feat: add Russian localization and independent STT languages

src/agent/subagents/demo-builder.ts

@@ -1,3 +1,4 @@
+import { learningLanguageInstruction, type Language } from "../../core/i18n.ts";
 import { Type } from "typebox";
 import type { AgentTool } from "@earendil-works/pi-agent-core";
 import type { Usage } from "@earendil-works/pi-ai";
@@ -31,6 +32,7 @@ export interface DemoBuilderInput {
 }
 
 export interface DemoBuilderOptions {
+	language?: Language;
 	/** Test seam: scripted engine; production uses the demoBuilder role. */
 	engine?: PiEngine;
 	headless?: HeadlessApi;
@@ -207,7 +209,7 @@ export async function runDemoBuilder(input: DemoBuilderInput, opts: DemoBuilderO
 		const result = await runSubagent(Object.values(createDemoBuilderTools(state, headless, opts.onActivity)), {
 			role: "demoBuilder",
 			engine: opts.engine,
-			systemPrompt: INSTRUCTIONS,
+			systemPrompt: `${INSTRUCTIONS}\n\n${learningLanguageInstruction(opts.language ?? "en")}`,
 			prompt,
 			maxTurns: opts.maxSteps ?? 24,
 			abortSignal: opts.abortSignal,

src/agent/subagents/extraction.ts

@@ -1,3 +1,4 @@
+import { learningLanguageInstruction, type Language } from "../../core/i18n.ts";
 import * as crypto from "node:crypto";
 import type { Usage } from "@earendil-works/pi-ai";
 import type { AgentTool } from "@earendil-works/pi-agent-core";
@@ -471,6 +472,7 @@ function runExtractor(
 	batchIds: ReadonlySet<string>,
 	signal: AbortSignal | undefined,
 	onStep: (usage: Usage, model: string, step: number, finishReason: string, tools: string) => void,
+	language: Language = "en",
 ): Promise<SessionFindings> {
 	// Chains in (the wire), chains out (the store) — no translation seam
 	// since chains-primary. Validation runs at the submit seam (error-back)
@@ -486,7 +488,7 @@ function runExtractor(
 		parameters: sessionFindingsSchema,
 		fill: fillFindings,
 		validate: (value) => validateSessionAnchors(value, batchIds) ?? validateFindingsChains(value, existingIds),
-		systemPrompt: EXTRACTOR_INSTRUCTIONS,
+		systemPrompt: `${EXTRACTOR_INSTRUCTIONS}\n\n${learningLanguageInstruction(language)}`,
 		prompt,
 		engine,
 		signal,
@@ -507,6 +509,7 @@ function runNotesExtractor(
 	existingIds: ReadonlySet<string>,
 	signal: AbortSignal | undefined,
 	onStep: (usage: Usage, model: string, step: number, finishReason: string, tools: string) => void,
+	language: Language = "en",
 ): Promise<NotesFindings> {
 	return runSubmitAgent<FindingsSubmit>({
 		role: "extractor",
@@ -517,7 +520,7 @@ function runNotesExtractor(
 		parameters: findingsSchema,
 		fill: fillNotesFindings,
 		validate: (value) => validateFindingsChains(value, existingIds),
-		systemPrompt: NOTES_EXTRACTOR_INSTRUCTIONS,
+		systemPrompt: `${NOTES_EXTRACTOR_INSTRUCTIONS}\n\n${learningLanguageInstruction(language)}`,
 		prompt,
 		engine,
 		signal,
@@ -725,6 +728,7 @@ export interface ExtractionModels {
 }
 
 export interface ExtractionPipelineDeps extends GraphDeps, ExtractionModels {
+	languageForSource?: (sourceId: string) => Language;
 	/** The chunk budget — the transcript-token share of the extractor's
 	 *  context; batches fill to it (the trigger fires on the same number,
 	 *  read fresh per turn by the composition root). OPTIONAL with the
@@ -735,6 +739,7 @@ export interface ExtractionPipelineDeps extends GraphDeps, ExtractionModels {
 }
 
