feat(agent): AgentHarness streams through a required Models instance (phase 6)

AgentHarnessOptions.models is required; the harness stream path,
compaction, and branch summarization go through models.streamSimple()/
completeSimple() instead of the compat globals. getApiKeyAndHeaders
stays and wins per-field over provider-resolved auth, but is no longer
required: without it, requests resolve through provider auth.

compact()/generateSummary()/generateBranchSummary() take a Models
parameter; explicit apiKey becomes optional. StreamFn is redefined
structurally (Models.streamSimple satisfies it), dropping the compat
type dependency from agent types.

Harness tests build per-file Models collections with fauxProvider()
and unique provider ids instead of mutating the global api-registry.

packages/agent/src/harness/agent-harness.ts

@@ -1,10 +1,4 @@
-import {
-	type AssistantMessage,
-	type ImageContent,
-	type Model,
-	streamSimple,
-	type UserMessage,
-} from "@earendil-works/pi-ai/compat";
+import type { AssistantMessage, ImageContent, Model, Models, UserMessage } from "@earendil-works/pi-ai";
 import { runAgentLoop } from "../agent-loop.ts";
 import type {
 	AgentContext,
@@ -178,6 +172,7 @@ export class AgentHarness<
 > {
 	readonly env: ExecutionEnv;
 	private session: Session;
+	readonly models: Models;
 	private phase: AgentHarnessPhase = "idle";
 	private runAbortController?: AbortController;
 	private runPromise?: Promise<void>;
@@ -200,6 +195,7 @@ export class AgentHarness<
 	constructor(options: AgentHarnessOptions<TSkill, TPromptTemplate, TTool>) {
 		this.env = options.env;
 		this.session = options.session;
+		this.models = options.models;
 		this.resources = options.resources ?? {};
 		this.streamOptions = cloneStreamOptions(options.streamOptions);
 		this.systemPrompt = options.systemPrompt;
@@ -382,7 +378,7 @@ export class AgentHarness<
 				headers: mergeHeaders(turnState.streamOptions.headers, auth?.headers),
 			};
 			const requestOptions = await this.emitBeforeProviderRequest(model, turnState.sessionId, snapshotOptions);
-			return streamSimple(model, context, {
+			return this.models.streamSimple(model, context, {
 				cacheRetention: requestOptions.cacheRetention,
 				headers: requestOptions.headers,
 				maxRetries: requestOptions.maxRetries,
@@ -713,8 +709,8 @@ export class AgentHarness<
 		try {
 			const model = this.model;
 			if (!model) throw new AgentHarnessError("invalid_state", "No model set for compaction");
+			// Explicit auth wins; otherwise the request resolves through provider auth.
 			const auth = await this.getApiKeyAndHeaders?.(model);
-			if (!auth) throw new AgentHarnessError("auth", "No auth available for compaction");
 			const branchEntries = await this.session.getBranch();
 			const preparationResult = prepareCompaction(branchEntries, DEFAULT_COMPACTION_SETTINGS);
 			if (!preparationResult.ok) throw preparationResult.error;
@@ -733,9 +729,10 @@ export class AgentHarness<
 				? { ok: true as const, value: provided }
 				: await compact(
 						preparation,
+						this.models,
 						model,
-						auth.apiKey,
-						auth.headers,
+						auth?.apiKey,
+						auth?.headers,
 						customInstructions,
 						undefined,
 						this.thinkingLevel,
@@ -792,12 +789,13 @@ export class AgentHarness<
 			if (!summaryText && options?.summarize && entries.length > 0) {
 				const model = this.model;
 				if (!model) throw new AgentHarnessError("invalid_state", "No model set for branch summary");
+				// Explicit auth wins; otherwise the request resolves through provider auth.
 				const auth = await this.getApiKeyAndHeaders?.(model);
-				if (!auth) throw new AgentHarnessError("auth", "No auth available for branch summary");
 				const branchSummary = await generateBranchSummary(entries, {
+					models: this.models,
 					model,
-					apiKey: auth.apiKey,
-					headers: auth.headers,
+					apiKey: auth?.apiKey,
+					headers: auth?.headers,
 					signal: new AbortController().signal,
 					customInstructions: hookResult?.customInstructions ?? options?.customInstructions,
 					replaceInstructions: hookResult?.replaceInstructions ?? options?.replaceInstructions,

