolved from OPENAI_COMPATIBLE_BASE_URLS[provider], overridable
+ *               by credentials[provider].baseURL or model metadata.baseURL.
+ *   - apiKey:   credentials[provider].apiKey (via the existing secret manager;
+ *               never stored in manifest/config rows).
+ *   - model:    manifest.spec.models[].model.
+ *   - params:   manifest.spec.models[].metadata (temperature, top_p, …).
+ *
+ * Adding a new OpenAI-compatible provider is therefore a config + credentials
+ * change (new AiModel row + creds entry), with no new adapter code required —
+ * unless it needs a base URL we don't know, which is a one-line registry entry.
+ *
+ * CACHING: these providers do AUTOMATIC (best-effort) prefix caching — the
+ * provider caches transparently, there is no billed cache write, and cached
+ * reads are reported back in usage.prompt_tokens_details.cached_tokens. We map
+ * that to cache_read_input_tokens and leave cache_creation_input_tokens at 0
+ * (there is no write concept here; the backend renders it as "—" / null based
+ * on the model's cacheMode). We never send Anthropic-style cache_control blocks
+ * — they are not part of the OpenAI wire format and must not leak through.
+ */
+import OpenAI from 'openai';
+import BaseExecutor from '../BaseExecutor.js';
+import { streamWithRetry, consumeOpenAIStream, STREAM_ABSOLUTE_BACKSTOP_MS } from './streaming.js';
+/** Known OpenAI-compatible provider base URLs. A provider not listed here is
+ *  still usable by supplying baseURL via credentials or model metadata. */
+export const OPENAI_COMPATIBLE_BASE_URLS = {
+    together: 'https://api.together.ai/v1',
+    fireworks: 'https://api.fireworks.ai/inference/v1',
+    deepinfra: 'https://api.deepinfra.com/v1/openai',
+    deepseek: 'https://api.deepseek.com/v1',
+};
+/** Providers routed through this adapter. Keep in sync with executorFactory. */
+export const OPENAI_COMPATIBLE_PROVIDERS = new Set(Object.keys(OPENAI_COMPATIBLE_BASE_URLS));
+export default class OpenAICompatibleExecutor extends BaseExecutor {
+    client;
+    providerName;
+    constructor(config) {
+        super(config);
+        // this.provider is set by BaseExecutor from the selected model config.
+        this.providerName = (this.provider || '').toLowerCase();
+        // Credentials are keyed by provider name. Real provider strings are
+        // lowercase ('together', 'deepseek'), but tolerate the original casing too.
+        const credMap = this.credentials;
+        const creds = credMap[this.provider] ?? credMap[this.providerName];
+        if (!creds) {
+            throw new Error(`[OpenAICompatibleExecutor] credentials.${this.providerName} is required`);
+        }
+        if (!creds.apiKey) {
+            throw new Error(`[OpenAICompatibleExecutor] credentials.${this.providerName}.apiKey is required`);
+        }
+        // baseURL resolution order: creds override → model metadata → registry.
+        const metadata = (this.primaryModelConfig.metadata || {});
+        const baseURL = creds.baseURL ||
+            metadata.baseURL ||
+            OPENAI_COMPATIBLE_BASE_URLS[this.providerName];
+        if (!baseURL) {
+            throw new Error(`[OpenAICompatibleExecutor] no baseURL for provider "${this.providerName}" — ` +
+                `add one to OPENAI_COMPATIBLE_BASE_URLS, credentials.${this.providerName}.baseURL, or model metadata.baseURL`);
+        }
+        // Initialize OpenAI-compatible client. invoke() STREAMS and bounds the
+        // response with an inter-chunk IDLE timeout (see streaming.ts), so a
+        // long-but-progressing turn — exactly Kimi doing multi-tool agentic work —
+        // never races a total-completion timeout. The client `timeout` here is only
+        // the SDK's own absolute request cap; set it to the streaming backstop and
+        // let the idle timeout be the operative ceiling. maxRetries matches the
+        // other streamed adapters.
+        this.client = new OpenAI({
+            apiKey: creds.apiKey,
+            baseURL,
+            maxRetries: 3,
+            timeout: STREAM_ABSOLUTE_BACKSTOP_MS,
+            dangerouslyAllowBrowser: creds.dangerouslyAllowBrowser,
+        });
+        this.log(`[OpenAICompatibleExecutor] Initialized ${this.providerName} @ ${baseURL} with model: ${this.model}`);
+    }
+    /**
+     * Invoke the OpenAI-compatible API.
+     * Returns: { message: { role, content, tool_calls }, usage: {...disjoint buckets} }
+     */
+    async invoke(messages, options = {}) {
+        const params = {
+            model: this.model,
+            messages: this.#formatMessages(messages),
+        };
+        // Pass through provider-specific params from model config (temperature, top_p, …).
+        Object.assign(params, this.#extractProviderParams());
+        if (options.tools && options.tools.length > 0) {
+            params.tools = options.tools.map(t => this.#formatTool(t));
+            // Explicit parallel tool calls — OpenAI-compatible default; set so it
+            // can't silently regress and matches parallel behavior across providers.
+            params.parallel_tool_calls = true;
+        }
+        if (options.tool_choice && options.tool_choice !== 'auto') {
+            params.tool_choice = this.#formatToolChoice(options.tool_choice);
+        }
+        this.log(`[OpenAICompatibleExecutor:${this.providerName}] Invoking:`, {
+            model: params.model,
+            temperature: params.temperature,
+            top_p: params.top_p,
+            tools: params.tools?.length || 0,
+        });
+        // STREAMED: consumeOpenAIStream reassembles content + tool_call arg deltas
+        // back into the exact non-streamed completion shape, so the extraction below
+        // is unchanged. Each chunk bumps the idle timer; a transient failure retries
+        // the whole prompt. stream_options.include_usage rides usage on the final
+        // chunk (guarded below — not all OpenAI-compatible hosts honor it).
+        const response = await streamWithRetry(async (guard) => {
+            const stream = await this.client.chat.completions.create({ ...params, stream: true, stream_options: { include_usage: true } }, { signal: guard.signal });
+            return await consumeOpenAIStream(stream, () => guard.bump());
+        }, {
+            externalSignal: options.signal,
+            isRetryable: (err) => this.isRetryableError(err),
+            log: (m) => this.log(m),
+        });
+        const choice = response.choices[0];
+        const message = choice.message;
+        // Automatic-cache providers report cached prefix tokens as a SUBSET of
+        // prompt_tokens (same convention as OpenAI direct). We subtract them out so
+        // input_tokens is the UNCACHED count and the four usage buckets stay
+        // disjoint across providers — otherwise cached reads get billed twice (once
+        // as input, once as read). cache_creation is 0: these providers have no
+        // billed write concept.
+        //
+        // Field location is NOT consistent across OpenAI-compatible hosts. Read all
+        // known spellings so a cache hit is never silently missed (a missed field
+        // reads as 0% hit rate → ~5x cost on our cache-heavy workload, and fails
+        // silently — the spec's highest-risk case):
+        //   - OpenAI direct / Together dedicated-inference: nested
+        //     prompt_tokens_details.cached_tokens
+        //   - Together OpenAI-compat surface: top-level cached_tokens
+        //   - DeepSeek: prompt_cache_hit_tokens (with prompt_cache_miss_tokens)
+        // All three are a SUBSET of prompt_tokens, so subtracting keeps input
+        // uncached. NOTE: prompt_tokens-is-cache-inclusive must be confirmed with a
+        // live Together call; if a host reports cached ADDITIVELY, this subtraction
+        // under-counts input and needs a per-host flag.
