qlogicagent 1.0.1 → 1.1.1

package/dist/orchestration.js

@@ -1,4 +1,4 @@
-function ce(e,t,n){return{role:"assistant",content:t||null,tool_calls:e,...n&&n.length>0?{thinkingBlocks:n}:{}}}function ue(e,t){let n=t.ok?typeof t.payload=="string"?t.payload:JSON.stringify(t.payload??""):`Error: ${t.error??"Tool execution failed"}`;return{role:"tool",tool_call_id:e,content:n}}var pe=/\b(?:daily|weekly|monthly)(?:\/(?:daily|weekly|monthly))* (?:usage )?limit(?:s)?(?: (?:exhausted|reached|exceeded))?\b/i,h={rateLimit:[/rate[_ ]limit|too many requests|429/,"model_cooldown","exceeded your current quota","resource has been exhausted","quota exceeded","resource_exhausted","usage limit",/\btpm\b/i,"tokens per minute","tokens per day"],overloaded:[/overloaded_error|"type"\s*:\s*"overloaded_error"/i,"overloaded",/service[_ ]unavailable.*(?:overload|capacity|high[_ ]demand)|(?:overload|capacity|high[_ ]demand).*service[_ ]unavailable/i,"high demand"],timeout:["timeout","timed out","service unavailable","deadline exceeded","context deadline exceeded","connection error","network error","network request failed","fetch failed","socket hang up",/\beconn(?:refused|reset|aborted)\b/i,/\benotfound\b/i,/\beai_again\b/i,/without sending (?:any )?chunks?/i,/\bstop reason:\s*(?:abort|error|network_error)\b/i,/\breason:\s*(?:abort|error|network_error)\b/i,/\bunhandled stop reason:\s*(?:abort|error|network_error)\b/i],billing:[/["']?(?:status|code)["']?\s*[:=]\s*402\b|\bhttp\s*402\b|\berror(?:\s+code)?\s*[:=]?\s*402\b|\b(?:got|returned|received)\s+(?:a\s+)?402\b|^\s*402\s+payment/i,"payment required","insufficient credits",/insufficient[_ ]quota/i,"credit balance","plans & billing","insufficient balance"],authPermanent:[/api[_ ]?key[_ ]?(?:revoked|invalid|deactivated|deleted)/i,"invalid_api_key","key has been disabled","key has been revoked","account has been deactivated",/could not (?:authenticate|validate).*(?:api[_ ]?key|credentials)/i,"permission_error","not allowed for this organization"],auth:[/invalid[_ ]?api[_ ]?key/,"incorrect api key","invalid token","authentication","re-authenticate","oauth token refresh failed","unauthorized","forbidden","access denied","insufficient permissions","insufficient permission",/missing scopes?:/i,"expired","token has expired",/\b401\b/,/\b403\b/,"no credentials found","no api key found"],format:["string should match pattern","tool_use.id","tool_use_id","messages.1.content.1.tool_use.id","invalid request format",/tool call id was.*must be/i]},de=/^(?:error[:\s-]+)?billing(?:\s+error)?(?:[:\s-]+|$)|^(?:error[:\s-]+)?(?:credit balance|insufficient credits?|payment required|http\s*402\b)/i,me=/["']?(?:status|code)["']?\s*[:=]\s*402\b|\bhttp\s*402\b|\berror(?:\s+code)?\s*[:=]?\s*402\b|^\s*402\s+payment/i,fe=512,ge=/^(?:http\s*)?(\d{3})(?:\s+([\s\S]+))?$/i;var ye=new Set([500,502,503,504,521,522,523,524,529]),he=["insufficient credits","insufficient quota","credit balance","insufficient balance","plans & billing","add more credits","top up"],Ce=["upgrade your plan","upgrade plan","current plan","subscription"],Te=["daily","weekly","monthly"],be=["try again","retry","temporary","cooldown"],ke=["usage limit","rate limit","organization usage"],_e=["organization","workspace"],Re=["billing period","exceeded","reached","exhausted"],xe=/["']?(?:status|code)["']?\s*[:=]\s*402\b|\bhttp\s*402\b|\berror(?:\s+code)?\s*[:=]?\s*402\b|\b(?:got|returned|received)\s+(?:a\s+)?402\b|^\s*402\s+payment required\b/i,Se=/^(?:error[:\s-]+)?(?:(?:http\s*)?402(?:\s+payment required)?|payment required)(?:[:\s-]+|$)/i;function C(e,t){if(!e)return!1;let n=e.toLowerCase();return t.some(o=>o instanceof RegExp?o.test(n):n.includes(o))}function Ae(e){return C(e,h.format)}function U(e){return C(e,h.rateLimit)}function ve(e){return C(e,h.timeout)}function Me(e){return pe.test(e)}function v(e){let t=e.toLowerCase();return t?e.length>fe?me.test(t):C(t,h.billing)?!0:de.test(e)?t.includes("upgrade")||t.includes("credits")||t.includes("payment")||t.includes("plan"):!1:!1}function Y(e){return C(e,h.authPermanent)}function we(e){return C(e,h.auth)}function W(e){return C(e,h.overloaded)}function y(e,t){return t.some(n=>e.includes(n))}function Ee(e){return y(e,he)||y(e,Ce)&&e.includes("limit")||e.includes("billing hard limit")||e.includes("hard limit reached")||e.includes("maximum allowed")&&e.includes("limit")}function Ie(e){let t=y(e,Te),n=e.includes("spend limit")||e.includes("spending limit"),o=y(e,_e);return y(e,be)&&y(e,ke)||t&&(e.includes("usage limit")||n)||t&&e.includes("limit")&&e.includes("reset")||o&&e.includes("limit")&&(n||y(e,Re))}function Le(e){return e.trim().toLowerCase().replace(Se,"").trim()}function K(e){let t=Le(e);return!t||Ee(t)?"billing":U(t)||Ie(t)?"rate_limit":"billing"}function Oe(e){return xe.test(e)?K(e):null}function J(e){let t=e.match(ge);if(!t)return null;let n=Number(t[1]);return Number.isFinite(n)?