packages/agent/src/harness/compaction/branch-summarization.ts

@@ -1,5 +1,5 @@
-import type { Model } from "@earendil-works/pi-ai/compat";
-import { completeSimple } from "@earendil-works/pi-ai/compat";
+import type { Model, Models } from "@earendil-works/pi-ai";
+
 import type { AgentMessage } from "../../types.ts";
 import {
 	convertToLlm,
@@ -49,10 +49,12 @@ export interface CollectEntriesResult {
 
 /** Options for generating a branch summary. */
 export interface GenerateBranchSummaryOptions {
+	/** Provider collection the summarization request goes through. */
+	models: Models;
 	/** Model used for summarization. */
 	model: Model<any>;
-	/** API key forwarded to the provider. */
-	apiKey: string;
+	/** Explicit API key; wins over provider-resolved auth. */
+	apiKey?: string;
 	/** Optional request headers forwarded to the provider. */
 	headers?: Record<string, string>;
 	/** Abort signal for the summarization request. */
@@ -202,7 +204,16 @@ export async function generateBranchSummary(
 	entries: SessionTreeEntry[],
 	options: GenerateBranchSummaryOptions,
 ): Promise<Result<BranchSummaryResult, BranchSummaryError>> {
-	const { model, apiKey, headers, signal, customInstructions, replaceInstructions, reserveTokens = 16384 } = options;
+	const {
+		models,
+		model,
+		apiKey,
+		headers,
+		signal,
+		customInstructions,
+		replaceInstructions,
+		reserveTokens = 16384,
+	} = options;
 	const contextWindow = model.contextWindow || 128000;
 	const tokenBudget = contextWindow - reserveTokens;
 
@@ -230,7 +241,7 @@ export async function generateBranchSummary(
 			timestamp: Date.now(),
 		},
 	];
-	const response = await completeSimple(
+	const response = await models.completeSimple(
 		model,
 		{ systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages },
 		{ apiKey, headers, signal, maxTokens: 2048 },

packages/agent/src/harness/compaction/compaction.ts

@@ -1,5 +1,4 @@
-import type { AssistantMessage, ImageContent, Model, TextContent, Usage } from "@earendil-works/pi-ai/compat";
-import { completeSimple } from "@earendil-works/pi-ai/compat";
+import type { AssistantMessage, ImageContent, Model, Models, TextContent, Usage } from "@earendil-works/pi-ai";
 import type { AgentMessage, ThinkingLevel } from "../../types.ts";
 import {
 	convertToLlm,
@@ -455,9 +454,10 @@ Keep each section concise. Preserve exact file paths, function names, and error
 /** Generate or update a conversation summary for compaction. */
 export async function generateSummary(
 	currentMessages: AgentMessage[],
+	models: Models,
 	model: Model<any>,
 	reserveTokens: number,
-	apiKey: string,
+	apiKey?: string,
 	headers?: Record<string, string>,
 	signal?: AbortSignal,
 	customInstructions?: string,
@@ -493,7 +493,7 @@ export async function generateSummary(
 			? { maxTokens, signal, apiKey, headers, reasoning: thinkingLevel }
 			: { maxTokens, signal, apiKey, headers };
 
