@wahack/pi-coding-agent 15.11.0

package/src/tiny/models.ts

@@ -0,0 +1,242 @@
+/** Default session-title model: the online pi/smol path (no local download / on-device inference). */
+export const ONLINE_TINY_TITLE_MODEL_KEY = "online";
+/** Local model the `tiny-models` CLI downloads when none is named. Not the session-title default — that is {@link ONLINE_TINY_TITLE_MODEL_KEY}. */
+export const DEFAULT_TINY_TITLE_LOCAL_MODEL_KEY = "lfm2-700m";
+
+export interface TinyTitleLocalModelSpec {
+	key: string;
+	repo: string;
+	dtype: "q4";
+	label: string;
+	description: string;
+	contextNote: string;
+}
+
+export const TINY_TITLE_LOCAL_MODELS = [
+	{
+		key: "lfm2-350m",
+		repo: "onnx-community/LFM2-350M-ONNX",
+		dtype: "q4",
+		label: "LFM2 350M",
+		description: "Recommended local model; best speed/quality balance, about 212 MB cached.",
+		contextNote: "Best local default from the title-generation spike.",
+	},
+	{
+		key: "qwen3-0.6b",
+		repo: "onnx-community/Qwen3-0.6B-ONNX",
+		dtype: "q4",
+		label: "Qwen3 0.6B",
+		description: "Most robust local option; slower first load, about 500 MB cached.",
+		contextNote: "Use when title quality matters more than local startup cost.",
+	},
+	{
+		key: "gemma-270m",
+		repo: "onnx-community/gemma-3-270m-it-ONNX",
+		dtype: "q4",
+		label: "Gemma 270M",
+		description: "Smallest viable local option; lower quality, lowest cache footprint.",
+		contextNote: "Use on constrained machines that still need local titles.",
+	},
+	{
+		key: "qwen2.5-0.5b",
+		repo: "onnx-community/Qwen2.5-0.5B-Instruct",
+		dtype: "q4",
+		label: "Qwen2.5 0.5B",
+		description: "Balanced local fallback; moderate quality and cache footprint.",
+		contextNote: "Useful when Qwen3 is too heavy but Gemma quality is insufficient.",
+	},
+	{
+		key: "lfm2-700m",
+		repo: "onnx-community/LFM2-700M-ONNX",
+		dtype: "q4",
+		label: "LFM2 700M",
+		description: "Highest-quality local option; larger and slower than LFM2 350M.",
+		contextNote: "Use when local title quality is preferred over startup cost.",
+	},
+] as const satisfies readonly TinyTitleLocalModelSpec[];
+
+export const TINY_TITLE_MODEL_VALUES = [
+	ONLINE_TINY_TITLE_MODEL_KEY,
+	"lfm2-350m",
+	"qwen3-0.6b",
+	"gemma-270m",
+	"qwen2.5-0.5b",
+	"lfm2-700m",
+] as const;
+
+export type TinyTitleModelKey = (typeof TINY_TITLE_MODEL_VALUES)[number];
+export type TinyTitleLocalModelKey = (typeof TINY_TITLE_LOCAL_MODELS)[number]["key"];
+
+type MissingTinyTitleModelValue = Exclude<
+	typeof ONLINE_TINY_TITLE_MODEL_KEY | TinyTitleLocalModelKey,
+	TinyTitleModelKey
+>;
+type ExtraTinyTitleModelValue = Exclude<TinyTitleModelKey, typeof ONLINE_TINY_TITLE_MODEL_KEY | TinyTitleLocalModelKey>;
+const TINY_TITLE_MODEL_VALUES_MATCH_REGISTRY: MissingTinyTitleModelValue extends never
+	? ExtraTinyTitleModelValue extends never
+		? true
+		: never
+	: never = true;
+void TINY_TITLE_MODEL_VALUES_MATCH_REGISTRY;
+
+export const TINY_TITLE_MODEL_OPTIONS = [
+	{
+		value: ONLINE_TINY_TITLE_MODEL_KEY,
+		label: "Online (pi/smol)",
+		description: "Current online title generation path; no local model download or on-device inference.",
+	},
+	...TINY_TITLE_LOCAL_MODELS.map(model => ({
+		value: model.key,
+		label: model.label,
+		description: model.description,
+	})),
+] satisfies ReadonlyArray<{ value: TinyTitleModelKey; label: string; description: string }>;
+
+export function isTinyTitleLocalModelKey(value: string): value is TinyTitleLocalModelKey {
+	return TINY_TITLE_LOCAL_MODELS.some(model => model.key === value);
+}
+
+export function getTinyTitleModelSpec(key: TinyTitleLocalModelKey): (typeof TINY_TITLE_LOCAL_MODELS)[number] {
+	const spec = TINY_TITLE_LOCAL_MODELS.find(model => model.key === key);
+	if (!spec) throw new Error(`Unknown tiny title model: ${key}`);
+	return spec;
+}
+
+/** Default memory model: the online path (the configured smol / remote LLM; no local download). */
+export const ONLINE_MEMORY_MODEL_KEY = "online";
+/** Recommended local model for memory tasks when none is named. */
+export const DEFAULT_MEMORY_LOCAL_MODEL_KEY = "qwen3-1.7b";
+
+/**
+ * Local models for Mnemopi memory tasks (fact extraction + consolidation).
