@roll-agent/runtime 0.2.0 → 0.3.0

package/dist/engine/agent-session.d.ts

@@ -1,5 +1,5 @@
 import { type ModelMessage } from "ai";
-import type { LanguageModelV3, SharedV3ProviderOptions } from "@ai-sdk/provider";
+import type { LanguageModelV4, SharedV4ProviderOptions } from "@ai-sdk/provider";
 import type { ContextCompactionReason, ContextCompactionStrategy, SessionEvent, SessionTokenUsage } from "../types/events.ts";
 import type { ToolPolicy } from "../types/policy.ts";
 import { type AgentToolSource } from "../tool-bridge/build-tools.ts";
@@ -12,7 +12,7 @@ export interface SessionCompactionSettings {
 }
 export interface AgentSessionOptions {
     readonly id: string;
-    readonly model: LanguageModelV3;
+    readonly model: LanguageModelV4;
     readonly sources: readonly AgentToolSource[];
     readonly maxSteps: number;
     readonly policy?: ToolPolicy;
@@ -22,7 +22,7 @@ export interface AgentSessionOptions {
     readonly contextWindow?: number;
     readonly compaction?: SessionCompactionSettings;
     readonly turnTimeoutMs?: number;
-    readonly providerOptions?: SharedV3ProviderOptions;
+    readonly providerOptions?: SharedV4ProviderOptions;
     readonly debugEvents?: boolean;
 }
 export declare class AgentSession {
@@ -55,7 +55,7 @@ export declare class AgentSession {
     getMessages(): readonly ModelMessage[];
     getContextWindow(): number | undefined;
     getSessionUsage(): SessionTokenUsage;
-    setProviderOptions(providerOptions: SharedV3ProviderOptions | undefined): void;
+    setProviderOptions(providerOptions: SharedV4ProviderOptions | undefined): void;
     private shouldAutoCompact;
     private runCompactionTurn;
     private runCompaction;