{code:n,rest:(t[2]??"").trim()}:null}function Pe(e){if(!e)return!1;let t=e.toLowerCase();return t.includes('"type":"api_error"')&&t.includes("internal server error")}function Fe(e){let t=e.trim();if(!t)return!1;let n=J(t);return n?ye.has(n.code):!1}function V(e,t){return typeof e!="number"||!Number.isFinite(e)?null:e===402?t?K(t):"billing":e===429?"rate_limit":e===401||e===403?t&&Y(t)?"auth_permanent":"auth":e===408?"timeout":e===503?t&&W(t)?"overloaded":"timeout":e===502||e===504?"timeout":e===529?"overloaded":e===400?t&&v(t)?"billing":"format":null}function Ne(e){if(!e)return!1;let t=e.toLowerCase();return!!(t.includes("unknown model")||t.includes("model not found")||t.includes("model_not_found")||t.includes("not_found_error")||t.includes("does not exist")&&t.includes("model")||t.includes("invalid model")&&!t.includes("invalid model reference")||/models\/[^\s]+ is not found/i.test(e)||/\b404\b/.test(e)&&/not[-_ ]?found/i.test(e))}function Be(e){if(!e)return!1;let t=e.toLowerCase();return t.includes("session not found")||t.includes("session does not exist")||t.includes("session expired")||t.includes("session invalid")||t.includes("conversation not found")||t.includes("conversation does not exist")||t.includes("conversation expired")||t.includes("conversation invalid")||t.includes("no such session")||t.includes("invalid session")||t.includes("session id not found")||t.includes("conversation id not found")}function X(e){if(Be(e))return"session_expired";if(Ne(e))return"model_not_found";let t=Oe(e);return t||(Me(e)?v(e)?"billing":"rate_limit":U(e)?"rate_limit":W(e)?"overloaded":Fe(e)?J(e.trim())?.code===529?"overloaded":"timeout":Pe(e)?"timeout":Ae(e)?"format":v(e)?"billing":ve(e)?"timeout":Y(e)?"auth_permanent":we(e)?"auth":null)}var De={timeout:"RETRYABLE_TRANSIENT",overloaded:"RETRYABLE_TRANSIENT",rate_limit:"RETRYABLE_DEGRADED",auth:"NON_RETRYABLE_AUTH",auth_permanent:"NON_RETRYABLE_AUTH",billing:"NON_RETRYABLE_QUOTA",format:"NON_RETRYABLE_CONTENT",model_not_found:"NON_RETRYABLE_CONTENT",session_expired:"NON_RETRYABLE_CONTENT",unknown:"RETRYABLE_TRANSIENT"},je=new Set(["RETRYABLE_TRANSIENT","RETRYABLE_DEGRADED","TOOL_EXECUTION_FAILED"]);function $e(e,t){let n=V(e,t)??(t?X(t):null);return n?De[n]:typeof e=="number"&&e>=400&&e<500?"NON_RETRYABLE_CONTENT":"RETRYABLE_TRANSIENT"}function qe(e){return je.has(e)}function B(e){return typeof e.compressAsync=="function"}var Q=4,M=class{constructor(t){this.estimateTokens=t}estimateTokens;compress(t,n){let o=[],r=[];for(let c of t)c.role==="system"?o.push(c):r.push(c);let a=n;for(let c of o)a-=this.estimateTokens(c);let i;for(let c of r)if(c.role==="user"){i=c;break}if(i&&(a-=this.estimateTokens(i)),a<=0)return{messages:i?[...o,i]:o,droppedCount:r.length-(i?1:0),strategy:"sliding-window"};let s=[],l=0;for(let c=r.length-1;c>=0;c--){let d=r[c];if(d===i)continue;let p=this.estimateTokens(d);if(a-p<0&&l>=Q)break;if(a-p<0&&l<Q){s.unshift(d),l++;continue}a-=p,s.unshift(d),l++}let u=[...o];return i&&!s.includes(i)&&u.push(i),u.push(...s),{messages:u,droppedCount:r.length-(s.length+(i&&!s.includes(i)?1:0)),strategy:"sliding-window"}}},w=class{constructor(t,n){this.recentCount=t;this.summarize=n}recentCount;summarize;compress(t,n){let o=t.filter(l=>l.role==="system"),r=t.filter(l=>l.role!=="system");if(r.length<=this.recentCount)return{messages:t,droppedCount:0,strategy:"summarize-old"};let a=r.slice(0,r.length-this.recentCount),i=r.slice(r.length-this.recentCount),s=this.summarize(a);return{messages:[...o,{role:"system",content:`[Conversation summary]