+ * These are larger (1B-1.7B) than the title models: structured extraction and
+ * faithful summarization need more capacity than 3-6 word titles. All q4.
+ * Ranking/recipe rationale lives in docs/local-models.md.
+ */
+export const TINY_MEMORY_LOCAL_MODELS = [
+	{
+		key: "qwen3-1.7b",
+		repo: "onnx-community/Qwen3-1.7B-ONNX",
+		dtype: "q4",
+		label: "Qwen3 1.7B",
+		description:
+			"Recommended; most disciplined extraction (ignores chit-chat), good consolidation, about 1.1 GB cached.",
+		contextNote: "Best single-model pick for memory from the local experiment.",
+	},
+	{
+		key: "gemma-3-1b",
+		repo: "onnx-community/gemma-3-1b-it-ONNX",
+		dtype: "q4",
+		label: "Gemma 3 1B",
+		description: "Best consolidation/dedup; lighter footprint, but leaks small talk during extraction.",
+		contextNote: "Use when consolidation quality and size matter most.",
+	},
+	{
+		key: "qwen2.5-1.5b",
+		repo: "onnx-community/Qwen2.5-1.5B-Instruct",
+		dtype: "q4",
+		label: "Qwen2.5 1.5B",
+		description: "Best extraction granularity (atomic facts); weaker consolidation.",
+		contextNote: "Use when fine-grained, deduplicatable facts matter more than summaries.",
+	},
+	{
+		key: "lfm2-1.2b",
+		repo: "onnx-community/LFM2-1.2B-ONNX",
+		dtype: "q4",
+		label: "LFM2 1.2B",
+		description: "Fastest load; solid all-rounder, slightly noisier extraction labels.",
+		contextNote: "Use when local startup cost is the priority.",
+	},
+] as const satisfies readonly TinyTitleLocalModelSpec[];
+
+export const TINY_MEMORY_MODEL_VALUES = [
+	ONLINE_MEMORY_MODEL_KEY,
+	"qwen3-1.7b",
+	"gemma-3-1b",
+	"qwen2.5-1.5b",
+	"lfm2-1.2b",
+] as const;
+
+export type TinyMemoryModelKey = (typeof TINY_MEMORY_MODEL_VALUES)[number];
+export type TinyMemoryLocalModelKey = (typeof TINY_MEMORY_LOCAL_MODELS)[number]["key"];
+
+type MissingTinyMemoryModelValue = Exclude<
+	typeof ONLINE_MEMORY_MODEL_KEY | TinyMemoryLocalModelKey,
+	TinyMemoryModelKey
+>;
+type ExtraTinyMemoryModelValue = Exclude<TinyMemoryModelKey, typeof ONLINE_MEMORY_MODEL_KEY | TinyMemoryLocalModelKey>;
+const TINY_MEMORY_MODEL_VALUES_MATCH_REGISTRY: MissingTinyMemoryModelValue extends never
+	? ExtraTinyMemoryModelValue extends never
+		? true
+		: never
+	: never = true;
+void TINY_MEMORY_MODEL_VALUES_MATCH_REGISTRY;
+
+export const TINY_MEMORY_MODEL_OPTIONS = [
+	{
+		value: ONLINE_MEMORY_MODEL_KEY,
+		label: "Online (smol/remote)",
+		description:
+			"Use the configured Mnemopi LLM mode (smol or remote); no local model download or on-device inference.",
+	},
+	...TINY_MEMORY_LOCAL_MODELS.map(model => ({
+		value: model.key,
+		label: model.label,
+		description: model.description,
+	})),
+] satisfies ReadonlyArray<{ value: TinyMemoryModelKey; label: string; description: string }>;
+
+export function isTinyMemoryLocalModelKey(value: string): value is TinyMemoryLocalModelKey {
+	return TINY_MEMORY_LOCAL_MODELS.some(model => model.key === value);
+}
+
+export function getTinyMemoryModelSpec(key: TinyMemoryLocalModelKey): (typeof TINY_MEMORY_LOCAL_MODELS)[number] {
+	const spec = TINY_MEMORY_LOCAL_MODELS.find(model => model.key === key);