-	const response = await completeSimple(
+	const response = await models.completeSimple(
 		model,
 		{ systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages },
 		completionOptions,
@@ -626,8 +626,9 @@ export { serializeConversation } from "./utils.ts";
 /** Generate compaction summary data from prepared session history. */
 export async function compact(
 	preparation: CompactionPreparation,
+	models: Models,
 	model: Model<any>,
-	apiKey: string,
+	apiKey?: string,
 	headers?: Record<string, string>,
 	customInstructions?: string,
 	signal?: AbortSignal,
@@ -655,6 +656,7 @@ export async function compact(
 			messagesToSummarize.length > 0
 				? generateSummary(
 						messagesToSummarize,
+						models,
 						model,
 						settings.reserveTokens,
 						apiKey,
@@ -667,6 +669,7 @@ export async function compact(
 				: Promise.resolve(ok<string, CompactionError>("No prior history.")),
 			generateTurnPrefixSummary(
 				turnPrefixMessages,
+				models,
 				model,
 				settings.reserveTokens,
 				apiKey,
@@ -681,6 +684,7 @@ export async function compact(
 	} else {
 		const summaryResult = await generateSummary(
 			messagesToSummarize,
+			models,
 			model,
 			settings.reserveTokens,
 			apiKey,
@@ -706,9 +710,10 @@ export async function compact(
 }
 async function generateTurnPrefixSummary(
 	messages: AgentMessage[],
+	models: Models,
 	model: Model<any>,
 	reserveTokens: number,
-	apiKey: string,
+	apiKey?: string,
 	headers?: Record<string, string>,
 	signal?: AbortSignal,
 	thinkingLevel?: ThinkingLevel,
@@ -728,7 +733,7 @@ async function generateTurnPrefixSummary(
 		},
 	];
 
-	const response = await completeSimple(
+	const response = await models.completeSimple(
 		model,
 		{ systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages },
 		model.reasoning && thinkingLevel && thinkingLevel !== "off"

packages/agent/src/harness/types.ts

@@ -1,4 +1,4 @@
-import type { ImageContent, Model, SimpleStreamOptions, TextContent, Transport } from "@earendil-works/pi-ai";
+import type { ImageContent, Model, Models, SimpleStreamOptions, TextContent, Transport } from "@earendil-works/pi-ai";
 import type { AgentEvent, AgentMessage, AgentTool, QueueMode, ThinkingLevel } from "../index.ts";
 import type { Session } from "./session/session.ts";
 
@@ -802,6 +802,12 @@ export interface AgentHarnessOptions<
 > {
 	env: ExecutionEnv;
 	session: Session;
+	/**
+	 * Provider collection used for all model requests (turn streaming,
+	 * compaction, branch summarization). Auth resolves through the providers'
+	 * auth; explicit per-request values (`getApiKeyAndHeaders`) win per field.
+	 */
+	models: Models;
 	tools?: TTool[];
 	/**
 	 * Concrete resources available to explicit invocation methods and system-prompt callbacks.

packages/agent/src/types.ts

@@ -1,19 +1,22 @@
 import type {
+	Api,
 	AssistantMessage,
 	AssistantMessageEvent,
+	AssistantMessageEventStream,
+	Context,
 	ImageContent,
 	Message,
 	Model,
 	SimpleStreamOptions,
-	streamSimple,
 	TextContent,
 	Tool,
 	ToolResultMessage,
-} from "@earendil-works/pi-ai/compat";
+} from "@earendil-works/pi-ai";
 import type { Static, TSchema } from "typebox";
 
 /**
- * Stream function used by the agent loop.
+ * Stream function used by the agent loop. `Models.streamSimple` satisfies
+ * this shape.
  *
  * Contract:
  * - Must not throw or return a rejected promise for request/model/runtime failures.
@@ -22,8 +25,10 @@ import type { Static, TSchema } from "typebox";
  *   final AssistantMessage with stopReason "error" or "aborted" and errorMessage.
  */
 export type StreamFn = (
-	...args: Parameters<typeof streamSimple>
-) => ReturnType<typeof streamSimple> | Promise<ReturnType<typeof streamSimple>>;
+	model: Model<Api>,
+	context: Context,
+	options?: SimpleStreamOptions,
+) => AssistantMessageEventStream | Promise<AssistantMessageEventStream>;
 
 /**
  * Configuration for how tool calls from a single assistant message are executed.