+        const usageTyped = response.usage;
+        const cachedTokens = usageTyped?.prompt_tokens_details?.cached_tokens ??
+            usageTyped?.cached_tokens ??
+            usageTyped?.prompt_cache_hit_tokens ??
+            0;
+        // Streamed usage rides the final include_usage chunk. Real OpenAI/Together
+        // send it, but guard so a host that omits it yields 0s instead of a crash
+        // on response.usage!.  If a host silently drops usage, cost → 0 and the
+        // credit engine under-bills — LIVE-VERIFY Together honors include_usage.
+        const promptTokens = response.usage?.prompt_tokens ?? 0;
+        const completionTokens = response.usage?.completion_tokens ?? 0;
+        return {
+            message: {
+                role: message.role,
+                content: message.content || '',
+                tool_calls: this.#extractToolCalls(message.tool_calls),
+            },
+            usage: {
+                input_tokens: Math.max(0, promptTokens - cachedTokens),
+                output_tokens: completionTokens,
+                cache_read_input_tokens: cachedTokens,
+                cache_creation_input_tokens: 0,
+            },
+        };
+    }
+    hasToolCalls(message) {
+        return Boolean(message?.tool_calls && message.tool_calls.length > 0);
+    }
+    /** Format messages for the OpenAI wire format. */
+    #formatMessages(messages) {
+        return messages.map(msg => {
+            if (msg.role === 'tool') {
+                return { role: 'tool', tool_call_id: msg.tool_call_id, content: msg.content };
+            }
+            return { role: msg.role, content: msg.content };
+        });
+    }
+    /** Format a tool definition to OpenAI format from any of our known shapes. */
+    #formatTool(tool) {
+        if (tool.type === 'function' && tool.function)
+            return tool;
+        if (tool.function)
+            return { type: 'function', function: tool.function };
+        if (tool.input_schema) {
+            return {
+                type: 'function',
+                function: { name: tool.name, description: tool.description || '', parameters: tool.input_schema },
+            };
+        }
+        if (tool.parameters) {
+            return {
+                type: 'function',
+                function: { name: tool.name, description: tool.description || '', parameters: tool.parameters },
+            };
+        }
+        this.log(`[OpenAICompatibleExecutor:${this.providerName}] Warning: Unrecognized tool format:`, JSON.stringify(tool));
+        return {
+            type: 'function',
+            function: {
+                name: tool.name || 'unknown',
+                description: tool.description || '',
+                parameters: tool.parameters || tool.input_schema || { type: 'object', properties: {} },
+            },
+        };
+    }
+    /** Format tool_choice to OpenAI format. */
+    #formatToolChoice(toolChoice) {
+        if (typeof toolChoice === 'string') {
+            if (toolChoice === 'required' || toolChoice === 'any')
+                return 'required';
+            return 'auto';
+        }
+        if (toolChoice.type === 'function' || toolChoice.function?.name)
+            return toolChoice;
+        // Anthropic format: { type: "tool", name: "..." }
+        if (toolChoice.type === 'tool' && toolChoice.name) {
+            return { type: 'function', function: { name: toolChoice.name } };
+        }
+        return 'auto';
+    }
+    /** Extract tool calls from an OpenAI-compatible response. */
+    #extractToolCalls(toolCalls) {
+        if (!toolCalls || toolCalls.length === 0)
+            return [];
+        return toolCalls.map(tc => ({
+            id: tc.id,
+            name: tc.function.name,
+            args: JSON.parse(tc.function.arguments),
+        }));
+    }
+    /** Provider-specific params from model config metadata (excludes non-API keys). */
+    #extractProviderParams() {
+        const metadata = this.primaryModelConfig.metadata || {};
+        // Strip keys that are adapter config, not model-call params.
+        const { displayName, baseURL, cacheMode, quantization, ...providerParams } = metadata;
+        return providerParams;
+    }
+}
+//# sourceMappingURL=openaiCompatible.js.map

package/dist/providers/openaiCompatible.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"openaiCompatible.js","sourceRoot":"","sources":["../../src/providers/openaiCompatible.