package/dist/engine/agent-session.js

@@ -1 +1 @@
-import{randomUUID as e}from"node:crypto";import{stepCountIs as t,streamText as s}from"ai";import{buildAgentToolset as o}from"../tool-bridge/build-tools.js";import{readIsError as n}from"../tool-bridge/normalize-result.js";import{ApprovalGate as i}from"../approval/approval-gate.js";import{compactMessages as a}from"./compactor.js";import{AsyncEventQueue as r}from"./event-queue.js";const l="你是 Roll chat 的运行时助手。你可以使用模型的 thinking/reasoning 能力完成内部推理，但必须把给用户看的最终回复写入普通 text 输出通道；不要只在 reasoning 中写最终答案。 工具调用完成后，也要在普通 text 输出通道给出简洁结论。最终回复不要重复。 不要复述用户输入；如果需要调用工具，直接调用工具，不要先输出用户原文或无意义前置文本。";function u(e){return e instanceof Error?e.message:String(e)}function p(e){return"object"==typeof e&&null!==e&&"output"in e}function h(e){return p(e)?u(e.output):u(e)}function c(e){const t=h(e);return t.startsWith("已取消执行")||t.startsWith("策略拒绝执行")?t:void 0}function g(e){return/context[_ -]?length|context window|maximum context|token limit|too many tokens|prompt is too long|input is too long/i.test(u(e))}function m(e){const t=e.inputTokenDetails?.cacheReadTokens,s=e.outputTokenDetails?.reasoningTokens;return{...void 0!==e.inputTokens?{inputTokens:e.inputTokens}:{},...void 0!==e.outputTokens?{outputTokens:e.outputTokens}:{},...void 0!==e.totalTokens?{totalTokens:e.totalTokens}:{},...void 0!==t?{cachedInputTokens:t}:{},...void 0!==s?{reasoningTokens:s}:{}}}function d(e,t){return void 0===e&&void 0===t?void 0:(e??0)+(t??0)}function v(e,t){const s=d(e.cachedInputTokens,t.cachedInputTokens),o=d(e.reasoningTokens,t.reasoningTokens);return{inputTokens:(e.inputTokens??0)+(t.inputTokens??0),outputTokens:(e.outputTokens??0)+(t.outputTokens??0),totalTokens:(e.totalTokens??0)+(t.totalTokens??0),...void 0!==s?{cachedInputTokens:s}:{},...void 0!==o?{reasoningTokens:o}:{}}}function T(e,t){return void 0===t?e:void 0===e?t:Math.max(e,t)}function b(e,t){const s=e.trim(),o=t.trim();return s.length>0&&o.length>0&&o.startsWith(s)}function k(e){return e.flatMap(e=>{if("assistant"!==e.role||!Array.isArray(e.content))return[e];const t=e.content.filter(e=>"reasoning"!==e.type);return 0===t.length?[]:t.length===e.content.length?[e]:[{...e,content:t}]})}export class AgentSession{id;model;maxSteps;messages;onPersist;onReplace;contextWindow;compaction;turnTimeoutMs;providerOptions;debugEvents;gate=new i;tools;registry;emit;activeTurn;sessionUsage={inputTokens:0,outputTokens:0,totalTokens:0};lastInputTokens;needsCompaction=!1;constructor(e){this.id=e.id,this.model=e.model,this.maxSteps=e.maxSteps,this.messages=e.initialMessages?k(e.initialMessages):[],this.onPersist=e.onPersist,this.onReplace=e.onReplace,this.contextWindow=e.contextWindow,this.compaction=e.compaction,this.turnTimeoutMs=e.turnTimeoutMs,this.providerOptions=e.providerOptions,this.debugEvents=e.debugEvents??!1;const t=o(e.sources,{...e.policy?{policy:e.policy}:{},requestApproval:e=>this.requestApproval(e)});this.tools=t.tools,this.registry=t.registry}async*send(e){if(this.activeTurn)throw new Error("session already has an active turn");const t=new r;this.emit=e=>t.push(e);const s={abortController:new AbortController,aborted:!1};this.activeTurn=s,this.runTurn(t,s,e).catch(e=>{t.push({type:"error",stage:"execute",message:u(e)}),t.close()});try{for await(const e of t)yield e}finally{this.activeTurn===s&&this.abort(),void 0!==this.emit&&(this.emit=void 0)}}async*compact(e="manual"){if(this.activeTurn)throw new Error("session already has an active turn");const t=new r;this.emit=e=>t.push(e);const s={abortController:new AbortController,aborted:!1};this.activeTurn=s,this.runCompactionTurn(t,s,e).catch(e=>{t.push({type:"error",stage:"plan",message:u(e)}),t.close()});try{for await(const e of t)yield e}finally{this.activeTurn===s&&this.abort(),void 0!==this.emit&&(this.emit=void 0)}}async runTurn(o,i,a){const r=Date.now();let p;try{if(this.debug(o,"turn","start",r,{messages:this.messages.length,tools:Object.keys(this.tools).length,maxSteps:this.maxSteps,...void 0!==this.contextWindow?{contextWindow:this.contextWindow}:{},...void 0!==this.lastInputTokens?{lastInputTokens:this.lastInputTokens}:{}}),this.shouldAutoCompact()){this.debug(o,"compaction","auto requested before turn",r,{messages:this.messages.length,...void 0!==this.lastInputTokens?{lastInputTokens:this.lastInputTokens}:{}});try{await this.runCompaction(o,"auto",i)}catch(e){return void o.push({type:"error",stage:"plan",message:u(e)})}}if(i.aborted||i.abortController.signal.aborted)return void o.push({type:"error",stage:"execute",message:"aborted"});p=this.messages.length,this.messages.push({role:"user",content:a}),o.push({type:"message-start",messageId:e()}),this.debug(o,"model","calling streamText",r,{messages:this.messages.length,tools:Object.keys(this.tools).length,...void 0!==this.turnTimeoutMs?{timeoutMs:this.turnTimeoutMs}:{}});const d=s({model:this.model,system:l,messages:this.messages,tools:this.tools,stopWhen:t(this.maxSteps),abortSignal:i.abortController.signal,...this.providerOptions?{providerOptions:this.providerOptions}:{},...void 0!==this.turnTimeoutMs?