+function ce(e,t,n,o){return{role:"assistant",content:t||null,tool_calls:e,...n&&n.length>0?{thinkingBlocks:n}:{},...o?{reasoning_content:o}:{}}}function ue(e,t){let n=t.ok?typeof t.payload=="string"?t.payload:JSON.stringify(t.payload??""):`Error: ${t.error??"Tool execution failed"}`;return{role:"tool",tool_call_id:e,content:n}}var pe=/\b(?:daily|weekly|monthly)(?:\/(?:daily|weekly|monthly))* (?:usage )?limit(?:s)?(?: (?:exhausted|reached|exceeded))?\b/i,h={rateLimit:[/rate[_ ]limit|too many requests|429/,"model_cooldown","exceeded your current quota","resource has been exhausted","quota exceeded","resource_exhausted","usage limit",/\btpm\b/i,"tokens per minute","tokens per day"],overloaded:[/overloaded_error|"type"\s*:\s*"overloaded_error"/i,"overloaded",/service[_ ]unavailable.*(?:overload|capacity|high[_ ]demand)|(?:overload|capacity|high[_ ]demand).*service[_ ]unavailable/i,"high demand"],timeout:["timeout","timed out","service unavailable","deadline exceeded","context deadline exceeded","connection error","network error","network request failed","fetch failed","socket hang up",/\beconn(?:refused|reset|aborted)\b/i,/\benotfound\b/i,/\beai_again\b/i,/without sending (?:any )?chunks?/i,/\bstop reason:\s*(?:abort|error|network_error)\b/i,/\breason:\s*(?:abort|error|network_error)\b/i,/\bunhandled stop reason:\s*(?:abort|error|network_error)\b/i],billing:[/["']?(?:status|code)["']?\s*[:=]\s*402\b|\bhttp\s*402\b|\berror(?:\s+code)?\s*[:=]?\s*402\b|\b(?:got|returned|received)\s+(?:a\s+)?402\b|^\s*402\s+payment/i,"payment required","insufficient credits",/insufficient[_ ]quota/i,"credit balance","plans & billing","insufficient balance"],authPermanent:[/api[_ ]?key[_ ]?(?:revoked|invalid|deactivated|deleted)/i,"invalid_api_key","key has been disabled","key has been revoked","account has been deactivated",/could not (?:authenticate|validate).*(?:api[_ ]?key|credentials)/i,"permission_error","not allowed for this organization"],auth:[/invalid[_ ]?api[_ ]?key/,"incorrect api key","invalid token","authentication","re-authenticate","oauth token refresh failed","unauthorized","forbidden","access denied","insufficient permissions","insufficient permission",/missing scopes?:/i,"expired","token has expired",/\b401\b/,/\b403\b/,"no credentials found","no api key found"],format:["string should match pattern","tool_use.id","tool_use_id","messages.1.content.1.tool_use.id","invalid request format",/tool call id was.*must be/i]},de=/^(?:error[:\s-]+)?billing(?:\s+error)?(?:[:\s-]+|$)|^(?:error[:\s-]+)?(?:credit balance|insufficient credits?|payment required|http\s*402\b)/i,me=/["']?(?:status|code)["']?\s*[:=]\s*402\b|\bhttp\s*402\b|\berror(?:\s+code)?\s*[:=]?\s*402\b|^\s*402\s+payment/i,fe=512,ge=/^(?:http\s*)?(\d{3})(?:\s+([\s\S]+))?$/i;var ye=new Set([500,502,503,504,521,522,523,524,529]),he=["insufficient credits","insufficient quota","credit balance","insufficient balance","plans & billing","add more credits","top up"],Ce=["upgrade your plan","upgrade plan","current plan","subscription"],Te=["daily","weekly","monthly"],be=["try again","retry","temporary","cooldown"],ke=["usage limit","rate limit","organization usage"],_e=["organization","workspace"],Re=["billing period","exceeded","reached","exhausted"],xe=/["']?(?:status|code)["']?\s*[:=]\s*402\b|\bhttp\s*402\b|\berror(?:\s+code)?\s*[:=]?\s*402\b|\b(?:got|returned|received)\s+(?:a\s+)?402\b|^\s*402\s+payment required\b/i,Se=/^(?:error[:\s-]+)?(?:(?:http\s*)?402(?:\s+payment required)?|payment required)(?:[:\s-]+|$)/i;function C(e,t){if(!e)return!1;let n=e.toLowerCase();return t.some(o=>o instanceof RegExp?o.test(n):n.includes(o))}function Ae(e){return C(e,h.format)}function U(e){return C(e,h.rateLimit)}function ve(e){return C(e,h.timeout)}function Me(e){return pe.test(e)}function v(e){let t=e.toLowerCase();return t?e.length>fe?me.test(t):C(t,h.billing)?!0:de.test(e)?t.includes("upgrade")||t.includes("credits")||t.includes("payment")||t.includes("plan"):!1:!1}function Y(e){return C(e,h.authPermanent)}function we(e){return C(e,h.auth)}function W(e){return C(e,h.overloaded)}function y(e,t){return t.some(n=>e.includes(n))}function Ee(e){return y(e,he)||y(e,Ce)&&e.includes("limit")||e.includes("billing hard limit")||e.includes("hard limit reached")||e.includes("maximum allowed")&&e.includes("limit")}function Ie(e){let t=y(e,Te),n=e.includes("spend limit")||e.includes("spending limit"),o=y(e,_e);return y(e,be)&&y(e,ke)||t&&(e.includes("usage limit")||n)||t&&e.includes("limit")&&e.includes("reset")||o&&e.includes("limit")&&(n||y(e,Re))}function Le(e){return e.trim().toLowerCase().replace(Se,"").trim()}function K(e){let t=Le(e);return!t||Ee(t)?"billing":U(t)||Ie(t)?"rate_limit":"billing"}function Oe(e){return xe.test(e)?K(e):null}function J(e){let t=e.match(ge);if(!t)return null;let n=Number(t[1]);return Number.isFinite(n)?