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;GA0BG;AAEH,OAAO,MAAM,MAAM,QAAQ,CAAC;AAC5B,OAAO,YAAY,MAAM,oBAAoB,CAAC;AAE9C,OAAO,EAAE,eAAe,EAAE,mBAAmB,EAAE,2BAA2B,EAAE,MAAM,gBAAgB,CAAC;AAEnG;2EAC2E;AAC3E,MAAM,CAAC,MAAM,2BAA2B,GAA2B;IACjE,QAAQ,EAAG,4BAA4B;IACvC,SAAS,EAAE,uCAAuC;IAClD,SAAS,EAAE,qCAAqC;IAChD,QAAQ,EAAG,6BAA6B;CACzC,CAAC;AAEF,gFAAgF;AAChF,MAAM,CAAC,MAAM,2BAA2B,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,2BAA2B,CAAC,CAAC,CAAC;AAE7F,MAAM,CAAC,OAAO,OAAO,wBAAyB,SAAQ,YAAY;IACxD,MAAM,CAAS;IACf,YAAY,CAAS;IAE7B,YAAY,MAA0B;QACpC,KAAK,CAAC,MAAM,CAAC,CAAC;QAEd,uEAAuE;QACvE,IAAI,CAAC,YAAY,GAAG,CAAC,IAAI,CAAC,QAAQ,IAAI,EAAE,CAAC,CAAC,WAAW,EAAE,CAAC;QAExD,oEAAoE;QACpE,4EAA4E;QAC5E,MAAM,OAAO,GAAG,IAAI,CAAC,WAAkC,CAAC;QACxD,MAAM,KAAK,GAAG,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,OAAO,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;QACnE,IAAI,CAAC,KAAK,EAAE,CAAC;YACX,MAAM,IAAI,KAAK,CAAC,0CAA0C,IAAI,CAAC,YAAY,cAAc,CAAC,CAAC;QAC7F,CAAC;QACD,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,CAAC;YAClB,MAAM,IAAI,KAAK,CAAC,0CAA0C,IAAI,CAAC,YAAY,qBAAqB,CAAC,CAAC;QACpG,CAAC;QAED,wEAAwE;QACxE,MAAM,QAAQ,GAAG,CAAE,IAAI,CAAC,kBAA0B,CAAC,QAAQ,IAAI,EAAE,CAAwB,CAAC;QAC1F,MAAM,OAAO,GACX,KAAK,CAAC,OAAO;YACb,QAAQ,CAAC,OAAO;YAChB,2BAA2B,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;QACjD,IAAI,CAAC,OAAO,EAAE,CAAC;YACb,MAAM,IAAI,KAAK,CACb,uDAAuD,IAAI,CAAC,YAAY,MAAM;gBAC9E,uDAAuD,IAAI,CAAC,YAAY,qCAAqC,CAC9G,CAAC;QACJ,CAAC;QAED,uEAAuE;QACvE,qEAAqE;QACrE,2EAA2E;QAC3E,4EAA4E;QAC5E,2EAA2E;QAC3E,wEAAwE;QACxE,2BAA2B;QAC3B,IAAI,CAAC,MAAM,GAAG,IAAI,MAAM,CAAC;YACvB,MAAM,EAAE,KAAK,CAAC,MAAM;YACpB,OAAO;YACP,UAAU,EAAE,CAAC;YACb,OAAO,EAAE,2BAA2B;YACpC,uBAAuB,EAAE,KAAK,CAAC,uBAAuB;SACvD,CAAC,CAAC;QAEH,IAAI,CAAC,GAAG,CAAC,0CAA0C,IAAI,CAAC,YAAY,MAAM,OAAO,gBAAgB,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC;IACjH,CAAC;IAED;;;OAGG;IACH,KAAK,CAAC,MAAM,CAAC,QAAmB,EAAE,UAAyB,EAAE;QAC3D,MAAM,MAAM,GAAQ;YAClB,KAAK,EAAE,IAAI,CAAC,KAAK;YACjB,QAAQ,EAAE,IAAI,CAAC,eAAe,CAAC,QAAQ,CAAC;SACzC,CAAC;QAEF,mFAAmF;QACnF,MAAM,CAAC,MAAM,CAAC,MAAM,EAAE,IAAI,CAAC,sBAAsB,EAAE,CAAC,CAAC;QAErD,IAAI,OAAO,CAAC,KAAK,IAAI,OAAO,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC9C,MAAM,CAAC,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,CAAC;YAC3D,sEAAsE;YACtE,yEAAyE;YACzE,MAAM,CAAC,mBAAmB,GAAG,IAAI,CAAC;QACpC,CAAC;QAED,IAAI,OAAO,CAAC,WAAW,IAAI,OAAO,CAAC,WAAW,KAAK,MAAM,EAAE,CAAC;YAC1D,MAAM,CAAC,WAAW,GAAG,IAAI,CAAC,iBAAiB,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC;QACnE,CAAC;QAED,IAAI,CAAC,GAAG,CAAC,6BAA6B,IAAI,CAAC,YAAY,aAAa,EAAE;YACpE,KAAK,EAAE,MAAM,CAAC,KAAK;YACnB,WAAW,EAAE,MAAM,CAAC,WAAW;YAC/B,KAAK,EAAE,MAAM,CAAC,KAAK;YACnB,KAAK,EAAE,MAAM,CAAC,KAAK,EAAE,MAAM,IAAI,CAAC;SACjC,CAAC,CAAC;QAEH,2EAA2E;QAC3E,6EAA6E;QAC7E,6EAA6E;QAC7E,0EAA0E;QAC1E,oEAAoE;QACpE,MAAM,QAAQ,GAAG,MAAM,eAAe,CACpC,KAAK,EAAE,KAAK,EAAE,EAAE;YACd,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,MAAM,CACtD,EAAE,GAAG,MAAM,EAAE,MAAM,EAAE,IAAI,EAAE,cAAc,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,EAAE,EACpE,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,CACzB,CAAC;YACF,OAAO,MAAM,mBAAmB,CAAC,MAAa,EAAE,GAAG,EAAE,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC;QACtE,CAAC,EACD;YACE,cAAc,EAAE,OAAO,CAAC,MAAM;YAC9B,WAAW,EAAE,CAAC,GAAG,EAAE,EAAE,CAAC,IAAI,CAAC,gBAAgB,CAAC,GAAG,CAAC;YAChD,GAAG,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;SACxB,CACF,CAAC;QAEF,MAAM,MAAM,GAAG,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QACnC,MAAM,OAAO,GAAG,MAAM,CAAC,OAAO,CAAC;QAE/B,uEAAuE;QACvE,4EAA4E;QAC5E,qEAAqE;QACrE,4EAA4E;QAC5E,wEAAwE;QACxE,wBAAwB;QACxB,EAAE;QACF,4EAA4E;QAC5E,0EAA0E;QAC1E,yEAAyE;QACzE,4CAA4C;QAC5C,2DAA2D;QAC3D,0CAA0C;QAC1C,8DAA8D;QAC9D,wEAAwE;QACxE,sEAAsE;QACtE,4EAA4E;QAC5E,4EAA4E;QAC5E,gDAAgD;QAChD,MAAM,UAAU,GAAG,QAAQ,CAAC,KAI3B,CAAC;QACF,MAAM,YAAY,GAChB,UAAU,EAAE,qBAAqB,EAAE,aAAa;YAChD,UAAU,EAAE,aAAa;YACzB,UAAU,EAAE,uBAAuB;YACnC,CAAC,CAAC;QACJ,2EAA2E;QAC3E,0EAA0E;QAC1E,wEAAwE;QACxE,yEAAyE;QACzE,MAAM,YAAY,GAAG,QAAQ,CAAC,KAAK,EAAE,aAAa,IAAI,CAAC,CAAC;QACxD,MAAM,gBAAgB,GAAG,QAAQ,CAAC,KAAK,EAAE,iBAAiB,IAAI,CAAC,CAAC;QAEhE,OAAO;YACL,OAAO,EAAE;gBACP,IAAI,EAAE,OAAO,CAAC,IAAuB;gBACrC,OAAO,EAAE,OAAO,CAAC,OAAO,IAAI,EAAE;gBAC9B,UAAU,EAAE,IAAI,CAAC,iBAAiB,CAAC,OAAO,CAAC,UAAU,CAAC;aACvD;YACD,KAAK,EAAE;gBACL,YAAY,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,YAAY,GAAG,YAAY,CAAC;gBACtD,aAAa,EAAE,gBAAgB;gBAC/B,uBAAuB,EAAE,YAAY;gBACrC,2BAA2B,EAAE,CAAC;aAC/B;SACF,CAAC;IACJ,CAAC;IAED,YAAY,CAAC,OAAgB;QAC3B,OAAO,OAAO,CAAC,OAAO,EAAE,UAAU,IAAI,OAAO,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;IACvE,CAAC;IAED,kDAAkD;IAClD,eAAe,CAAC,QAAmB;QACjC,OAAO,QAAQ,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE;YACxB,IAAI,GAAG,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;gBACxB,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,YAAY,EAAE,GAAG,CAAC,YAAY,EAAE,OAAO,EAAE,GAAG,CAAC,OAAO,EAAE,CAAC;YAChF,CAAC;YACD,OAAO,EAAE,IAAI,EAAE,GAAG,CAAC,IAAI,EAAE,OAAO,EAAE,GAAG,CAAC,OAAO,EAAE,CAAC;QAClD,CAAC,CAAC,CAAC;IACL,CAAC;IAED,8EAA8E;IAC9E,WAAW,CAAC,IAAS;QACnB,IAAI,IAAI,CAAC,IAAI,KAAK,UAAU,IAAI,IAAI,CAAC,QAAQ;YAAE,OAAO,IAAI,CAAC;QAC3D,IAAI,IAAI,CAAC,QAAQ;YAAE,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,QAAQ,EAAE,IAAI,CAAC,QAAQ,EAAE,CAAC;QACxE,IAAI,IAAI,CAAC,YAAY,EAAE,CAAC;YACtB,OAAO;gBACL,IAAI,EAAE,UAAU;gBAChB,QAAQ,EAAE,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE,WAAW,EAAE,IAAI,CAAC,WAAW,IAAI,EAAE,EAAE,UAAU,EAAE,IAAI,CAAC,YAAY,EAAE;aAClG,CAAC;QACJ,CAAC;QACD,IAAI,IAAI,CAAC,UAAU,EAAE,CAAC;YACpB,OAAO;gBACL,IAAI,EAAE,UAAU;gBAChB,QAAQ,EAAE,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE,WAAW,EAAE,IAAI,CAAC,WAAW,IAAI,EAAE,EAAE,UAAU,EAAE,IAAI,CAAC,UAAU,EAAE;aAChG,CAAC;QACJ,CAAC;QACD,IAAI,CAAC,GAAG,CAAC,6BAA6B,IAAI,CAAC,YAAY,sCAAsC,EAAE,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC;QACrH,OAAO;YACL,IAAI,EAAE,UAAU;YAChB,QAAQ,EAAE;gBACR,IAAI,EAAE,IAAI,CAAC,IAAI,IAAI,SAAS;gBAC5B,WAAW,EAAE,IAAI,CAAC,WAAW,IAAI,EAAE;gBACnC,UAAU,EAAE,IAAI,CAAC,UAAU,IAAI,IAAI,CAAC,YAAY,IAAI,EAAE,IAAI,EAAE