 export interface NotesImportInput {
+	language?: Language;
 	/** The source's file path — its identity (deterministic src id) AND its
 	 *  provenance label in the detail card. */
 	filePath: string;
@@ -940,6 +945,7 @@ export function createExtractionPipeline(deps: ExtractionPipelineDeps): Extracti
 				recordStep(jobKey, "extractor", callUsage(usage), model);
 				console.info(`[extraction] extractor step ${step}: ${finishReason}, tools: ${tools}`);
 			},
+			deps.languageForSource?.(sourceId) ?? chatStore.getChat(sourceId)?.language ?? "en",
 		);
 		console.info(`[extraction ${jobKey}] extractor done in ${Math.round(performance.now() - t0)}ms`);
 		return findings;
@@ -1262,7 +1268,7 @@ export function createExtractionPipeline(deps: ExtractionPipelineDeps): Extracti
 
 	async function notesExtract(
 		jobKey: string,
-		input: { id: string; filePath: string; text: string; part?: { i: number; n: number } },
+		input: { id: string; filePath: string; text: string; language?: Language; part?: { i: number; n: number } },
 	): Promise<NotesFindings> {
 		// jobKey (per-chunk jobs, owner 2026-09-28): the registry key —
 		// "src#i" for chunk jobs — so the abort signal reaches the call.
@@ -1290,6 +1296,7 @@ export function createExtractionPipeline(deps: ExtractionPipelineDeps): Extracti
 				recordStep(jobKey, "extractor", callUsage(usage), model);
 				console.info(`[extraction] extractor step ${step}: ${finishReason}, tools: ${tools}`);
 			},
+			input.language ?? deps.languageForSource?.(input.id) ?? "en",
 		);
 		console.info(`[extraction ${jobKey}] notes extractor done in ${Math.round(performance.now() - t0)}ms`);
 		return findings;
@@ -1298,7 +1305,7 @@ export function createExtractionPipeline(deps: ExtractionPipelineDeps): Extracti
 	// The notes auditor is the SAME identity-linker as the chat one (one
 	// schema now that attribution is gone) — audit() above, verbatim.
 
-	async function runNotesExtraction(input: { id: string; filePath: string; text: string; importedAt: string }): Promise<EmissionStats> {
+	async function runNotesExtraction(input: { id: string; filePath: string; text: string; importedAt: string; language?: Language }): Promise<EmissionStats> {
 		const { id } = input;
 		// importNotes owns the source exclusion, including metadata and cleanup.
 		const t0 = performance.now();

src/agent/subagents/pdf-reader.ts

@@ -1,3 +1,4 @@
+import { learningLanguageInstruction, type Language } from "../../core/i18n.ts";
 import { randomUUID } from "node:crypto";
 import type { AssistantMessage, Message, ToolResultMessage, Usage, Tool } from "@earendil-works/pi-ai";
 import { normalizeContext } from "@earendil-works/pi-ai";
@@ -62,6 +63,7 @@ const RENDER_BATCH_CAP = 6;
 const EXTRACT_CAP = 24_000;
 
 export interface PdfReaderOptions {
+	language?: Language;
 	engine?: PiEngine;
 	abortSignal?: AbortSignal;
 	onUsage?: (usage: Usage) => void;
@@ -206,7 +208,7 @@ export async function runPdfReaderAgent(
 		for (let iteration = 0; iteration < maxIterations; iteration++) {
 			const result = await streamOnce(
 				engine,
-				INSTRUCTIONS,
+				`${INSTRUCTIONS}\n\n${learningLanguageInstruction(opts.language ?? "en")}`,
 				messages,
 				toolDeclarations,
 				sessionId,

src/agent/subagents/researcher.ts

@@ -1,3 +1,4 @@
+import { learningLanguageInstruction, type Language } from "../../core/i18n.ts";
 import type { AgentTool } from "@earendil-works/pi-agent-core";
 import type { Usage } from "@earendil-works/pi-ai";
 import { createWebTools } from "../tools/web.ts";
@@ -68,6 +69,7 @@ Numbered findings with inline source citations:
 What couldn't be answered. Suggested next steps.`;
 