{code:n,rest:(t[2]??"").trim()}:null}function Pe(e){if(!e)return!1;let t=e.toLowerCase();return t.includes('"type":"api_error"')&&t.includes("internal server error")}function Fe(e){let t=e.trim();if(!t)return!1;let n=J(t);return n?ye.has(n.code):!1}function V(e,t){return typeof e!="number"||!Number.isFinite(e)?null:e===402?t?K(t):"billing":e===429?"rate_limit":e===401||e===403?t&&Y(t)?"auth_permanent":"auth":e===408?"timeout":e===503?t&&W(t)?"overloaded":"timeout":e===502||e===504?"timeout":e===529?"overloaded":e===400?t&&v(t)?"billing":"format":null}function Ne(e){if(!e)return!1;let t=e.toLowerCase();return!!(t.includes("unknown model")||t.includes("model not found")||t.includes("model_not_found")||t.includes("not_found_error")||t.includes("does not exist")&&t.includes("model")||t.includes("invalid model")&&!t.includes("invalid model reference")||/models\/[^\s]+ is not found/i.test(e)||/\b404\b/.test(e)&&/not[-_ ]?found/i.test(e))}function Be(e){if(!e)return!1;let t=e.toLowerCase();return t.includes("session not found")||t.includes("session does not exist")||t.includes("session expired")||t.includes("session invalid")||t.includes("conversation not found")||t.includes("conversation does not exist")||t.includes("conversation expired")||t.includes("conversation invalid")||t.includes("no such session")||t.includes("invalid session")||t.includes("session id not found")||t.includes("conversation id not found")}function X(e){if(Be(e))return"session_expired";if(Ne(e))return"model_not_found";let t=Oe(e);return t||(Me(e)?v(e)?"billing":"rate_limit":U(e)?"rate_limit":W(e)?"overloaded":Fe(e)?J(e.trim())?.code===529?"overloaded":"timeout":Pe(e)?"timeout":Ae(e)?"format":v(e)?"billing":ve(e)?"timeout":Y(e)?"auth_permanent":we(e)?"auth":null)}var De={timeout:"RETRYABLE_TRANSIENT",overloaded:"RETRYABLE_TRANSIENT",rate_limit:"RETRYABLE_DEGRADED",auth:"NON_RETRYABLE_AUTH",auth_permanent:"NON_RETRYABLE_AUTH",billing:"NON_RETRYABLE_QUOTA",format:"NON_RETRYABLE_CONTENT",model_not_found:"NON_RETRYABLE_CONTENT",session_expired:"NON_RETRYABLE_CONTENT",unknown:"RETRYABLE_TRANSIENT"},je=new Set(["RETRYABLE_TRANSIENT","RETRYABLE_DEGRADED","TOOL_EXECUTION_FAILED"]);function $e(e,t){let n=V(e,t)??(t?X(t):null);return n?De[n]:typeof e=="number"&&e>=400&&e<500?"NON_RETRYABLE_CONTENT":"RETRYABLE_TRANSIENT"}function qe(e){return je.has(e)}function B(e){return typeof e.compressAsync=="function"}var Q=4,M=class{constructor(t){this.estimateTokens=t}estimateTokens;compress(t,n){let o=[],r=[];for(let c of t)c.role==="system"?o.push(c):r.push(c);let a=n;for(let c of o)a-=this.estimateTokens(c);let i;for(let c of r)if(c.role==="user"){i=c;break}if(i&&(a-=this.estimateTokens(i)),a<=0)return{messages:i?[...o,i]:o,droppedCount:r.length-(i?1:0),strategy:"sliding-window"};let s=[],l=0;for(let c=r.length-1;c>=0;c--){let d=r[c];if(d===i)continue;let p=this.estimateTokens(d);if(a-p<0&&l>=Q)break;if(a-p<0&&l<Q){s.unshift(d),l++;continue}a-=p,s.unshift(d),l++}let u=[...o];return i&&!s.includes(i)&&u.push(i),u.push(...s),{messages:u,droppedCount:r.length-(s.length+(i&&!s.includes(i)?1:0)),strategy:"sliding-window"}}},w=class{constructor(t,n){this.recentCount=t;this.summarize=n}recentCount;summarize;compress(t,n){let o=t.filter(l=>l.role==="system"),r=t.filter(l=>l.role!=="system");if(r.length<=this.recentCount)return{messages:t,droppedCount:0,strategy:"summarize-old"};let a=r.slice(0,r.length-this.recentCount),i=r.slice(r.length-this.recentCount),s=this.summarize(a);return{messages:[...o,{role:"system",content:`[Conversation summary]
 ${s}`},...i],droppedCount:a.length,strategy:"summarize-old"}}},E=class{constructor(t=8e3){this.maxToolResultChars=t}maxToolResultChars;compress(t,n){let o=0;return{messages:t.map(a=>a.role!=="tool"||typeof a.content!="string"||a.content.length<=this.maxToolResultChars?a:(o++,{...a,content:Ge(a.content,this.maxToolResultChars)})),droppedCount:o,strategy:"tool-result-trim"}}};function Ge(e,t){if(e.length<=t)return e;let n=e.slice(0,t);if(e.trimStart().startsWith("{")||e.trimStart().startsWith("[")){let a=Math.max(n.lastIndexOf("},"),n.lastIndexOf("],"),n.lastIndexOf(`}
 `),n.lastIndexOf(`]
 `));if(a>t*.5)return n.slice(0,a+1)+`