,QAAQ,EAAE,UAAU,EAAE,EAAE,EAAE;aACvF;SACF,CAAC;IACJ,CAAC;IAED,2CAA2C;IAC3C,iBAAiB,CAAC,UAAe;QAC/B,IAAI,OAAO,UAAU,KAAK,QAAQ,EAAE,CAAC;YACnC,IAAI,UAAU,KAAK,UAAU,IAAI,UAAU,KAAK,KAAK;gBAAE,OAAO,UAAU,CAAC;YACzE,OAAO,MAAM,CAAC;QAChB,CAAC;QACD,IAAI,UAAU,CAAC,IAAI,KAAK,UAAU,IAAI,UAAU,CAAC,QAAQ,EAAE,IAAI;YAAE,OAAO,UAAU,CAAC;QACnF,kDAAkD;QAClD,IAAI,UAAU,CAAC,IAAI,KAAK,MAAM,IAAI,UAAU,CAAC,IAAI,EAAE,CAAC;YAClD,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,QAAQ,EAAE,EAAE,IAAI,EAAE,UAAU,CAAC,IAAI,EAAE,EAAE,CAAC;QACnE,CAAC;QACD,OAAO,MAAM,CAAC;IAChB,CAAC;IAED,6DAA6D;IAC7D,iBAAiB,CAAC,SAA4B;QAC5C,IAAI,CAAC,SAAS,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO,EAAE,CAAC;QACpD,OAAO,SAAS,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;YAC1B,EAAE,EAAE,EAAE,CAAC,EAAE;YACT,IAAI,EAAE,EAAE,CAAC,QAAQ,CAAC,IAAI;YACtB,IAAI,EAAE,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,QAAQ,CAAC,SAAS,CAAC;SACxC,CAAC,CAAC,CAAC;IACN,CAAC;IAED,mFAAmF;IACnF,sBAAsB;QACpB,MAAM,QAAQ,GAAI,IAAI,CAAC,kBAA0B,CAAC,QAAQ,IAAI,EAAE,CAAC;QACjE,6DAA6D;QAC7D,MAAM,EAAE,WAAW,EAAE,OAAO,EAAE,SAAS,EAAE,YAAY,EAAE,GAAG,cAAc,EAAE,GAAG,QAAQ,CAAC;QACtF,OAAO,cAAc,CAAC;IACxB,CAAC;CACF"}

package/dist/providers/streaming.d.ts

@@ -0,0 +1,109 @@
+/**
+ * streaming.ts — shared streaming orchestration for the provider adapters.
+ *
+ * WHY: model responses used to be awaited as a single non-streamed completion,
+ * so the SDK's per-attempt `timeout` measured *total* time-to-complete. A
+ * legitimately long-but-progressing response (a full slide deck, a long report)
+ * could exceed that ceiling, get retried — re-sending the whole prompt and
+ * re-billing each attempt — and eventually fail as "Request timed out".
+ *
+ * Streaming removes the ceiling: we consume the provider's token stream and
+ * reset an INTER-CHUNK IDLE timer on every event. A response that keeps making
+ * progress never times out no matter how long it runs; only a genuinely STALLED
+ * stream (no token for `idleTimeoutMs`) is aborted and retried. A generous
+ * ABSOLUTE BACKSTOP guards the pathological "one token every 59 s forever" case.
+ *
+ * This module owns the timeout/abort/retry *orchestration* (provider-agnostic).
+ * Each provider adapter owns its SDK-specific stream consumption and assembles
+ * the SAME final response object its non-streamed path produced — so the
+ * executor's consumption (and token accounting) is byte-for-byte unchanged.
+ */
+/** No token for this long → the stream is stalled; abort + retry the attempt. */
+export declare const STREAM_IDLE_TIMEOUT_MS: number;
+/** Absolute per-attempt ceiling. Only the pathological dribble case reaches it;
+ *  a steadily-streaming multi-minute response never does. It is TERMINAL (not
+ *  retried) so a misbehaving stream can't be retried 3× and blow the worker's
+ *  14-min checkpoint. Comfortably above any legitimate single-turn duration. */
+export declare const STREAM_ABSOLUTE_BACKSTOP_MS: number;
+/** Whole-attempt retries on a transient mid-stream failure (idle stall, network
+ *  drop, 5xx after the stream opened). Matches the pre-existing retry budget. */
+export declare const STREAM_MAX_RETRIES: number;
+export type StreamAbortReason = 'idle' | 'backstop' | 'external';
+/** Raised when the guard aborts a stream. `reason` decides retry policy:
+ *  `idle` is transient (retry); `backstop` and `external` are terminal. */
+export declare class StreamAbortError extends Error {
+    reason: StreamAbortReason;
+    constructor(reason: StreamAbortReason, message: string);
+}
+export interface StreamGuard {
+    /** Pass to the provider SDK's stream call; aborts on idle/backstop/external. */
+    readonly signal: AbortSignal;
+    /** Call on every stream event/chunk to reset the idle timer. */
+    bump(): void;
+    /** Why the guard aborted, or null if it hasn't. */
+    reason(): StreamAbortReason | null;
+    /** Clear all timers + listeners. Idempotent; always call in a finally. */
+    dispose(): void;
+}
+export interface StreamGuardOptions {
+    idleTimeoutMs?: number;
+    backstopMs?: number;
+    /** The caller's AbortSignal (backend stop/timeout). Aborts the stream with
+     *  reason `external` — terminal, never retried. */
+    externalSignal?: AbortSignal;
+}
+/**
+ * Build an abort guard around a single stream attempt: an idle timer reset by
+ * `bump()`, an absolute backstop, and linkage to an optional external signal.
+ */
+export declare function createStreamGuard(opts?: StreamGuardOptions): StreamGuard;
+/** True for the abort error a provider SDK throws when we call `controller.abort()`. */
+export declare function isAbortError(err: any): boolean;
+export interface StreamRetryOptions {
+    maxRetries?: number;
+    idleTimeoutMs?: number;
+    backstopMs?: number;
+    externalSignal?: AbortSignal;
+    /** Classify a NON-abort mid-stream error as transient (retry) vs fatal. */
+    isRetryable?: (err: any) => boolean;
+    log?: (message: string) => void;
+}
+/**
+ * Run one stream `attempt` with a fresh guard, retrying the WHOLE prompt on a
+ * transient failure (idle stall or a caller-classified retryable error). The
+ * partial output is discarded on failure. Backstop and external aborts are
+ * terminal. Bounded by `maxRetries`.