{timeout:{totalMs:this.turnTimeoutMs}}:{}});this.debug(o,"model","streamText returned",r);let y,f,x,w,C,I="",N="",M=!1,R=0,A=!1;const S=this.scheduleDebug(o,"model","waiting for first stream event",r,{messages:this.messages.length});try{for await(const e of d.fullStream){switch(A||(A=!0,this.clearDebugTimer(S),this.debug(o,"model","first stream event",r,{part:e.type})),e.type){case"text-delta":if(!M){const t=N+e.text;if(b(t,a)){N=t;break}if(N.length>0){const t=N+e.text;N="",I+=t,o.push({type:"text-delta",delta:t});break}}I+=e.text,o.push({type:"text-delta",delta:e.text});break;case"tool-call":{M=!0,N="";const t=this.registry.resolve(e.toolName);o.push({type:"tool-call",toolCallId:e.toolCallId,agentName:t?.agentName??e.toolName,toolName:t?.toolName??e.toolName,input:e.input});break}case"tool-result":{const t=this.registry.resolve(e.toolName),s=n(e.output);o.push({type:"tool-result",toolCallId:e.toolCallId,agentName:t?.agentName??e.toolName,toolName:t?.toolName??e.toolName,output:e.output,isError:s}),i.aborted||i.abortController.signal.aborted||!s||(C=c(e.output));break}case"tool-error":{const t=this.registry.resolve(e.toolName),s=h(e.error);o.push({type:"tool-result",toolCallId:e.toolCallId,agentName:t?.agentName??e.toolName,toolName:t?.toolName??e.toolName,output:s,isError:!0}),i.aborted||i.abortController.signal.aborted||(C=c(e.error));break}case"finish-step":{const t=m(e.usage);f=T(f,t.inputTokens),R+=1,x=e.finishReason,o.push({type:"step-finish",finishReason:e.finishReason,usage:t});break}case"finish":!M&&N.length>0&&(I+=N,o.push({type:"text-delta",delta:N}),N=""),y=m(e.totalUsage);break;case"error":g(e.error)&&(this.needsCompaction=!0),w=u(e.error),o.push({type:"error",stage:"execute",message:w});break;case"abort":i.aborted=!0}if(void 0!==C)break}}finally{this.clearDebugTimer(S)}if(this.debug(o,"model","fullStream finished",r,{textChars:I.length}),void 0!==w)return this.messages.splice(p),void(this.needsCompaction&&await this.recoverFromContextError(o,i));if(i.aborted||i.abortController.signal.aborted)return this.messages.splice(p),void o.push({type:"error",stage:"execute",message:"aborted"});if(void 0!==C){const e=C,t=0===I.length?e:`\n${e}`;o.push({type:"text-delta",delta:t}),this.messages.push({role:"assistant",content:e}),this.debug(o,"persist","persisting messages",r,{appendedMessages:this.messages.length-p}),this.onPersist?.(this.messages.slice(p)),this.debug(o,"persist","messages persisted",r,{totalMessages:this.messages.length}),y&&(this.sessionUsage=v(this.sessionUsage,y));const s=f??y?.inputTokens;return void 0!==s&&(this.lastInputTokens=s),void o.push({type:"message-finish",text:e,...y?{totalUsage:y}:{},sessionUsage:{...this.sessionUsage},...void 0!==f?{contextInputTokens:f}:{}})}let E;this.debug(o,"model","awaiting response messages",r);const U=this.scheduleDebug(o,"model","still awaiting response messages",r);try{E=(await d.response).messages}catch(e){return this.clearDebugTimer(U),g(e)&&(this.needsCompaction=!0),this.messages.splice(p),o.push({type:"error",stage:"execute",message:u(e)}),void(this.needsCompaction&&await this.recoverFromContextError(o,i))}finally{this.clearDebugTimer(U)}this.debug(o,"model","response messages ready",r,{responseMessages:E.length});const W=k(E);this.messages.push(...W),this.debug(o,"persist","persisting messages",r,{appendedMessages:this.messages.length-p}),this.onPersist?.(this.messages.slice(p)),this.debug(o,"persist","messages persisted",r,{totalMessages:this.messages.length}),y&&(this.sessionUsage=v(this.sessionUsage,y));const D=f??y?.inputTokens;void 0!==D&&(this.lastInputTokens=D);const O=R>=this.maxSteps&&"tool-calls"===x;o.push({type:"message-finish",text:I,...y?{totalUsage:y}:{},sessionUsage:{...this.sessionUsage},...void 0!==f?{contextInputTokens:f}:{},...O?{stoppedAtStepLimit:!0}:{}})}catch(e){void 0!==p&&this.messages.splice(p),g(e)&&(this.needsCompaction=!0),this.debug(o,"turn","error",r,{message:u(e)}),o.push({type:"error",stage:"execute",message:u(e)}),this.needsCompaction&&await this.recoverFromContextError(o,i)}finally{this.activeTurn===i&&(this.activeTurn=void 0),o.close()}}requestApproval(t){const s=e(),o=this.gate.request(s);return void 0===this.emit?(this.gate.resolve(s,{approved:!1,reason:"approval event could not be delivered"}),o):(this.emit({type:"confirmation-required",approvalId:s,agentName:t.agentName,toolName:t.toolName,input:t.input,...t.reason?{reason:t.reason}:{}}),o)}approve(e){return this.gate.resolve(e,{approved:!0})}reject(e,t){return this.gate.resolve(e,{approved:!1,...t?{reason:t}:{}})}getMessages(){return this.messages}getContextWindow(){return this.contextWindow}getSessionUsage(){return{...this.sessionUsage}}setProviderOptions(e){this.providerOptions=e}shouldAutoCompact(){const e=this.compaction;return void 0!==e&&e.enabled&&(this.needsCompaction||void 0!==this.contextWindow&&void 0!