fix: use native Pi identity and refresh provider catalogs

src/agent/models.ts

@@ -11,6 +11,14 @@ import { builtinProviders } from "@earendil-works/pi-ai/providers/all";
 import type { ThinkingLevel } from "@earendil-works/pi-agent-core";
 import { type ModelRole } from "../../config.ts";
 import { DEFAULT_THINKING_ORDER } from "../core/settings.ts";
+import packageMetadata from "../../package.json" with { type: "json" };
+
+/** App identity is immutable and shared by Models and every native Agent.
+ * Agent's default Pi attribution otherwise overrides the Models identity. */
+export const APP_IDENTITY = Object.freeze({
+	userAgent: Object.freeze({ custom: `learnharness/${packageMetadata.version}` }),
+	attribution: Object.freeze({ name: packageMetadata.name, url: packageMetadata.homepage }),
+});
 
 /**
  * The one engine factory every agent's model comes from (the pi-ai swap,
@@ -47,7 +55,7 @@ let collection: ReturnType<typeof createModels> | undefined;
  *  keyPresent. */
 export function piModels() {
 	if (!collection) {
-		collection = createModels();
+		collection = createModels({ identity: APP_IDENTITY });
 		for (const provider of builtinProviders()) collection.setProvider(provider);
 	}
 	return collection;

src/agent/subagents/pdf-reader.ts

@@ -1,3 +1,4 @@
+import { randomUUID } from "node:crypto";
 import type { AssistantMessage, Message, ToolResultMessage, Usage, Tool } from "@earendil-works/pi-ai";
 import { normalizeContext } from "@earendil-works/pi-ai";
 import { Type } from "typebox";
@@ -183,6 +184,7 @@ export async function runPdfReaderAgent(
 	// BYOK (owner, 2026-09-22): the level rides the injected engine — no
 	// factory pin to consult (engine.level is the resolved/clamped value).
 	const reasoning = engine.level;
+	const sessionId = randomUUID();
 	const prompt = [
 		`PDF path: ${pdfPath}`,
 		`Question (what the teacher needs from this document): ${input.question || "(no question — give a structural overview)"}`,
@@ -207,6 +209,7 @@ export async function runPdfReaderAgent(
 				INSTRUCTIONS,
 				messages,
 				toolDeclarations,
+				sessionId,
 				reasoning,
 				opts.abortSignal,
 				opts.chatId ? providerEventObserver(opts.chatId, "job:pdfReader") : undefined,
@@ -395,6 +398,7 @@ async function streamOnce(
 	systemPrompt: string,
 	messages: Message[],
 	tools: Tool[],
+	sessionId: string,
 	reasoning: ThinkingLevel | undefined,
 	abortSignal?: AbortSignal,
 	onProviderStreamEvent?: ProviderEventObserver,
@@ -403,6 +407,7 @@ async function streamOnce(
 	// and tool declarations fold into a leading system message.
 	// normalizeContext is the library's only Context→TranscriptContext entry.
 	const stream = engine.streamFn(engine.model, normalizeContext({ systemPrompt, messages, tools }), {
+		sessionId,
 		reasoning: reasoning && reasoning !== "off" ? reasoning : undefined,
 		signal: abortSignal,
 		onProviderStreamEvent: onProviderStreamEvent as never,

src/agent/subagents/pi-run.ts

@@ -1,6 +1,7 @@
 import { Agent, type AgentEvent, type AgentMessage, type AgentTool } from "@earendil-works/pi-agent-core";
 import type { AssistantMessage, ToolResultMessage, Usage } from "@earendil-works/pi-ai";
-import { piEngineFor, type PiEngine } from "../models.ts";
+import { randomUUID } from "node:crypto";
+import { APP_IDENTITY, piEngineFor, type PiEngine } from "../models.ts";
 import { providerEventObserver } from "../provider-events.ts";
 import type { ModelRole } from "../../../config.ts";
 