 export interface ResearcherRunOptions {
+	language?: Language;
 	engine?: PiEngine;
 	/** The search seam (search port wave, 2026-09-20): resolves the LIVE
 	 *  provider per call, null = honestly unavailable (the web_search tool
@@ -98,7 +100,7 @@ export async function runResearcherAgent(prompt: string, opts: ResearcherRunOpti
 		role: "researcher",
 		prompt,
 		engine: opts.engine,
-		systemPrompt: INSTRUCTIONS,
+		systemPrompt: `${INSTRUCTIONS}\n\n${learningLanguageInstruction(opts.language ?? "en")}`,
 		maxTurns: 12,
 		abortSignal: opts.abortSignal,
 		onUsage: opts.onUsage,

src/agent/subagents/runner.ts

@@ -1,3 +1,4 @@
+import type { Language } from "../../core/i18n.ts";
 import type { PiEngine } from "../models.ts";
 import type { SearchProvider } from "../../core/search.ts";
 import {
@@ -64,6 +65,7 @@ export const pdfModelOverride: { current?: PiEngine } = {};
 type JobToolInput = ResearchToolInput | DemoToolInput | VisualToolInput | PdfToolInput;
 
 export interface JobRunnerDeps {
+	languageForChat?: (chatId: string) => Language;
 	chatStore: ChatStore;
 	jobStore: JobStore;
 	/**
@@ -364,6 +366,7 @@ export function createJobRunner(deps: JobRunnerDeps): JobRunner {
 			jobStore.updateJob(jobId, { activity: "researching" });
 			const brief = await runResearcherAgent(prompt, {
 				engine: model,
+					language: deps.languageForChat?.(job.chatId) ?? chatStore.getChat(job.chatId)?.language ?? "en",
 				search: deps.searchProvider,
 				pdfDir,
 				abortSignal: loop.abortSignal,
@@ -417,6 +420,7 @@ export function createJobRunner(deps: JobRunnerDeps): JobRunner {
 				{ title, brief },
 				{
 					engine: model,
+					language: deps.languageForChat?.(job.chatId) ?? chatStore.getChat(job.chatId)?.language ?? "en",
 					abortSignal: loop.abortSignal,
 					chatId: loop.chatId,
 					onUsage: (usage) => loop.onUsage(piUsageToStep(usage)),
@@ -453,6 +457,7 @@ export function createJobRunner(deps: JobRunnerDeps): JobRunner {
 				{ brief },
 				{
 					engine: model,
+					language: deps.languageForChat?.(job.chatId) ?? chatStore.getChat(job.chatId)?.language ?? "en",
 					abortSignal: loop.abortSignal,
 					chatId: loop.chatId,
 					onUsage: (usage) => loop.onUsage(piUsageToStep(usage)),
@@ -492,6 +497,7 @@ export function createJobRunner(deps: JobRunnerDeps): JobRunner {
 				{ path, question },
 			{
 				engine: model,
+					language: deps.languageForChat?.(job.chatId) ?? chatStore.getChat(job.chatId)?.language ?? "en",
 				abortSignal: loop.abortSignal,
 				chatId: loop.chatId,
 				onUsage: (usage) => loop.onUsage(piUsageToStep(usage)),

src/agent/subagents/visual-maker.ts

@@ -1,3 +1,4 @@
+import { learningLanguageInstruction, type Language } from "../../core/i18n.ts";
 import { randomUUID } from "node:crypto";
 import type { AssistantMessage, Message, Tool, ToolResultMessage, Usage } from "@earendil-works/pi-ai";
 import { normalizeContext } from "@earendil-works/pi-ai";
@@ -35,6 +36,7 @@ export interface VisualMakerInput {
 }
 
 export interface VisualMakerOptions {
+	language?: Language;
 	/** Test seam: scripted engine; production uses the visualMaker role. */
 	engine?: PiEngine;
 	headless?: HeadlessApi;
@@ -163,7 +165,7 @@ export async function runVisualMaker(input: VisualMakerInput, opts: VisualMakerO
 			// loop owns tool execution and the image-in-the-loop injection).
 			const result = await streamOnce(
 				engine,
-				INSTRUCTIONS,
+				`${INSTRUCTIONS}\n\n${learningLanguageInstruction(opts.language ?? "en")}`,
 				messages,
 				toolDeclarations,
 				sessionId,