==this.lastInputTokens&&this.lastInputTokens/this.contextWindow>=e.threshold)}async runCompactionTurn(e,t,s){try{await this.runCompaction(e,s,t)}catch(t){e.push({type:"error",stage:"plan",message:u(t)})}finally{this.activeTurn===t&&(this.activeTurn=void 0),e.close()}}async runCompaction(e,t,s){const o=Date.now(),n=this.compaction,i=n?.strategy??"summarize";if(this.debug(e,"compaction","start",o,{reason:t,messages:this.messages.length}),e.push({type:"compaction-start",reason:t}),this.isTurnAborted(s))return void e.push({type:"error",stage:"plan",message:"aborted"});if(!n)return void e.push({type:"context-compacted",reason:t,strategy:i,removed:0,kept:this.messages.length});const r=this.lastInputTokens;let l=n.strategy;const p=s?.abortController.signal;let h;try{h=await a({messages:this.messages,strategy:l,keepRecentTurns:n.keepRecentTurns,keepRecentTokens:n.keepRecentTokens,model:this.model,...p?{abortSignal:p}:{}})}catch(t){if(this.isTurnAborted(s))return void e.push({type:"error",stage:"plan",message:"aborted"});if("summarize"!==l)throw t;this.debug(e,"compaction","summarize failed, fallback to truncate",o,{message:u(t)}),l="truncate",h=await a({messages:this.messages,strategy:l,keepRecentTurns:n.keepRecentTurns,keepRecentTokens:n.keepRecentTokens,model:this.model,...p?{abortSignal:p}:{}})}if(this.isTurnAborted(s))return void e.push({type:"error",stage:"plan",message:"aborted"});const c=h.removed>0||h.truncatedTools>0;c&&(this.onReplace?.(h.messages),this.messages.splice(0,this.messages.length,...h.messages),this.lastInputTokens=void 0,this.needsCompaction=!1),e.push({type:"context-compacted",reason:t,strategy:l,removed:h.removed,kept:h.kept,...h.truncatedTools>0?{truncatedTools:h.truncatedTools}:{},...void 0!==r?{beforeInputTokens:r}:{}}),this.debug(e,"compaction","finish",o,{reason:t,strategy:l,removed:h.removed,kept:h.kept,truncatedTools:h.truncatedTools,progressed:c})}async recoverFromContextError(e,t){if(this.compaction?.enabled)try{await this.runCompaction(e,"auto",t)}catch(t){e.push({type:"error",stage:"plan",message:u(t)})}}isTurnAborted(e){return!0===e?.aborted||!0===e?.abortController.signal.aborted}abort(){this.activeTurn&&(this.activeTurn.aborted=!0),this.gate.abortAll(),this.activeTurn?.abortController.abort()}debug(e,t,s,o,n){this.debugEvents&&e.push({type:"debug",stage:t,message:s,...void 0!==o?{elapsedMs:Date.now()-o}:{},...void 0!==n?{data:n}:{}})}scheduleDebug(e,t,s,o,n){if(this.debugEvents)return setTimeout(()=>{this.debug(e,t,s,o,n)},5e3)}clearDebugTimer(e){void 0!==e&&clearTimeout(e)}}
+import{randomUUID as e}from"node:crypto";import{stepCountIs as t,streamText as s}from"ai";import{buildAgentToolset as o}from"../tool-bridge/build-tools.js";import{readIsError as n}from"../tool-bridge/normalize-result.js";import{ApprovalGate as i}from"../approval/approval-gate.js";import{compactMessages as r}from"./compactor.js";import{AsyncEventQueue as a}from"./event-queue.js";const u="你是 Roll chat 的运行时助手。你可以使用模型的 thinking/reasoning 能力完成内部推理，但必须把给用户看的最终回复写入普通 text 输出通道；不要只在 reasoning 中写最终答案。 工具调用完成后，也要在普通 text 输出通道给出简洁结论。最终回复不要重复。 不要复述用户输入；如果需要调用工具，直接调用工具，不要先输出用户原文或无意义前置文本。";function l(e){return e instanceof Error?e.message:String(e)}function p(e){return"object"==typeof e&&null!==e&&"output"in e}function h(e){return p(e)?l(e.output):l(e)}function c(e){const t=h(e);return t.startsWith("已取消执行")||t.startsWith("策略拒绝执行")?t:void 0}function d(e){return/context[_ -]?length|context window|maximum context|token limit|too many tokens|prompt is too long|input is too long/i.test(l(e))}function m(e){const t=e.inputTokenDetails?.cacheReadTokens,s=e.inputTokenDetails?.cacheWriteTokens,o=e.outputTokenDetails?.reasoningTokens;return{...void 0!==e.inputTokens?{inputTokens:e.inputTokens}:{},...void 0!==e.outputTokens?{outputTokens:e.outputTokens}:{},...void 0!==e.totalTokens?{totalTokens:e.totalTokens}:{},...void 0!==t?{cachedInputTokens:t}:{},...void 0!==s?{cacheWriteTokens:s}:{},...void 0!==o?{reasoningTokens:o}:{}}}function g(e,t){return void 0===e&&void 0===t?void 0:(e??0)+(t??0)}function v(e,t){const s=g(e.cachedInputTokens,t.cachedInputTokens),o=g(e.cacheWriteTokens,t.cacheWriteTokens),n=g(e.reasoningTokens,t.reasoningTokens);return{inputTokens:(e.inputTokens??0)+(t.inputTokens??0),outputTokens:(e.outputTokens??0)+(t.outputTokens??0),totalTokens:(e.totalTokens??0)+(t.totalTokens??0),...void 0!==s?{cachedInputTokens:s}:{},...void 0!==o?{cacheWriteTokens:o}:{},...void 0!==n?{reasoningTokens:n}:{}}}function T(e,t){return void 0===t?e:void 0===e?t:Math.max(e,t)}function b(e,t){const s=e.trim(),o=t.trim();return s.length>0&&o.length>0&&o.startsWith(s)}function k(e){return e.flatMap(e=>{if("assistant"!==e.role||!Array.isArray(e.content))return[e];const t=e.content.filter(e=>"reasoning"!==e.type);return 0===t.length?[]:t.length===e.content.length?[e]:[{...e,content:t}]})}export class AgentSession{id;model;maxSteps;messages;onPersist;onReplace;contextWindow;compaction;turnTimeoutMs;providerOptions;debugEvents;gate=new i;tools;registry;emit;activeTurn;sessionUsage={inputTokens:0,outputTokens:0,totalTokens:0};lastInputTokens;needsCompaction=!1;constructor(e){this.id=e.id,this.model=e.model,this.maxSteps=e.maxSteps,this.messages=e.initialMessages?k(e.initialMessages):[],this.onPersist=e.onPersist,this.onReplace=e.onReplace,this.contextWindow=e.contextWindow,this.compaction=e.compaction,this.turnTimeoutMs=e.turnTimeoutMs,this.providerOptions=e.providerOptions,this.debugEvents=e.debugEvents??!1;const t=o(e.sources,{...e.policy?