package/dist/types/agent/tool-loop.d.ts

@@ -78,6 +78,8 @@ export interface ToolLoopParams {
     }) => Promise<void> | void>;
     /** Max concurrent tool executions (0 = unlimited, CC parity: semantic concurrency control). */
     maxConcurrentTools?: number;
+    /** Model requires streaming — disable non-streaming fallback. */
+    streamRequired?: boolean;
     signal?: AbortSignal;
 }
 /**

package/dist/types/agent/types.d.ts

@@ -13,6 +13,7 @@
  */
 import type { SkillInstruction } from "../orchestration/index.js";
 import type { HookRegistry } from "../contracts/hooks.js";
+import type { MediaCapability } from "../llm/provider-def.js";
 export type ChatMessageRole = "system" | "user" | "assistant" | "tool";
 /** Thinking block from Anthropic Messages API (DeepSeek, Claude, etc.) */
 export interface ThinkingBlock {
@@ -30,6 +31,38 @@ export interface ChatMessage {
     name?: string;
     /** Anthropic thinking blocks — must be passed back to API in subsequent requests */
     thinkingBlocks?: ThinkingBlock[];
+    /**
+     * Image URLs for vision-capable models.
+     * Transports convert these into the provider's native content block format:
+     *   - OpenAI-chat: [{type:"text",text:...},{type:"image_url",image_url:{url:...}}]
+     *   - Anthropic: [{type:"text",text:...},{type:"image",source:{type:"url",url:...}}]
+     * Only valid on role="user" messages.
+     */
+    imageUrls?: string[];
+    /** Image detail level for vision understanding: 'auto' | 'low' | 'high' | 'xhigh' */
+    imageDetail?: "auto" | "low" | "high" | "xhigh";
+    /** Max image pixels budget for vision understanding (Volcengine image_pixel_limit) */
+    imagePixelLimit?: {
+        minPixels?: number;
+        maxPixels?: number;
+    };
+    /** Video URLs for video-understanding models. Only valid on role="user" messages. */
+    videoUrls?: string[];
+    /** Per-video fps for video understanding (0.2-5, default 1) */
+    videoFps?: number;
+    /** Audio format hint for audio understanding: 'mp3' | 'wav' | 'aac' | 'm4a' */
+    audioFormat?: "mp3" | "wav" | "aac" | "m4a";
+    /** Audio URLs for audio-understanding models. Only valid on role="user" messages. */
+    audioUrls?: string[];
+    /**
+     * Pre-uploaded file IDs for document/media understanding.
+     * Used with Files API — each entry is a { id, mimeType } pair.
+     * Only valid on role="user" messages.
+     */
+    fileIds?: Array<{
+        id: string;
+        mimeType?: string;
+    }>;
 }
 export interface ToolCallMessage {
     id: string;
@@ -130,7 +163,7 @@ export interface TurnRequest {
 export interface TurnConfig {
     /** LLM provider id, e.g. "deepseek" */
     provider?: string;
-    /** Model id, e.g. "deepseek-chat" */
+    /** Model id, e.g. "deepseek-v4-flash" */
     model?: string;
     /** User API key */
     apiKey?: string;
@@ -160,6 +193,18 @@ export interface TurnConfig {
     summaryModel?: string;
     /** Max concurrent tool executions (0 = unlimited). */
     maxConcurrentTools?: number;
+    /** Model requires streaming — disable non-streaming fallback. */
+    streamRequired?: boolean;
+    /**
+     * User-designated provider+model per media category.
+     * Each entry maps a media type (image/video/music/tts/3d/...) to
+     * the specific provider and model the user chose. No failover —
+     * if the designated provider fails, the error propagates directly.
+     */
+    mediaProviders?: Partial<Record<MediaCapability, {
+        provider: string;
+        model: string;
+    }>>;
 }
 export type TurnEvent = {
     type: "start";

package/dist/types/cli/stdio-server.d.ts

@@ -100,6 +100,16 @@ export declare class StdioServer {
      * Supports optional category filter.
      */
     private handleToolsList;
+    /**
+     * `media.listModels` — List available media generation models.
+     * Desktop app uses this to populate the model selector per category.
+     */
+    private handleMediaListModels;
+    /**
+     * `provider.list` — List all available LLM providers grouped by logical family.
+     * Returns protocol variants so callers can pick which transport to use.
+     */
+    private handleProviderList;
     /**
      * `config.get` — Read current agent runtime configuration.
      * Returns merged user-level + project-level settings.

package/dist/types/cli/tool-bootstrap.d.ts

@@ -2,12 +2,24 @@ import type { PortableTool } from "../skills/portable-tool.js";
 import { type ExecProgress } from "../skills/tools/exec-tool.js";
 import type { AgentLogger } from "../agent/types.js";
 import type { MediaClient } from "../llm/media-client.js";
+import type { MediaCapability } from "../llm/provider-def.js";
+import type { ProviderToolAPI } from "../llm/provider-tool-api.js";
+/**
+ * Set the provider's tool API for C-level cascade.
+ * When set, web_search tool will try provider-native search first,
+ * falling back to SearXNG. Called from stdio-server per session.
+ */
+export declare function setProviderToolAPI(api: ProviderToolAPI | undefined): void;
 /**
  * Set the media client + API keys for generation tools.
  * Tools will call vendor APIs directly through MediaTransport.
  * Keys map: { providerId: apiKey, ... }
+ * Providers map: { "image": { provider: "volcengine", model: "doubao-seedream-5-0-260128" }, ... }
  */
-export declare function setMediaClientConfig(client: MediaClient | undefined, apiKeys?: Record<string, string>, onMediaUsage?: (model: string, billingUnit: string, quantity: number) => void): void;
+export declare function setMediaClientConfig(client: MediaClient | undefined, apiKeys?: Record<string, string>, onMediaUsage?: (model: string, billingUnit: string, quantity: number) => void, providers?: Partial<Record<MediaCapability, {
+    provider: string;
+    model: string;
+}>>): void;
 export interface BootstrapConfig {
     workdir?: string;
     log?: AgentLogger;

package/dist/types/llm/index.d.ts

@@ -3,7 +3,7 @@
  *
  * Provides: ProviderDef + LLMTransport + ProviderRegistry + LLMClient factory
  */
-export type { ProviderDef, ModelInfo, TransportType, AuthType, MediaCapability } from "./provider-def.js";
+export type { ProviderDef, ModelInfo, TransportType, AuthType, MediaCapability, MediaCapabilities, VideoCapabilities, ImageCapabilities, MusicCapabilities, TtsCapabilities, ThreeDCapabilities, SttCapabilities, EmbeddingCapabilities, VideoUnderstandingCapabilities, ImageUnderstandingCapabilities, VoiceCloneCapabilities, RerankCapabilities, DocumentParsingCapabilities, VideoOperation, ImageOperation, MusicOperation, TtsOperation, ThreeDOperation } from "./provider-def.js";
 export type { LLMTransport, LLMRequest, LLMChunk, AccumulatedToolCall, } from "./transport.js";
 export { accumulateToolCalls } from "./transport.js";
 export type { MediaTransport, MediaRequest, MediaResult, MediaType } from "./media-transport.js";