+	if (!spec) throw new Error(`Unknown tiny memory model: ${key}`);
+	return spec;
+}
+
+/** Any local model key (title or memory), used by the shared inference worker. */
+export type TinyLocalModelKey = TinyTitleLocalModelKey | TinyMemoryLocalModelKey;
+
+/** Resolve a local model spec by key across both the title and memory registries. */
+export function getTinyLocalModelSpec(key: string): TinyTitleLocalModelSpec | undefined {
+	return (
+		TINY_TITLE_LOCAL_MODELS.find(model => model.key === key) ??
+		TINY_MEMORY_LOCAL_MODELS.find(model => model.key === key)
+	);
+}
+
+export function isTinyLocalModelKey(value: string): value is TinyLocalModelKey {
+	return getTinyLocalModelSpec(value) !== undefined;
+}
+
+/** Combined local model registry (title + memory) for the shared tiny-models CLI. */
+export const TINY_LOCAL_MODELS = [
+	...TINY_TITLE_LOCAL_MODELS,
+	...TINY_MEMORY_LOCAL_MODELS,
+] as const satisfies readonly TinyTitleLocalModelSpec[];
+
+/**
+ * Difficulty-classifier model for the `auto` thinking level. Defaults to the
+ * online smol path; the local options reuse the memory-model registry because
+ * the shared worker's `complete()` only accepts memory local keys, and the
+ * 1B+ memory models classify coding difficulty far more reliably than the
+ * sub-1B title models.
+ */
+export const ONLINE_AUTO_THINKING_MODEL_KEY = ONLINE_MEMORY_MODEL_KEY;
+export const AUTO_THINKING_MODEL_VALUES = TINY_MEMORY_MODEL_VALUES;
+export type AutoThinkingModelKey = TinyMemoryModelKey;
+
+export const AUTO_THINKING_MODEL_OPTIONS = [
+	{
+		value: ONLINE_AUTO_THINKING_MODEL_KEY,
+		label: "Online (smol)",
+		description: "Classify prompt difficulty with the online smol model; no local download or on-device inference.",
+	},
+	...TINY_MEMORY_LOCAL_MODELS.map(model => ({
+		value: model.key,
+		label: model.label,
+		description: model.description,
+	})),
+] satisfies ReadonlyArray<{ value: AutoThinkingModelKey; label: string; description: string }>;

package/src/tiny/text.ts

@@ -0,0 +1,165 @@
+export const MAX_TITLE_INPUT_CHARS = 2000;
+
+/**
+ * Minimum length of code-stripped input below which we fall back to the
+ * original message. Guards against messages that are (almost) entirely a code
+ * block — stripping would otherwise leave the model nothing to title from.
+ */
+const MIN_STRIPPED_TITLE_CHARS = 12;
+/** Matches a fenced code block (3+ backticks), including an unterminated trailing fence. */
+const FENCED_CODE_BLOCK = /```+[\s\S]*?(?:```+|$)/g;
+
+export function truncateTitleInput(message: string): string {
+	return message.length > MAX_TITLE_INPUT_CHARS ? `${message.slice(0, MAX_TITLE_INPUT_CHARS)}…` : message;
+}
+
+/**
+ * Strip fenced code blocks from a message before titling.
+ *
+ * Small title models latch onto literal text inside code blocks — e.g. a pasted
+ * UI mockup containing "Welcome to Claude Code v2.1.158" yields that string as
+ * the title instead of the surrounding intent. Removing fenced blocks leaves the
+ * prose that actually describes the task. Inline code (single backticks) is kept
+ * — it is short, high-signal context like `/login`.