{policy:e.policy}:{},requestApproval:e=>this.requestApproval(e)});this.tools=t.tools,this.registry=t.registry}async*send(e){if(this.activeTurn)throw new Error("session already has an active turn");const t=new a;this.emit=e=>t.push(e);const s={abortController:new AbortController,aborted:!1};this.activeTurn=s,this.runTurn(t,s,e).catch(e=>{t.push({type:"error",stage:"execute",message:l(e)}),t.close()});try{for await(const e of t)yield e}finally{this.activeTurn===s&&this.abort(),void 0!==this.emit&&(this.emit=void 0)}}async*compact(e="manual"){if(this.activeTurn)throw new Error("session already has an active turn");const t=new a;this.emit=e=>t.push(e);const s={abortController:new AbortController,aborted:!1};this.activeTurn=s,this.runCompactionTurn(t,s,e).catch(e=>{t.push({type:"error",stage:"plan",message:l(e)}),t.close()});try{for await(const e of t)yield e}finally{this.activeTurn===s&&this.abort(),void 0!==this.emit&&(this.emit=void 0)}}async runTurn(o,i,r){const a=Date.now();let p;try{if(this.debug(o,"turn","start",a,{messages:this.messages.length,tools:Object.keys(this.tools).length,maxSteps:this.maxSteps,...void 0!==this.contextWindow?{contextWindow:this.contextWindow}:{},...void 0!==this.lastInputTokens?{lastInputTokens:this.lastInputTokens}:{}}),this.shouldAutoCompact()){this.debug(o,"compaction","auto requested before turn",a,{messages:this.messages.length,...void 0!==this.lastInputTokens?{lastInputTokens:this.lastInputTokens}:{}});try{await this.runCompaction(o,"auto",i)}catch(e){return void o.push({type:"error",stage:"plan",message:l(e)})}}if(i.aborted||i.abortController.signal.aborted)return void o.push({type:"error",stage:"execute",message:"aborted"});p=this.messages.length,this.messages.push({role:"user",content:r}),o.push({type:"message-start",messageId:e()}),this.debug(o,"model","calling streamText",a,{messages:this.messages.length,tools:Object.keys(this.tools).length,...void 0!==this.turnTimeoutMs?{timeoutMs:this.turnTimeoutMs}:{}});const g=s({model:this.model,system:u,messages:this.messages,tools:this.tools,stopWhen:t(this.maxSteps),abortSignal:i.abortController.signal,...this.providerOptions?{providerOptions:this.providerOptions}:{},...void 0!==this.turnTimeoutMs?{timeout:{totalMs:this.turnTimeoutMs}}:{}});this.debug(o,"model","streamText returned",a);let y,f,x,w,C,I,N="",M="",S=!1,W=0,R=!1;const A=this.scheduleDebug(o,"model","waiting for first stream event",a,{messages:this.messages.length});try{for await(const e of g.fullStream){switch(R||(R=!0,this.clearDebugTimer(A),this.debug(o,"model","first stream event",a,{part:e.type})),e.type){case"text-delta":if(!S){const t=M+e.text;if(b(t,r)){M=t;break}if(M.length>0){const t=M+e.text;M="",N+=t,o.push({type:"text-delta",delta:t});break}}N+=e.text,o.push({type:"text-delta",delta:e.text});break;case"tool-call":{S=!0,M="";const t=this.registry.resolve(e.toolName);o.push({type:"tool-call",toolCallId:e.toolCallId,agentName:t?.agentName??e.toolName,toolName:t?.toolName??e.toolName,input:e.input});break}case"tool-result":{const t=this.registry.resolve(e.toolName),s=n(e.output);o.push({type:"tool-result",toolCallId:e.toolCallId,agentName:t?.agentName??e.toolName,toolName:t?.toolName??e.toolName,output:e.output,isError:s}),i.aborted||i.abortController.signal.aborted||!s||(I=c(e.output));break}case"tool-error":{const t=this.registry.resolve(e.toolName),s=h(e.error);o.push({type:"tool-result",toolCallId:e.toolCallId,agentName:t?.agentName??e.toolName,toolName:t?.toolName??e.toolName,output:s,isError:!0}),i.aborted||i.abortController.signal.aborted||(I=c(e.error));break}case"finish-step":{const t=m(e.usage);f=T(f,t.inputTokens),W+=1,w=e.finishReason;const s=e.performance.outputTokensPerSecond??e.performance.effectiveOutputTokensPerSecond;void 0!==s&&Number.isFinite(s)&&(x=s),o.push({type:"step-finish",finishReason:e.finishReason,usage:t});break}case"finish":!S&&M.length>0&&(N+=M,o.push({type:"text-delta",delta:M}),M=""),y=m(e.totalUsage);break;case"error":d(e.error)&&(this.needsCompaction=!0),C=l(e.error),o.push({type:"error",stage:"execute",message:C});break;case"abort":i.aborted=!0}if(void 0!==I)break}}finally{this.clearDebugTimer(A)}if(this.debug(o,"model","fullStream finished",a,{textChars:N.length}),void 0!==C)return this.messages.splice(p),void(this.needsCompaction&&await this.recoverFromContextError(o,i));if(i.aborted||i.abortController.signal.aborted)return this.messages.splice(p),void o.push({type:"error",stage:"execute",message:"aborted"});if(void 0!==I){const e=I,t=0===N.length?e:`\n${e}`;o.push({type:"text-delta",delta:t}),this.messages.push({role:"assistant",content:e}),this.debug(o,"persist","persisting messages",a,{appendedMessages:this.messages.length-p}),this.onPersist?.