package/dist/types/llm/llm-client.d.ts

@@ -10,7 +10,7 @@ import { ProviderRegistry } from "./provider-registry.js";
 export interface LLMClientConfig {
     /** Provider id, e.g. "deepseek" */
     provider: string;
-    /** Model id, e.g. "deepseek-chat" */
+    /** Model id, e.g. "deepseek-v4-flash" */
     model: string;
     /** User API key */
     apiKey: string;

package/dist/types/llm/media-client.d.ts

@@ -25,11 +25,10 @@ export declare class MediaClient {
     private transports;
     constructor(config: MediaClientConfig);
     /**
-     * Find the best available model for a given media type.
-     * Scans all providers for models with matching mediaType.
-     * If preferredProvider is set, search that first.
+     * Find a specific model by provider + model id for a given media type.
+     * Used by the user-designated model routing (no failover).
      */
-    resolveModel(mediaType: MediaCapability, preferredProvider?: string): ResolvedMediaModel | undefined;
+    resolveModelById(providerId: string, modelId: string, mediaType: MediaCapability): ResolvedMediaModel | undefined;
     /**
      * Get a MediaTransport for a specific provider.
      * Creates and caches the adapter on first access.

package/dist/types/llm/media-transport.d.ts

@@ -9,7 +9,7 @@
  * Each provider adapter implements this interface and hides vendor-specific
  * auth, endpoint paths, request shapes, and polling logic.
  */
-export type MediaType = "image" | "video" | "music" | "tts" | "3d";
+export type MediaType = "image" | "video" | "music" | "tts" | "3d" | "stt" | "embedding" | "video_understanding" | "image_understanding" | "voice_clone" | "rerank" | "document_parsing";
 export interface MediaRequest {
     /** Generation model id, e.g. "doubao-seedream-5-0-260128", "gpt-image-2" */
     model: string;
@@ -39,12 +39,79 @@ export interface MediaRequest {
     channel?: string;
     /** Source video URLs for edit/merge operations */
     sourceVideos?: string[];
-    /** Reference images for edit operations */
+    /** Reference images (no role / first_frame / last_frame) */
     referenceImages?: string[];
+    /** Reference image roles — parallel array with referenceImages. */
+    imageRoles?: Array<"first_frame" | "last_frame" | "reference_image">;
+    /** Reference video URLs for multimodal reference generation (Seedance 2.0) */
+    referenceVideos?: string[];
+    /** Reference audio URLs for multimodal reference generation (Seedance 2.0) */
+    referenceAudios?: string[];
+    /** Generate synchronized audio track (Seedance 2.0 / 1.5 pro) */
+    generateAudio?: boolean;
     /** Output resolution for upscale, e.g. "1080p" */
     resolution?: string;
-    /** Operation variant: generate (default), edit, merge, upscale */
-    operation?: "generate" | "edit" | "merge" | "upscale";
+    /** Operation variant: generate (default), edit, merge, upscale, multimodal_reference, extend */
+    operation?: "generate" | "edit" | "merge" | "upscale" | "multimodal_reference" | "extend";
+    /** Image/video quality, e.g. "auto", "high", "low", "hd" */
+    quality?: string;
+    /** Seed for reproducible generation */
+    seed?: number;
+    /** TTS voice, e.g. "alloy", "nova", "shimmer" */
+    voice?: string;
+    /** TTS speech speed multiplier */
+    speed?: number;
+    /** Pure instrumental mode (MiniMax music) */
+    isInstrumental?: boolean;
+    /** Source audio URL for cover/remix (MiniMax music) */
+    audioUrl?: string;
+    /** Output audio format, e.g. "mp3", "wav", "flac" */
+    audioFormat?: string;
+    /** Video frames per second */
+    fps?: number;
+    /** Whether to add AI watermark */
+    watermark?: boolean;
+    /** img2img guidance/control strength (0-1), e.g. Volcengine */
+    guidanceScale?: number;
+    /** Auto-generate lyrics when none provided (MiniMax) */
+    lyricsOptimizer?: boolean;
+    /** 3D output format, e.g. "glb", "obj", "usd", "usdz" */
+    outputFormat?: string;
+    /** Image background mode (OpenAI gpt-image-2): "transparent", "opaque", "auto" */
+    background?: string;
+    /** Provider-specific request metadata (e.g. rerank documents, parsing options) */
+    metadata?: Record<string, unknown>;
+    /** Ask the provider to optimize/enhance the prompt before generation */
+    enhancePrompt?: boolean;
+    /**
+     * Progress callback for async polling operations (video gen, 3D gen, etc.).
+     * Called periodically with estimated progress percentage and status text.
+     */
+    onProgress?: (percent: number, status: string) => void;
+    /** Enable progressive/streaming image generation (Volcengine Seedream stream:true) */
+    streamImage?: boolean;
+    /** Lock camera position (Seedance 1.0/1.5, not Seedance 2.0) */
+    cameraFixed?: boolean;
+    /** Return last frame URL for chaining continuous video segments */
+    returnLastFrame?: boolean;
+    /** Service tier: 'default' (online) or 'flex' (offline, ~50% cost) — not all models support flex */
+    serviceTier?: "default" | "flex";
+    /** Task expiration in seconds (for flex/offline scheduling) */
+    executionExpiresAfterSeconds?: number;
+    /** Draft mode — low-cost preview (Seedance 1.5 pro only) */
+    draft?: boolean;
+    /** Draft task ID to promote to final video */
+    draftTaskId?: string;
+    /** Video-level builtin tools, e.g. ["web_search"] (Seedance 2.0) */
+    videoTools?: string[];
+    /** End-user safety identifier for content moderation */
+    safetyIdentifier?: string;
+    /** Callback URL for async task status push notifications */
+    callbackUrl?: string;
+    /** Image detail level control: 'auto' | 'low' | 'high' */
+    detail?: "auto" | "low" | "high";
+    /** Max image pixels budget (Volcengine image_pixel_limit) */
+    imagePixelLimit?: number;
 }
 export interface MediaResult {
     /** URLs of generated media files */
@@ -61,6 +128,10 @@ export interface MediaResult {
     billingQuantity?: number;
     /** Provider-specific metadata */
     metadata?: Record<string, unknown>;
+    /** Last frame image URL for chaining continuous video generation */
+    lastFrameUrl?: string;
+    /** Task ID (for continuing draft→final or querying) */
+    taskId?: string;
 }
 export interface MediaTransport {
     /**