+ */
+export declare function streamWithRetry<T>(attempt: (guard: StreamGuard) => Promise<T>, opts?: StreamRetryOptions): Promise<T>;
+/** The subset of an OpenAI chat completion the executor consumes. Assembling
+ *  the stream back into this shape lets the non-streamed extraction run
+ *  unchanged. */
+export interface OpenAICompletionLike {
+    choices: Array<{
+        message: {
+            role: string;
+            content: string | null;
+            tool_calls?: Array<{
+                id: string;
+                type: 'function';
+                function: {
+                    name: string;
+                    arguments: string;
+                };
+            }>;
+        };
+    }>;
+    usage?: any;
+}
+/**
+ * Fold an OpenAI (or OpenAI-compatible) chat completion STREAM back into the
+ * non-streamed completion shape.
+ *
+ * Deltas arrive fragmented: `content` in pieces, and each tool call as an
+ * index-keyed run of deltas — the id + function.name usually in the first, the
+ * JSON `arguments` string across many. We reassemble per index and join in
+ * index order. Usage arrives on the final chunk (requires
+ * `stream_options.include_usage: true`).
+ */
+export declare function consumeOpenAIStream(stream: AsyncIterable<any>, bump: () => void): Promise<OpenAICompletionLike>;
+//# sourceMappingURL=streaming.d.ts.map

package/dist/providers/streaming.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"streaming.d.ts","sourceRoot":"","sources":["../../src/providers/streaming.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;GAmBG;AAEH,iFAAiF;AACjF,eAAO,MAAM,sBAAsB,QACwB,CAAC;AAE5D;;;gFAGgF;AAChF,eAAO,MAAM,2BAA2B,QACgB,CAAC;AAEzD;iFACiF;AACjF,eAAO,MAAM,kBAAkB,QAAmD,CAAC;AAEnF,MAAM,MAAM,iBAAiB,GAAG,MAAM,GAAG,UAAU,GAAG,UAAU,CAAC;AAEjE;2EAC2E;AAC3E,qBAAa,gBAAiB,SAAQ,KAAK;IACzC,MAAM,EAAE,iBAAiB,CAAC;gBACd,MAAM,EAAE,iBAAiB,EAAE,OAAO,EAAE,MAAM;CAKvD;AAED,MAAM,WAAW,WAAW;IAC1B,gFAAgF;IAChF,QAAQ,CAAC,MAAM,EAAE,WAAW,CAAC;IAC7B,gEAAgE;IAChE,IAAI,IAAI,IAAI,CAAC;IACb,mDAAmD;IACnD,MAAM,IAAI,iBAAiB,GAAG,IAAI,CAAC;IACnC,0EAA0E;IAC1E,OAAO,IAAI,IAAI,CAAC;CACjB;AAED,MAAM,WAAW,kBAAkB;IACjC,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB;uDACmD;IACnD,cAAc,CAAC,EAAE,WAAW,CAAC;CAC9B;AAOD;;;GAGG;AACH,wBAAgB,iBAAiB,CAAC,IAAI,GAAE,kBAAuB,GAAG,WAAW,CAqD5E;AAED,wFAAwF;AACxF,wBAAgB,YAAY,CAAC,GAAG,EAAE,GAAG,GAAG,OAAO,CAI9C;AAED,MAAM,WAAW,kBAAkB;IACjC,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,cAAc,CAAC,EAAE,WAAW,CAAC;IAC7B,2EAA2E;IAC3E,WAAW,CAAC,EAAE,CAAC,GAAG,EAAE,GAAG,KAAK,OAAO,CAAC;IACpC,GAAG,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,IAAI,CAAC;CACjC;AAED;;;;;GAKG;AACH,wBAAsB,eAAe,CAAC,CAAC,EACrC,OAAO,EAAE,CAAC,KAAK,EAAE,WAAW,KAAK,OAAO,CAAC,CAAC,CAAC,EAC3C,IAAI,GAAE,kBAAuB,GAC5B,OAAO,CAAC,CAAC,CAAC,CAyCZ;AAMD;;iBAEiB;AACjB,MAAM,WAAW,oBAAoB;IACnC,OAAO,EAAE,KAAK,CAAC;QACb,OAAO,EAAE;YACP,IAAI,EAAE,MAAM,CAAC;YACb,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;YACvB,UAAU,CAAC,EAAE,KAAK,CAAC;gBAAE,EAAE,EAAE,MAAM,CAAC;gBAAC,IAAI,EAAE,UAAU,CAAC;gBAAC,QAAQ,EAAE;oBAAE,IAAI,EAAE,MAAM,CAAC;oBAAC,SAAS,EAAE,MAAM,CAAA;iBAAE,CAAA;aAAE,CAAC,CAAC;SACrG,CAAC;KACH,CAAC,CAAC;IACH,KAAK,CAAC,EAAE,GAAG,CAAC;CACb;AAED;;;;;;;;;GASG;AACH,wBAAsB,mBAAmB,CACvC,MAAM,EAAE,aAAa,CAAC,GAAG,CAAC,EAC1B,IAAI,EAAE,MAAM,IAAI,GACf,OAAO,CAAC,oBAAoB,CAAC,CAgD/B"}

package/dist/providers/streaming.js

@@ -0,0 +1,226 @@
+/**
+ * streaming.ts — shared streaming orchestration for the provider adapters.
+ *
+ * WHY: model responses used to be awaited as a single non-streamed completion,
+ * so the SDK's per-attempt `timeout` measured *total* time-to-complete. A
+ * legitimately long-but-progressing response (a full slide deck, a long report)
+ * could exceed that ceiling, get retried — re-sending the whole prompt and
+ * re-billing each attempt — and eventually fail as "Request timed out".
+ *
+ * Streaming removes the ceiling: we consume the provider's token stream and
+ * reset an INTER-CHUNK IDLE timer on every event. A response that keeps making
+ * progress never times out no matter how long it runs; only a genuinely STALLED
+ * stream (no token for `idleTimeoutMs`) is aborted and retried. A generous
+ * ABSOLUTE BACKSTOP guards the pathological "one token every 59 s forever" case.
+ *
+ * This module owns the timeout/abort/retry *orchestration* (provider-agnostic).
+ * Each provider adapter owns its SDK-specific stream consumption and assembles
+ * the SAME final response object its non-streamed path produced — so the
+ * executor's consumption (and token accounting) is byte-for-byte unchanged.
+ */
+/** No token for this long → the stream is stalled; abort + retry the attempt. */
+export const STREAM_IDLE_TIMEOUT_MS = Number(process.env.AGNT_STREAM_IDLE_TIMEOUT_MS) || 60_000;
+/** Absolute per-attempt ceiling. Only the pathological dribble case reaches it;
+ *  a steadily-streaming multi-minute response never does. It is TERMINAL (not
+ *  retried) so a misbehaving stream can't be retried 3× and blow the worker's
+ *  14-min checkpoint. Comfortably above any legitimate single-turn duration. */
+export const STREAM_ABSOLUTE_BACKSTOP_MS = Number(process.env.AGNT_STREAM_BACKSTOP_MS) || 600_000;
+/** Whole-attempt retries on a transient mid-stream failure (idle stall, network
+ *  drop, 5xx after the stream opened). Matches the pre-existing retry budget. */
+export const STREAM_MAX_RETRIES = Number(process.env.AGNT_STREAM_MAX_RETRIES) || 3;
+/** Raised when the guard aborts a stream. `reason` decides retry policy:
+ *  `idle` is transient (retry); `backstop` and `external` are terminal. */
+export class StreamAbortError extends Error {
+    reason;
+    constructor(reason, message) {
+        super(message);
+        this.name = 'StreamAbortError';
+        this.reason = reason;
+    }
+}
+function unref(t) {
+    // Don't let the idle/backstop timer keep the event loop (or a Lambda) alive.