(this.messages.slice(p)),this.debug(o,"persist","messages persisted",a,{totalMessages:this.messages.length}),y&&(this.sessionUsage=v(this.sessionUsage,y));const s=f??y?.inputTokens;return void 0!==s&&(this.lastInputTokens=s),void o.push({type:"message-finish",text:e,...y?{totalUsage:y}:{},sessionUsage:{...this.sessionUsage},...void 0!==f?{contextInputTokens:f}:{},...void 0!==x?{outputTokensPerSecond:x}:{}})}let D;this.debug(o,"model","awaiting response messages",a);const E=this.scheduleDebug(o,"model","still awaiting response messages",a);try{D=(await g.response).messages}catch(e){return this.clearDebugTimer(E),d(e)&&(this.needsCompaction=!0),this.messages.splice(p),o.push({type:"error",stage:"execute",message:l(e)}),void(this.needsCompaction&&await this.recoverFromContextError(o,i))}finally{this.clearDebugTimer(E)}this.debug(o,"model","response messages ready",a,{responseMessages:D.length});const U=k(D);this.messages.push(...U),this.debug(o,"persist","persisting messages",a,{appendedMessages:this.messages.length-p}),this.onPersist?.(this.messages.slice(p)),this.debug(o,"persist","messages persisted",a,{totalMessages:this.messages.length}),y&&(this.sessionUsage=v(this.sessionUsage,y));const O=f??y?.inputTokens;void 0!==O&&(this.lastInputTokens=O);const P=W>=this.maxSteps&&"tool-calls"===w;o.push({type:"message-finish",text:N,...y?{totalUsage:y}:{},sessionUsage:{...this.sessionUsage},...void 0!==f?{contextInputTokens:f}:{},...void 0!==x?{outputTokensPerSecond:x}:{},...P?{stoppedAtStepLimit:!0}:{}})}catch(e){void 0!==p&&this.messages.splice(p),d(e)&&(this.needsCompaction=!0),this.debug(o,"turn","error",a,{message:l(e)}),o.push({type:"error",stage:"execute",message:l(e)}),this.needsCompaction&&await this.recoverFromContextError(o,i)}finally{this.activeTurn===i&&(this.activeTurn=void 0),o.close()}}requestApproval(t){const s=e(),o=this.gate.request(s);return void 0===this.emit?(this.gate.resolve(s,{approved:!1,reason:"approval event could not be delivered"}),o):(this.emit({type:"confirmation-required",approvalId:s,agentName:t.agentName,toolName:t.toolName,input:t.input,...t.reason?{reason:t.reason}:{}}),o)}approve(e){return this.gate.resolve(e,{approved:!0})}reject(e,t){return this.gate.resolve(e,{approved:!1,...t?{reason:t}:{}})}getMessages(){return this.messages}getContextWindow(){return this.contextWindow}getSessionUsage(){return{...this.sessionUsage}}setProviderOptions(e){this.providerOptions=e}shouldAutoCompact(){const e=this.compaction;return void 0!==e&&e.enabled&&(this.needsCompaction||void 0!==this.contextWindow&&void 0!==this.lastInputTokens&&this.lastInputTokens/this.contextWindow>=e.threshold)}async runCompactionTurn(e,t,s){try{await this.runCompaction(e,s,t)}catch(t){e.push({type:"error",stage:"plan",message:l(t)})}finally{this.activeTurn===t&&(this.activeTurn=void 0),e.close()}}async runCompaction(e,t,s){const o=Date.now(),n=this.compaction,i=n?.strategy??"summarize";if(this.debug(e,"compaction","start",o,{reason:t,messages:this.messages.length}),e.push({type:"compaction-start",reason:t}),this.isTurnAborted(s))return void e.push({type:"error",stage:"plan",message:"aborted"});if(!n)return void e.push({type:"context-compacted",reason:t,strategy:i,removed:0,kept:this.messages.length});const a=this.lastInputTokens;let u=n.strategy;const p=s?.abortController.signal;let h;try{h=await r({messages:this.messages,strategy:u,keepRecentTurns:n.keepRecentTurns,keepRecentTokens:n.keepRecentTokens,model:this.model,...p?{abortSignal:p}:{}})}catch(t){if(this.isTurnAborted(s))return void e.push({type:"error",stage:"plan",message:"aborted"});if("summarize"!==u)throw t;this.debug(e,"compaction","summarize failed, fallback to truncate",o,{message:l(t)}),u="truncate",h=await r({messages:this.messages,strategy:u,keepRecentTurns:n.keepRecentTurns,keepRecentTokens:n.keepRecentTokens,model:this.model,...p?{abortSignal:p}:{}})}if(this.isTurnAborted(s))return void e.push({type:"error",stage:"plan",message:"aborted"});const c=h.removed>0||h.truncatedTools>0;c&&(this.onReplace?.(h.messages),this.messages.splice(0,this.messages.length,...h.messages),this.lastInputTokens=void 0,this.needsCompaction=!1),e.push({type:"context-compacted",reason:t,strategy:u,removed:h.removed,kept:h.kept,...h.truncatedTools>0?{truncatedTools:h.truncatedTools}:{},...void 0!==a?{beforeInputTokens:a}:{}}),this.debug(e,"compaction","finish",o,{reason:t,strategy:u,removed:h.removed,kept:h.kept,truncatedTools:h.truncatedTools,progressed:c})}async recoverFromContextError(e,t){if(this.compaction?.enabled)try{await this.runCompaction(e,"auto",t)}catch(t){e.push({type:"error",stage:"plan",message:l(t)})}}isTurnAborted(e){return!0===e?.aborted||!0===e?.abortController.signal.aborted}abort(){this.activeTurn&&(this.activeTurn.aborted=!0),this.gate.abortAll(),this.activeTurn?.abortController.abort()}debug(e,t,s,o,n){this.debugEvents&&e.push({type:"debug",stage:t,message:s,...void 0!==o?{elapsedMs:Date.now()-o}:{},...void 0!==n?{data:n}:{}})}scheduleDebug(e,t,s,o,n){if(this.debugEvents)return setTimeout(()=>{this.debug(e,t,s,o,n)},5e3)}clearDebugTimer(e){void 0!==e&&clearTimeout(e)}}