package/dist/types/llm/provider-def.d.ts

@@ -9,9 +9,116 @@
  *   Layer 2: model-catalog.ts remote (models.dev)
  *   Layer 3: user config (from agent.turn.config)
  */
-export type TransportType = "openai-chat" | "anthropic-messages";
+export type TransportType = "openai-chat" | "anthropic-messages" | "volcengine-responses";
 export type AuthType = "bearer" | "x-api-key" | "none";
-export type MediaCapability = "image" | "video" | "music" | "tts" | "3d";
+export type MediaCapability = "image" | "video" | "music" | "tts" | "3d" | "stt" | "embedding" | "video_understanding" | "image_understanding" | "voice_clone" | "rerank" | "document_parsing";
+export type VideoOperation = "text2video" | "img2video" | "video2video" | "edit" | "merge" | "upscale";
+export type ImageOperation = "text2image" | "img2img" | "inpainting" | "outpainting";
+export type MusicOperation = "text2music" | "cover";
+export type TtsOperation = "text2speech" | "voice_clone";
+export type ThreeDOperation = "text2_3d" | "img2_3d";
+export interface VideoCapabilities {
+    type: "video";
+    operations: VideoOperation[];
+    maxDurationSeconds?: number;
+    resolutions?: string[];
+    aspectRatios?: string[];
+    fps?: number[];
+}
+export interface ImageCapabilities {
+    type: "image";
+    operations: ImageOperation[];
+    sizes?: string[];
+    transparentBackground?: boolean;
+}
+export interface MusicCapabilities {
+    type: "music";
+    operations: MusicOperation[];
+    maxDurationSeconds?: number;
+    formats?: string[];
+}
+export interface TtsCapabilities {
+    type: "tts";
+    operations?: TtsOperation[];
+    voices?: string[];
+    maxCharacters?: number;
+    formats?: string[];
+}
+export interface ThreeDCapabilities {
+    type: "3d";
+    operations: ThreeDOperation[];
+    outputFormats?: string[];
+}
+export interface SttCapabilities {
+    type: "stt";
+    languages?: string[];
+    maxDurationSeconds?: number;
+    formats?: string[];
+}
+export interface EmbeddingCapabilities {
+    type: "embedding";
+    dimensions?: number;
+    maxTokens?: number;
+}
+export interface VideoUnderstandingCapabilities {
+    type: "video_understanding";
+    maxDurationSeconds?: number;
+    formats?: string[];
+}
+export interface ImageUnderstandingCapabilities {
+    type: "image_understanding";
+    formats?: string[];
+}
+export interface VoiceCloneCapabilities {
+    type: "voice_clone";
+    maxSampleDurationSeconds?: number;
+    maxSampleSizeMB?: number;
+    formats?: string[];
+}
+export interface RerankCapabilities {
+    type: "rerank";
+    maxDocuments?: number;
+    maxQueryLength?: number;
+    maxDocumentLength?: number;
+}
+export interface DocumentParsingCapabilities {
+    type: "document_parsing";
+    supportedFormats?: string[];
+    maxPageCount?: number;
+    maxFileSizeMB?: number;
+}
+export type MediaCapabilities = VideoCapabilities | ImageCapabilities | MusicCapabilities | TtsCapabilities | ThreeDCapabilities | SttCapabilities | EmbeddingCapabilities | VideoUnderstandingCapabilities | ImageUnderstandingCapabilities | VoiceCloneCapabilities | RerankCapabilities | DocumentParsingCapabilities;
+/**
+ * Provider-specific quirks — drives conditional logic in transports.
+ * CC parity: provider detection via quirks flags instead of hardcoded if/else.
+ * altcode parity: provider auto-detect + per-provider parameter translation.
+ */
+export interface ProviderQuirks {
+    /** Provider doesn't support thinking content blocks (Qwen) */
+    filterThinkingBlocks?: boolean;
+    /** Provider doesn't support image content blocks — strip imageUrls before sending (DeepSeek, MiniMax) */
+    filterImageBlocks?: boolean;
+    /** DeepSeek: budget_tokens ignored, use output_config.effort instead */
+    useEffortInsteadOfBudget?: boolean;
+    /** Provider supports reasoning_effort param (Kimi K2, OpenAI o-series) */
+    supportsReasoningEffort?: boolean;
+    /** Provider has built-in web search (Kimi: builtin_function.$web_search, GLM: web_search) */
+    builtinWebSearch?: boolean;
+    /** Provider has built-in code interpreter */
+    builtinCodeInterpreter?: boolean;
+    /** Supports thinking.type="enabled"/"disabled" body param (Kimi K2, GLM).
+     *  Disambiguation: GLM also sets supportsToolStream; Kimi does not. */
+    supportsThinkingParam?: boolean;
+    /** GLM-only: supports tool_stream=true for incremental tool call streaming */
+    supportsToolStream?: boolean;
+    /** DeepSeek only maps to "high"|"max"; low/medium→high */
+    maxReasoningEffort?: "high" | "max";
+    /** Supports prefix completion via /beta endpoint (DeepSeek Beta) */
+    supportsPrefixCompletion?: boolean;
+    /** MiniMax OpenAI route: inject reasoning_split=true to split thinking into reasoning_details.
+     *  Streaming uses cumulative string updates (not incremental deltas). */
+    supportsReasoningSplit?: boolean;
+}
 export interface ProviderDef {
     /** Unique provider id, e.g. "deepseek", "openai", "anthropic" */
     id: string;
@@ -21,6 +128,12 @@ export interface ProviderDef {
     transport: TransportType;
     /** API base URL, e.g. "https://api.deepseek.com" */
     baseUrl: string;
+    /**
+     * Logical provider group — links protocol variants of the same vendor.
+     * e.g. both "zhipu" (anthropic) and "zhipu-openai" share group "zhipu".
+     * Defaults to provider id if unset.
+     */
+    group?: string;
     /** Env var names for API key (priority order) */
     apiKeyEnvVars: string[];
     /** Auth header style */
@@ -37,9 +150,11 @@ export interface ProviderDef {
     supportsStreamOptions?: boolean;
     /** Whether to omit temperature when it equals 0 (some providers reject 0) */
     omitZeroTemperature?: boolean;
+    /** Provider-specific quirks for transport-level conditional logic */
+    quirks?: ProviderQuirks;
 }
 export interface ModelInfo {
-    /** Model id, e.g. "deepseek-chat" */
+    /** Model id, e.g. "deepseek-v4-flash" */
     id: string;
     /** Display name, e.g. "DeepSeek Chat V3" */
     name: string;
@@ -51,6 +166,10 @@ export interface ModelInfo {
     toolCall: boolean;
     /** Has reasoning/thinking mode */
     reasoning: boolean;
+    /** Thinking is forced on — cannot be toggled off (e.g. QwQ, DeepSeek-R1) */
+    reasoningRequired?: boolean;
+    /** Model only supports streaming (non-stream requests will fail) */
+    streamRequired?: boolean;
     /** Supports vision (image input) */
     vision: boolean;
     /** Cost per 1M input tokens (USD) */
@@ -63,4 +182,6 @@ export interface ModelInfo {
     costCacheWrite?: number;
     /** Media generation capability — undefined means chat/reasoning model */
     mediaType?: MediaCapability;
+    /** Fine-grained media capabilities — operations, formats, limits */
+    mediaCapabilities?: MediaCapabilities;
 }