+    if (t && typeof t.unref === 'function')
+        t.unref();
+}
+/**
+ * Build an abort guard around a single stream attempt: an idle timer reset by
+ * `bump()`, an absolute backstop, and linkage to an optional external signal.
+ */
+export function createStreamGuard(opts = {}) {
+    const idleMs = opts.idleTimeoutMs ?? STREAM_IDLE_TIMEOUT_MS;
+    const backstopMs = opts.backstopMs ?? STREAM_ABSOLUTE_BACKSTOP_MS;
+    const controller = new AbortController();
+    let abortedReason = null;
+    let idleTimer = null;
+    let backstopTimer = null;
+    let externalCleanup = null;
+    const abort = (reason) => {
+        if (abortedReason)
+            return; // first reason wins
+        abortedReason = reason;
+        clearTimers();
+        controller.abort();
+    };
+    const clearTimers = () => {
+        if (idleTimer) {
+            clearTimeout(idleTimer);
+            idleTimer = null;
+        }
+        if (backstopTimer) {
+            clearTimeout(backstopTimer);
+            backstopTimer = null;
+        }
+    };
+    const bump = () => {
+        if (abortedReason)
+            return;
+        if (idleTimer)
+            clearTimeout(idleTimer);
+        idleTimer = setTimeout(() => abort('idle'), idleMs);
+        unref(idleTimer);
+    };
+    backstopTimer = setTimeout(() => abort('backstop'), backstopMs);
+    unref(backstopTimer);
+    if (opts.externalSignal) {
+        const ext = opts.externalSignal;
+        if (ext.aborted) {
+            abort('external');
+        }
+        else {
+            const onAbort = () => abort('external');
+            ext.addEventListener('abort', onAbort, { once: true });
+            externalCleanup = () => ext.removeEventListener('abort', onAbort);
+        }
+    }
+    bump(); // arm the idle timer for the connection-open window
+    return {
+        signal: controller.signal,
+        bump,
+        reason: () => abortedReason,
+        dispose: () => {
+            clearTimers();
+            if (externalCleanup) {
+                externalCleanup();
+                externalCleanup = null;
+            }
+        },
+    };
+}
+/** True for the abort error a provider SDK throws when we call `controller.abort()`. */
+export function isAbortError(err) {
+    const name = err?.name ?? err?.constructor?.name;
+    if (name === 'AbortError' || name === 'APIUserAbortError')
+        return true;
+    return typeof err?.message === 'string' && /\baborted\b/i.test(err.message);
+}
+/**
+ * Run one stream `attempt` with a fresh guard, retrying the WHOLE prompt on a
+ * transient failure (idle stall or a caller-classified retryable error). The
+ * partial output is discarded on failure. Backstop and external aborts are
+ * terminal. Bounded by `maxRetries`.
+ */
+export async function streamWithRetry(attempt, opts = {}) {
+    const maxRetries = opts.maxRetries ?? STREAM_MAX_RETRIES;
+    const isRetryable = opts.isRetryable ?? (() => true);
+    let lastError;
+    for (let i = 0; i < maxRetries; i++) {
+        if (opts.externalSignal?.aborted) {
+            throw new StreamAbortError('external', '[streaming] aborted before attempt');
+        }
+        const guard = createStreamGuard({
+            idleTimeoutMs: opts.idleTimeoutMs,
+            backstopMs: opts.backstopMs,
+            externalSignal: opts.externalSignal,
+        });
+        try {
+            return await attempt(guard);
+        }
+        catch (err) {
+            // Map a guard-triggered abort to a typed error carrying its reason.
+            const reason = guard.reason();
+            const mapped = reason && isAbortError(err)
+                ? new StreamAbortError(reason, `[streaming] stream aborted: ${reason}`)
+                : err;
+            lastError = mapped;
+            // Terminal: the caller asked to stop, or the stream dribbled to the
+            // backstop — retrying either would be wrong/wasteful.
+            if (mapped instanceof StreamAbortError && mapped.reason !== 'idle') {
+                throw mapped;
+            }
+            const retryable = mapped instanceof StreamAbortError ? mapped.reason === 'idle' : isRetryable(mapped);
+            if (!retryable || i === maxRetries - 1)
+                throw mapped;
+            opts.log?.(`[streaming] attempt ${i + 1}/${maxRetries} failed (${mapped?.message ?? mapped}) — discarding partial, retrying`);
+        }
+        finally {
+            guard.dispose();
+        }
+    }
+    throw lastError;
+}
+/**
+ * Fold an OpenAI (or OpenAI-compatible) chat completion STREAM back into the
+ * non-streamed completion shape.
+ *
+ * Deltas arrive fragmented: `content` in pieces, and each tool call as an
+ * index-keyed run of deltas — the id + function.name usually in the first, the
+ * JSON `arguments` string across many. We reassemble per index and join in
+ * index order. Usage arrives on the final chunk (requires
+ * `stream_options.include_usage: true`).
+ */
+export async function consumeOpenAIStream(stream, bump) {
+    let role = 'assistant';
+    let content = '';
+    let sawContent = false;
+    let usage = undefined;
+    const byIndex = new Map();
+    for await (const chunk of stream) {
+        bump();
+        const choice = chunk?.choices?.[0];
+        if (choice) {
+            const delta = choice.delta || {};
+            if (delta.role)
+                role = delta.role;
+            if (typeof delta.content === 'string') {
+                content += delta.content;
+                sawContent = true;
+            }
+            if (Array.isArray(delta.tool_calls)) {
+                for (const tc of delta.tool_calls) {
+                    const idx = typeof tc.index === 'number' ? tc.index : 0;
+                    let acc = byIndex.get(idx);
+                    if (!acc) {
+                        acc = { id: '', name: '', args: '' };
+                        byIndex.set(idx, acc);
+                    }
+                    if (tc.id)
+                        acc.id = tc.id;
+                    if (tc.function?.name)
+                        acc.name += tc.function.name;
+                    if (typeof tc.function?.arguments === 'string')
+                        acc.args += tc.function.arguments;
+                }
+            }
+        }
+        // Usage rides its own final chunk (choices may be empty there).
+        if (chunk?.usage)
+            usage = chunk.usage;
+    }
+    const tool_calls = byIndex.size
+        ? [...byIndex.entries()]
+            .sort((a, b) => a[0] - b[0])
+            // Default empty args to "{}" — a zero-arg tool call may stream an id+name
+            // with no `arguments` fragment (real OpenAI always sends "{}", but
+            // DeepSeek / OpenAI-compatible endpoints may not). The downstream
+            // extractor does JSON.parse(arguments), which throws on "". Mirror the
+            // non-streamed shape so a no-arg tool call parses to {} instead of crashing.