package/dist/engine/compactor.d.ts

@@ -1,12 +1,12 @@
 import { type ModelMessage } from "ai";
-import type { LanguageModelV3 } from "@ai-sdk/provider";
+import type { LanguageModelV4 } from "@ai-sdk/provider";
 import type { ContextCompactionStrategy } from "../types/events.ts";
 export interface CompactionInput {
     readonly messages: readonly ModelMessage[];
     readonly strategy: ContextCompactionStrategy;
     readonly keepRecentTurns: number;
     readonly keepRecentTokens: number;
-    readonly model: LanguageModelV3;
+    readonly model: LanguageModelV4;
     readonly abortSignal?: AbortSignal;
 }
 export interface CompactionResult {

package/dist/engine/context-window.js

@@ -1 +1 @@
-const w=[{match:"claude-fable-5",window:1e6},{match:"claude-mythos-5",window:1e6},{match:"claude-opus-4-8",window:1e6},{match:"claude-sonnet-4-6",window:1e6},{match:"claude-haiku-4-5",window:2e5},{match:"claude",window:2e5},{match:"gpt-5.5",window:105e4},{match:"gpt-5.4-mini",window:4e5},{match:"gpt-5.4-nano",window:4e5},{match:"gpt-5.4",window:105e4},{match:"gpt-5-mini",window:4e5},{match:"gpt-5-nano",window:4e5},{match:"gpt-5",window:4e5},{match:"gpt-4.1",window:128e3},{match:"gpt-4o",window:128e3},{match:"gpt-4-turbo",window:128e3},{match:"o3",window:2e5},{match:"o1-mini",window:128e3},{match:"o1",window:2e5},{match:"deepseek-v4",window:1e6},{match:"deepseek-chat",window:1e6},{match:"deepseek-reasoner",window:1e6},{match:"deepseek",window:128e3},{match:"qwen-long",window:1e7},{match:"qwen3-coder",window:1e6},{match:"qwen3.7-plus",window:1e6},{match:"qwen3.7-max",window:1e6},{match:"qwen3.6-plus",window:1e6},{match:"qwen3.5-plus",window:1e6},{match:"qwen3.5-flash",window:1e6},{match:"qwen-plus",window:1e6},{match:"qwen-flash",window:1e6},{match:"qwen3-max",window:262144},{match:"qwen3-vl-plus",window:262144},{match:"qwen-vl-plus",window:131072},{match:"qwen",window:131072},{match:"gemini",window:1e6},{match:"kimi",window:128e3},{match:"glm",window:128e3}];export function lookupContextWindow(e){const n=e.toLowerCase();return w.find(w=>n.includes(w.match))?.window}export function resolveContextWindow(w,e){return void 0!==e?e:lookupContextWindow(w)}
+const w=[{match:"claude-fable-5",window:1e6},{match:"claude-mythos-5",window:1e6},{match:"claude-opus-4-8",window:1e6},{match:"claude-opus-4-7",window:1e6},{match:"claude-opus-4-6",window:1e6},{match:"claude-sonnet-5",window:1e6},{match:"claude-sonnet-4-6",window:1e6},{match:"claude-haiku-4-5",window:2e5},{match:"claude",window:2e5},{match:"gpt-5.5",window:105e4},{match:"gpt-5.4-mini",window:4e5},{match:"gpt-5.4-nano",window:4e5},{match:"gpt-5.4",window:105e4},{match:"gpt-5-mini",window:4e5},{match:"gpt-5-nano",window:4e5},{match:"gpt-5",window:4e5},{match:"gpt-4.1",window:128e3},{match:"gpt-4o",window:128e3},{match:"gpt-4-turbo",window:128e3},{match:"o3",window:2e5},{match:"o1-mini",window:128e3},{match:"o1",window:2e5},{match:"deepseek-v4",window:1e6},{match:"deepseek-chat",window:1e6},{match:"deepseek-reasoner",window:1e6},{match:"deepseek",window:128e3},{match:"qwen-long",window:1e7},{match:"qwen3-coder",window:1e6},{match:"qwen3.7-plus",window:1e6},{match:"qwen3.7-max",window:1e6},{match:"qwen3.6-plus",window:1e6},{match:"qwen3.5-plus",window:1e6},{match:"qwen3.5-flash",window:1e6},{match:"qwen-plus",window:1e6},{match:"qwen-flash",window:1e6},{match:"qwen3-max",window:262144},{match:"qwen3-vl-plus",window:262144},{match:"qwen-vl-plus",window:131072},{match:"qwen",window:131072},{match:"gemini",window:1e6},{match:"kimi",window:128e3},{match:"glm",window:128e3}];export function lookupContextWindow(e){const n=e.toLowerCase();return w.find(w=>n.includes(w.match))?.window}export function resolveContextWindow(w,e){return void 0!==e?e:lookupContextWindow(w)}