+ *
+ * Falls back to the original message when stripping leaves too little to title
+ * (a message that is essentially just a code block).
+ */
+export function stripCodeBlocks(message: string): string {
+	const cleaned = message
+		.replace(FENCED_CODE_BLOCK, " ")
+		.replace(/[ \t]+/g, " ")
+		.replace(/\n{3,}/g, "\n\n")
+		.trim();
+	return cleaned.length >= MIN_STRIPPED_TITLE_CHARS ? cleaned : message;
+}
+
+/** Prepare a raw user message for titling: drop code blocks, then bound length. */
+export function prepareTitleInput(message: string): string {
+	return truncateTitleInput(stripCodeBlocks(message));
+}
+
+export function formatTitleUserMessage(message: string): string {
+	return `<user-message>\n${prepareTitleInput(message)}\n</user-message>`;
+}
+
+/**
+ * Greeting / acknowledgement / filler tokens. A first user message composed
+ * entirely of these (or of bare numbers / punctuation / emoji) carries no
+ * concrete task, so titling is deferred to a later message instead of latching
+ * onto "hi". See {@link isLowSignalTitleInput}.
+ */
+const FILLER_TITLE_TOKENS = new Set<string>([
+	// greetings
+	"hi",
+	"hii",
+	"hiii",
+	"hiya",
+	"hey",
+	"heya",
+	"hello",
+	"helo",
+	"hullo",
+	"yo",
+	"ya",
+	"sup",
+	"wassup",
+	"whatsup",
+	"howdy",
+	"greetings",
+	"hola",
+	"ciao",
+	"aloha",
+	"gm",
+	"gn",
+	"good",
+	"morning",
+	"afternoon",
+	"evening",
+	"night",
+	"day",
+	// politeness / acknowledgement
+	"thanks",
+	"thank",
+	"thx",
+	"ty",
+	"tysm",
+	"cheers",
+	"please",
+	"pls",
+	"plz",
+	"ok",
+	"okay",
+	"okey",
+	"k",
+	"kk",
+	"yep",
+	"yes",
+	"yeah",
+	"yup",
+	"nope",
+	"no",
+	"nah",
+	"sure",
+	"cool",
+	"nice",
+	"great",
+	"awesome",
+	"perfect",
+	"lol",
+	"lmao",
+	"haha",
+	"hehe",
+	// poking the agent / fillers
+	"test",
+	"tests",
+	"testing",
+	"ping",
+	"pong",
+	"there",
+	"you",
+	"u",
+	"hmm",
+	"hmmm",
+	"um",
+	"uh",
+	"so",
+	"well",
+	"anyway",
+]);
+
+const TITLE_WORD = /[\p{L}\p{N}]+/gu;
+
+/**
+ * True when a first user message is too low-signal to title (greeting, ack,
+ * bare number, or empty once code/punctuation/emoji are stripped).
+ *
+ * Deterministic pre-filter: the default tiny title model (~350M local) cannot
+ * reliably follow a "respond with none" instruction and tends to hallucinate a
+ * title for trivial input, so we never ask it — the caller defers titling to
+ * the next message instead.
+ */
+export function isLowSignalTitleInput(message: string): boolean {
+	const tokens = stripCodeBlocks(message).toLowerCase().match(TITLE_WORD);
+	if (!tokens) return true;
+	return tokens.every(token => FILLER_TITLE_TOKENS.has(token) || /^\d+$/.test(token));
+}
+
+/**
+ * Sentinel a capable title model may emit when a message carries no concrete
+ * task. Treated as "no title yet" so the caller can defer titling. Backstop for
+ * the deterministic {@link isLowSignalTitleInput} filter; kept in sync with the
+ * `none` instruction in `prompts/system/title-system.md`.
+ */
+export const NO_TITLE_SENTINEL = "none";
+
+export function normalizeGeneratedTitle(value: string | null | undefined): string | null {
+	const firstLine = value?.trim().split(/\r?\n/, 1)[0]?.trim();
+	if (!firstLine) return null;
+	const title = firstLine
+		.replace(/^["']|["']$/g, "")
+		.replace(/[.!?]$/, "")
+		.trim();
+	if (!title || title.toLowerCase() === NO_TITLE_SENTINEL) return null;
+	return title;
+}