+            .map(([, v]) => ({ id: v.id, type: 'function', function: { name: v.name, arguments: v.args || '{}' } }))
+        : undefined;
+    return {
+        // Mirror the non-streamed shape: content is null when the turn is tool-only.
+        choices: [{ message: { role, content: sawContent ? content : (tool_calls ? null : ''), tool_calls } }],
+        usage,
+    };
+}
+//# sourceMappingURL=streaming.js.map

package/dist/providers/streaming.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"streaming.js","sourceRoot":"","sources":["../../src/providers/streaming.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;GAmBG;AAEH,iFAAiF;AACjF,MAAM,CAAC,MAAM,sBAAsB,GACjC,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,2BAA2B,CAAC,IAAI,MAAM,CAAC;AAE5D;;;gFAGgF;AAChF,MAAM,CAAC,MAAM,2BAA2B,GACtC,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,uBAAuB,CAAC,IAAI,OAAO,CAAC;AAEzD;iFACiF;AACjF,MAAM,CAAC,MAAM,kBAAkB,GAAG,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,uBAAuB,CAAC,IAAI,CAAC,CAAC;AAInF;2EAC2E;AAC3E,MAAM,OAAO,gBAAiB,SAAQ,KAAK;IACzC,MAAM,CAAoB;IAC1B,YAAY,MAAyB,EAAE,OAAe;QACpD,KAAK,CAAC,OAAO,CAAC,CAAC;QACf,IAAI,CAAC,IAAI,GAAG,kBAAkB,CAAC;QAC/B,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;IACvB,CAAC;CACF;AAqBD,SAAS,KAAK,CAAC,CAAM;IACnB,6EAA6E;IAC7E,IAAI,CAAC,IAAI,OAAO,CAAC,CAAC,KAAK,KAAK,UAAU;QAAE,CAAC,CAAC,KAAK,EAAE,CAAC;AACpD,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,iBAAiB,CAAC,OAA2B,EAAE;IAC7D,MAAM,MAAM,GAAG,IAAI,CAAC,aAAa,IAAI,sBAAsB,CAAC;IAC5D,MAAM,UAAU,GAAG,IAAI,CAAC,UAAU,IAAI,2BAA2B,CAAC;IAClE,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;IACzC,IAAI,aAAa,GAA6B,IAAI,CAAC;IACnD,IAAI,SAAS,GAAQ,IAAI,CAAC;IAC1B,IAAI,aAAa,GAAQ,IAAI,CAAC;IAC9B,IAAI,eAAe,GAAwB,IAAI,CAAC;IAEhD,MAAM,KAAK,GAAG,CAAC,MAAyB,EAAQ,EAAE;QAChD,IAAI,aAAa;YAAE,OAAO,CAAC,oBAAoB;QAC/C,aAAa,GAAG,MAAM,CAAC;QACvB,WAAW,EAAE,CAAC;QACd,UAAU,CAAC,KAAK,EAAE,CAAC;IACrB,CAAC,CAAC;IAEF,MAAM,WAAW,GAAG,GAAS,EAAE;QAC7B,IAAI,SAAS,EAAE,CAAC;YAAC,YAAY,CAAC,SAAS,CAAC,CAAC;YAAC,SAAS,GAAG,IAAI,CAAC;QAAC,CAAC;QAC7D,IAAI,aAAa,EAAE,CAAC;YAAC,YAAY,CAAC,aAAa,CAAC,CAAC;YAAC,aAAa,GAAG,IAAI,CAAC;QAAC,CAAC;IAC3E,CAAC,CAAC;IAEF,MAAM,IAAI,GAAG,GAAS,EAAE;QACtB,IAAI,aAAa;YAAE,OAAO;QAC1B,IAAI,SAAS;YAAE,YAAY,CAAC,SAAS,CAAC,CAAC;QACvC,SAAS,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,KAAK,CAAC,MAAM,CAAC,EAAE,MAAM,CAAC,CAAC;QACpD,KAAK,CAAC,SAAS,CAAC,CAAC;IACnB,CAAC,CAAC;IAEF,aAAa,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,KAAK,CAAC,UAAU,CAAC,EAAE,UAAU,CAAC,CAAC;IAChE,KAAK,CAAC,aAAa,CAAC,CAAC;IAErB,IAAI,IAAI,CAAC,cAAc,EAAE,CAAC;QACxB,MAAM,GAAG,GAAG,IAAI,CAAC,cAAc,CAAC;QAChC,IAAI,GAAG,CAAC,OAAO,EAAE,CAAC;YAChB,KAAK,CAAC,UAAU,CAAC,CAAC;QACpB,CAAC;aAAM,CAAC;YACN,MAAM,OAAO,GAAG,GAAG,EAAE,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC;YACxC,GAAG,CAAC,gBAAgB,CAAC,OAAO,EAAE,OAAO,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;YACvD,eAAe,GAAG,GAAG,EAAE,CAAC,GAAG,CAAC,mBAAmB,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;QACpE,CAAC;IACH,CAAC;IAED,IAAI,EAAE,CAAC,CAAC,oDAAoD;IAE5D,OAAO;QACL,MAAM,EAAE,UAAU,CAAC,MAAM;QACzB,IAAI;QACJ,MAAM,EAAE,GAAG,EAAE,CAAC,aAAa;QAC3B,OAAO,EAAE,GAAG,EAAE;YACZ,WAAW,EAAE,CAAC;YACd,IAAI,eAAe,EAAE,CAAC;gBAAC,eAAe,EAAE,CAAC;gBAAC,eAAe,GAAG,IAAI,CAAC;YAAC,CAAC;QACrE,CAAC;KACF,CAAC;AACJ,CAAC;AAED,wFAAwF;AACxF,MAAM,UAAU,YAAY,CAAC,GAAQ;IACnC,MAAM,IAAI,GAAG,GAAG,EAAE,IAAI,IAAI,GAAG,EAAE,WAAW,EAAE,IAAI,CAAC;IACjD,IAAI,IAAI,KAAK,YAAY,IAAI,IAAI,KAAK,mBAAmB;QAAE,OAAO,IAAI,CAAC;IACvE,OAAO,OAAO,GAAG,EAAE,OAAO,KAAK,QAAQ,IAAI,cAAc,CAAC,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;AAC9E,CAAC;AAYD;;;;;GAKG;AACH,MAAM,CAAC,KAAK,UAAU,eAAe,CACnC,OAA2C,EAC3C,OAA2B,EAAE;IAE7B,MAAM,UAAU,GAAG,IAAI,CAAC,UAAU,IAAI,kBAAkB,CAAC;IACzD,MAAM,WAAW,GAAG,IAAI,CAAC,WAAW,IAAI,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,CAAC;IACrD,IAAI,SAAc,CAAC;IAEnB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,UAAU,EAAE,CAAC,EAAE,EAAE,CAAC;QACpC,IAAI,IAAI,CAAC,cAAc,EAAE,OAAO,EAAE,CAAC;YACjC,MAAM,IAAI,gBAAgB,CAAC,UAAU,EAAE,oCAAoC,CAAC,CAAC;QAC/E,CAAC;QACD,MAAM,KAAK,GAAG,iBAAiB,CAAC;YAC9B,aAAa,EAAE,IAAI,CAAC,aAAa;YACjC,UAAU,EAAE,IAAI,CAAC,UAAU;YAC3B,cAAc,EAAE,IAAI,CAAC,cAAc;SACpC,CAAC,CAAC;QACH,IAAI,CAAC;YACH,OAAO,MAAM,OAAO,CAAC,KAAK,CAAC,CAAC;QAC9B,CAAC;QAAC,OAAO,GAAQ,EAAE,CAAC;YAClB,oEAAoE;YACpE,MAAM,MAAM,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC;YAC9B,MAAM,MAAM,GACV,MAAM,IAAI,YAAY,CAAC,GAAG,CAAC;gBACzB,CAAC,CAAC,IAAI,gBAAgB,CAAC,MAAM,EAAE,+BAA+B,MAAM,EAAE,CAAC;gBACvE,CAAC,CAAC,GAAG,CAAC;YACV,SAAS,GAAG,MAAM,CAAC;YAEnB,oEAAoE;YACpE,sDAAsD;YACtD,IAAI,MAAM,YAAY,gBAAgB,IAAI,MAAM,CAAC,MAAM,KAAK,MAAM,EAAE,CAAC;gBACnE,MAAM,MAAM,CAAC;YACf,CAAC;YACD,MAAM,SAAS,GACb,MAAM,YAAY,gBAAgB,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,KAAK,MAAM,CAAC,CAAC,CAAC,WAAW,CAAC,MAAM,CAAC,CAAC;YACtF,IAAI,CAAC,SAAS,IAAI,CAAC,KAAK,UAAU,GAAG,CAAC;gBAAE,MAAM,MAAM,CAAC;YACrD,IAAI,CAAC,GAAG,EAAE,CACR,uBAAuB,CAAC,GAAG,CAAC,IAAI,UAAU,YAAY,MAAM,EAAE,OAAO,IAAI,MAAM,kCAAkC,CAClH,CAAC;QACJ,CAAC;gBAAS,CAAC;YACT,KAAK,CAAC,OAAO,EAAE,CAAC;QAClB,CAAC;IACH,CAAC;IACD,MAAM,SAAS,CAAC;AAClB,CAAC;AAoBD;;;;;;;;;GASG;AACH,MAAM,CAAC,KAAK,UAAU,mBAAmB,CACvC,MAA0B,EAC1B,IAAgB;IAEhB,IAAI,IAAI,GAAG,WAAW,CAAC;IACvB,IAAI,OAAO,GAAG,EAAE,CAAC;IACjB,IAAI,UAAU,GAAG,KAAK,CAAC;IACvB,IAAI,KAAK,GAAQ,SAAS,CAAC;IAC3B,MAAM,OAAO,GAAG,IAAI,GAAG,EAAsD,CAAC;IAE9E,IAAI,KAAK,EAAE,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;QACjC,IAAI,EAAE,CAAC;QACP,MAAM,MAAM,GAAG,KAAK,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC,CAAC;QACnC,IAAI,MAAM,EAAE,CAAC;YACX,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,IAAI,EAAE,CAAC;YACjC,IAAI,KAAK,CAAC,IAAI;gBAAE,IAAI,GAAG,KAAK,CAAC,IAAI,CAAC;YAClC,IAAI,OAAO,KAAK,CAAC,OAAO,KAAK,QAAQ,EAAE,CAAC;gBACtC,OAAO,IAAI,KAAK,CAAC,OAAO,CAAC;gBACzB,UAAU,GAAG,IAAI,CAAC;YACpB,CAAC;YACD,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,UAAU,CAAC,EAAE,CAAC;gBACpC,KAAK,MAAM,EAAE,IAAI,KAAK,CAAC,UAAU,EAAE,CAAC;oBAClC,MAAM,GAAG,GAAG,OAAO,EAAE,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,EAAE,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;oBACxD,IAAI,GAAG,GAAG,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;oBAC3B,IAAI,CAAC,GAAG,EAAE,CAAC;wBAAC,GAAG,GAAG,EAAE,EAAE,EAAE,EAAE,EAAE,IAAI,EAAE,EAAE,EAAE,IAAI,EAAE,EAAE,EAAE,CAAC;wBAAC,OAAO,CAAC,GAAG,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC;oBAAC,CAAC;oBAC1E,IAAI,EAAE,CAAC,EAAE;wBAAE,GAAG,CAAC,EAAE,GAAG,EAAE,CAAC,EAAE,CAAC;oBAC1B,IAAI,EAAE,CAAC,QAAQ,EAAE,IAAI;wBAAE,GAAG,CAAC,IAAI,IAAI,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC;oBACpD,IAAI,OAAO,EAAE,CAAC,QAAQ,EAAE,SAAS,KAAK,QAAQ;wBAAE,GAAG,CAAC,IAAI,IAAI,EAAE,CAAC,QAAQ,CAAC,SAAS,CAAC;gBACpF,CAAC;YACH,CAAC;QACH,CAAC;QACD,gEAAgE;QAChE,IAAI,KAAK,EAAE,KAAK;YAAE,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC;IACxC,CAAC;IAED,MAAM,UAAU,GAAG,OAAO,CAAC,IAAI;QAC7B,CAAC,CAAC,CAAC,GAAG,OAAO,CAAC,OAAO,EAAE,CAAC;aACnB,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;YAC5B,0EAA0E;YAC1E,mEAAmE;YACnE,kEAAkE;YAClE,uEAAuE;YACvE,6EAA6E;aAC5E,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC,EAAE,EAAE,IAAI,EAAE,UAAmB,EAAE,QAAQ,EAAE,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,SAAS,EAAE,CAAC,CAAC,IAAI,IAAI,IAAI,EAAE,EAAE,CAAC,CAAC;QACrH,CAAC,CAAC,SAAS,CAAC;IAEd,OAAO;QACL,6EAA6E;QAC7E,OAAO,EAAE,CAAC,EAAE,OAAO,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,UAAU,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,UAAU,EAAE,EAAE,CAAC;QACtG,KAAK;KACN,CAAC;AACJ,CAAC"}

package/dist/types.d.ts

@@ -48,14 +48,21 @@ export interface ProviderCredentials {
         accessKeyId?: string;
         secretAccessKey?: string;
     };
-    deepseek?: {
-        apiKey: string;
-        dangerouslyAllowBrowser?: boolean;
-    };
+    deepseek?: OpenAICompatibleCredentials;
     google?: {
         apiKey: string;
         dangerouslyAllowBrowser?: boolean;
     };
+    together?: OpenAICompatibleCredentials;
+    fireworks?: OpenAICompatibleCredentials;
+    deepinfra?: OpenAICompatibleCredentials;
+    [provider: string]: any;
+}
+export interface OpenAICompatibleCredentials {
+    apiKey: string;
+    /** Override the provider registry / model-metadata base URL. */
+    baseURL?: string;
+    dangerouslyAllowBrowser?: boolean;
 }
 export interface ToolHandler {
     execute: (args: any) => Promise<any>;
@@ -159,6 +166,10 @@ export interface InvokeOptions {
      *  no follow-up turn to hit the cache — avoids paying the 25% write surcharge
      *  with zero benefit. Agent/Prime multi-turn runs should NOT set this. */
     disableCache?: boolean;
+    /** Caller's abort signal (backend stop/timeout). Threaded to the provider's
+     *  streaming call so an in-flight token stream stops cleanly. Set by
+     *  BaseExecutor from its per-run AbortController (see cancel()). */
+    signal?: AbortSignal;
 }
 /** Leaf condition: variable op value */
 export interface LeafCondition {