package/dist/engine/conversation-engine.d.ts

@@ -1,4 +1,4 @@
-import type { LanguageModelV3, SharedV3ProviderOptions } from "@ai-sdk/provider";
+import type { LanguageModelV4, SharedV4ProviderOptions } from "@ai-sdk/provider";
 import { McpClientManager } from "@roll-agent/core/mcp/client-manager";
 import type { RollConfig } from "@roll-agent/core/config/schema";
 import type { RegisteredAgent } from "@roll-agent/core/types/agent";
@@ -10,13 +10,13 @@ export type EnsureAgentReady = (agent: RegisteredAgent, env: Readonly<Record<str
 export interface ConversationEngineOptions {
     readonly config: RollConfig;
     readonly agents?: readonly RegisteredAgent[];
-    readonly model?: LanguageModelV3;
+    readonly model?: LanguageModelV4;
     readonly sources?: readonly AgentToolSource[];
     readonly clientManager?: McpClientManager;
     readonly store?: ThreadStore;
     readonly policy?: ToolPolicy;
     readonly maxSteps?: number;
-    readonly providerOptions?: SharedV3ProviderOptions;
+    readonly providerOptions?: SharedV4ProviderOptions;
     readonly ensureAgentReady?: EnsureAgentReady;
     readonly debugEvents?: boolean;
     readonly onAgentBootstrapIssue?: (issue: AgentBootstrapIssue) => void;

package/dist/server/protocol.d.ts

@@ -70,23 +70,23 @@ export declare const approveParamsSchema: z.ZodObject<{
     sessionId: z.ZodString;
     approvalId: z.ZodString;
 }, "strip", z.ZodTypeAny, {
-    approvalId: string;
     sessionId: string;
-}, {
     approvalId: string;
+}, {
     sessionId: string;
+    approvalId: string;
 }>;
 export declare const rejectParamsSchema: z.ZodObject<{
     sessionId: z.ZodString;
     approvalId: z.ZodString;
     reason: z.ZodOptional<z.ZodString>;
 }, "strip", z.ZodTypeAny, {
-    approvalId: string;
     sessionId: string;
+    approvalId: string;
     reason?: string | undefined;
 }, {
-    approvalId: string;
     sessionId: string;
+    approvalId: string;
     reason?: string | undefined;
 }>;
 export declare const abortParamsSchema: z.ZodObject<{

package/dist/types/events.d.ts

@@ -3,6 +3,7 @@ export interface SessionTokenUsage {
     readonly outputTokens?: number;
     readonly totalTokens?: number;
     readonly cachedInputTokens?: number;
+    readonly cacheWriteTokens?: number;
     readonly reasoningTokens?: number;
 }
 export type SessionEventStage = "bootstrap" | "plan" | "execute";
@@ -50,6 +51,7 @@ export type SessionEvent = {
     readonly totalUsage?: SessionTokenUsage;
     readonly sessionUsage?: SessionTokenUsage;
     readonly contextInputTokens?: number;
+    readonly outputTokensPerSecond?: number;
     readonly stoppedAtStepLimit?: boolean;
 } | {
     readonly type: "compaction-start";

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@roll-agent/runtime",
-  "version": "0.2.0",
+  "version": "0.3.0",
   "type": "module",
   "repository": {
     "type": "git",
@@ -26,14 +26,14 @@
     "node": ">=22.6.0"
   },
   "dependencies": {
-    "@ai-sdk/provider": "^3.0.8",
+    "@ai-sdk/provider": "^4.0.1",
     "@modelcontextprotocol/sdk": "^1.12.0",
-    "ai": "^6.0.154",
+    "ai": "^7.0.9",
     "zod": "^3.25.76"
   },
   "devDependencies": {
     "@types/node": "^22.0.0",
-    "@roll-agent/core": "0.12.0"
+    "@roll-agent/core": "0.13.0"
   },
   "scripts": {
     "build": "rm -rf dist && tsc -p tsconfig.build.json && node ../../scripts/obfuscate.mjs",