package/dist/types/llm/provider-registry.d.ts

@@ -41,6 +41,11 @@ export declare class ProviderRegistry {
      * Merges: Layer 3 override > Layer 1 builtin > Layer 2 catalog enrichment.
      */
     listModels(providerId: string): ModelInfo[];
+    /**
+     * Look up a single model's info by provider + model id.
+     * Returns undefined if the model is not found.
+     */
+    getModelInfo(providerId: string, modelId: string): ModelInfo | undefined;
     /**
      * Trigger background refresh of the remote model catalog.
      */

package/dist/types/llm/provider-tool-api.d.ts

@@ -0,0 +1,44 @@
+/**
+ * ProviderToolAPI — interface for provider-specific utility endpoints
+ * that are neither LLM chat nor media generation.
+ *
+ * Examples: web search, content reader, tokenizer, moderation, realtime voice.
+ * Each provider can expose its own set of tool APIs; the agent's tool cascade
+ * mechanism (Q1) routes to these when the provider has a native capability.
+ */
+export interface WebSearchResult {
+    title: string;
+    url: string;
+    snippet: string;
+    /** Full page content if available */
+    content?: string;
+}
+export interface ReaderResult {
+    title: string;
+    content: string;
+    url: string;
+}
+export interface TokenizerResult {
+    tokenCount: number;
+    model: string;
+}
+export interface ModerationResult {
+    flagged: boolean;
+    categories: Record<string, boolean>;
+    scores?: Record<string, number>;
+}
+export interface ProviderToolAPI {
+    /** Which tool APIs this provider supports */
+    readonly capabilities: readonly ProviderToolCapability[];
+    /** Web search — returns search result list */
+    webSearch?(query: string, options?: {
+        maxResults?: number;
+    }): Promise<WebSearchResult[]>;
+    /** URL reader — extracts content from a web page */
+    reader?(url: string): Promise<ReaderResult>;
+    /** Tokenizer — count tokens for given text/model */
+    tokenize?(text: string, model: string): Promise<TokenizerResult>;
+    /** Content moderation — check text for policy violations */
+    moderate?(text: string): Promise<ModerationResult>;
+}
+export type ProviderToolCapability = "web_search" | "reader" | "tokenizer" | "moderations";

package/dist/types/llm/retry.d.ts

@@ -0,0 +1,37 @@
+/**
+ * Shared retry/backoff utilities for LLM transport implementations.
+ *
+ * Provides common constants and helper functions used by all transports
+ * (anthropic-messages, openai-chat, volcengine-responses) to handle
+ * transient errors with exponential backoff.
+ */
+/** Default maximum number of retry attempts */
+export declare const DEFAULT_MAX_RETRIES = 3;
+/** Base delay for exponential backoff (doubles each attempt, capped at 30s) */
+export declare const RETRY_BASE_DELAY_MS = 1000;
+/** Maximum backoff delay */
+export declare const RETRY_MAX_DELAY_MS = 30000;
+/** HTTP status codes considered transient (worth retrying) */
+export declare const TRANSIENT_STATUS_CODES: Set<number>;
+/** Default timeout for idle stream detection (no data received) */
+export declare const STREAM_IDLE_TIMEOUT_MS = 90000;
+/**
+ * Calculate the delay for a given retry attempt using exponential backoff with jitter.
+ * @param attempt 1-based attempt number (1 = first retry)
+ * @returns delay in milliseconds
+ */
+export declare function retryDelay(attempt: number): number;
+/**
+ * Check if an HTTP status code indicates a transient error worth retrying.
+ */
+export declare function isTransientStatus(status: number | null | undefined): boolean;
+/**
+ * Sleep with abort signal support. Resolves after `ms` milliseconds
+ * or rejects if the signal is aborted.
+ */
+export declare function retrySleep(ms: number, signal?: AbortSignal): Promise<void>;
+/**
+ * Extract HTTP status from various error shapes.
+ * Works with fetch Response errors, Axios errors, and generic errors with status property.
+ */
+export declare function extractHttpStatus(error: unknown): number | null;