@@ -64,6 +65,9 @@ export async function runSubagent(tools: AgentTool<any, any>[], opts: SubagentRu
 	// Debug recorder (owner, 2026-10-04): raw provider events per job run.
 	const providerEvents = opts.chatId ? providerEventObserver(opts.chatId, `job:${opts.role}`) : undefined;
 	const agent = new Agent({
+		identity: APP_IDENTITY,
+		// One unsaved conversation per invocation, stable through every turn.
+		sessionId: randomUUID(),
 		initialState: {
 			systemPrompt: opts.systemPrompt,
 			model: engine.model,

src/agent/subagents/visual-maker.ts

@@ -1,3 +1,4 @@
+import { randomUUID } from "node:crypto";
 import type { AssistantMessage, Message, Tool, ToolResultMessage, Usage } from "@earendil-works/pi-ai";
 import { normalizeContext } from "@earendil-works/pi-ai";
 import { Type } from "typebox";
@@ -142,6 +143,7 @@ export async function runVisualMaker(input: VisualMakerInput, opts: VisualMakerO
 	// BYOK (owner, 2026-09-22): the level rides the injected engine — no
 	// factory pin to consult (engine.level is the resolved/clamped value).
 	const reasoning = engine.level;
+	const sessionId = randomUUID();
 	const prompt = [
 		"Brief for ONE precise figure (from the teacher, verbatim — it is everything you know):",
 		brief,
@@ -164,6 +166,7 @@ export async function runVisualMaker(input: VisualMakerInput, opts: VisualMakerO
 				INSTRUCTIONS,
 				messages,
 				toolDeclarations,
+				sessionId,
 				reasoning,
 				opts.abortSignal,
 				opts.chatId ? providerEventObserver(opts.chatId, "job:visualMaker") : undefined,
@@ -317,6 +320,7 @@ async function streamOnce(
 	systemPrompt: string,
 	messages: Message[],
 	tools: Tool[],
+	sessionId: string,
 	reasoning: ThinkingLevel | undefined,
 	abortSignal?: AbortSignal,
 	onProviderStreamEvent?: ProviderEventObserver,
@@ -325,6 +329,7 @@ async function streamOnce(
 	// and tool declarations fold into a leading system message.
 	// normalizeContext is the library's only Context→TranscriptContext entry.
 	const stream = engine.streamFn(engine.model, normalizeContext({ systemPrompt, messages, tools }), {
+		sessionId,
 		reasoning: reasoning && reasoning !== "off" ? reasoning : undefined,
 		signal: abortSignal,
 		onProviderStreamEvent: onProviderStreamEvent as never,

src/agent/teacher-agent.ts

@@ -4,7 +4,7 @@ import { invalidInputResult, questionResult } from "./tool-execute.ts";
 import { providerEventObserver } from "./provider-events.ts";
 import type { Usage } from "@earendil-works/pi-ai";
 import { PATHS } from "../../config.ts";
-import { piEngineFor, type PiEngine } from "./models.ts";
+import { APP_IDENTITY, piEngineFor, type PiEngine } from "./models.ts";
 import { effectivePromptSegments } from "./prompt-segments.ts";
 import {
 	askUserQuestionTool,
@@ -116,7 +116,9 @@ export interface AgentBuildOptions {
 	 *  fresh per turn, honoring enabled overrides (switch on + file
 	 *  exists); absent → the cached pure-source composition. */
 	promptSwitches?: Record<string, boolean>;
-	/** Debug: when set AND LH_PROVIDER_EVENTS=1, raw provider stream events
+	/** Native conversation identity: the existing persisted chat id, reused
+	 * across reconstructed prober, teacher, and practice Agents.
+	 * Debug: when set AND LH_PROVIDER_EVENTS=1, raw provider stream events
 	 *  (pre-normalization) append to data/debug/provider-events/
 	 *  <chatId>.jsonl via providerEventObserver — the GLM-truncation /
 	 *  DeepSeek-id incident layer. Absent or env-off → no observer. */
@@ -177,6 +179,8 @@ function baseAgentOptions(kind: AgentKind, opts: AgentBuildOptions, tools: Agent
 	// the normal no-observer path (tests, env-off).
 	const providerEvents = opts.chatId ? providerEventObserver(opts.chatId, kind) : undefined;
 	const agent = new Agent({
+		identity: APP_IDENTITY,
+		sessionId: opts.chatId,
 		initialState: {
 			systemPrompt,
 			model: engine.model,

src/core/auth-keys.ts

@@ -77,7 +77,7 @@ export interface AuthProviderInfo {
  *
  * EXCLUDED by owner spec — providers whose auth is not a plain key this
  * wave: github-copilot (OAuth), amazon-bedrock (AWS creds), google-vertex
- * (ADC), azure-openai-responses (needs a base URL), the cloudflare
+ * (ADC), azure (needs a base URL), the cloudflare
  * account-scoped pair. Anthropic keeps ANTHROPIC_API_KEY only — its
  * AUTH_TOKEN/OAUTH_TOKEN alternates stay env-only. openai-codex exposes
  * no env-key (naturally absent). Providers sharing one env var (moonshotai