@oh-my-pi/pi-coding-agent 16.1.22 → 16.2.0

package/src/cli/usage-cli.ts

@@ -326,7 +326,26 @@ function formatAccountHeader(
 	const planType = report.metadata?.planType;
 	if (typeof planType === "string" && planType) header += chalk.dim(` · plan: ${planType}`);
 	const savedResets = report.resetCredits?.availableCount ?? 0;
-	if (savedResets > 0) header += chalk.cyan(` · ✦ ${savedResets} saved reset${savedResets === 1 ? "" : "s"}`);
+	if (savedResets > 0) {
+		header += chalk.cyan(` · ✦ ${savedResets} saved reset${savedResets === 1 ? "" : "s"}`);
+		const credits = report.resetCredits?.credits;
+		if (credits) {
+			const expiries = credits
+				.filter(c => c.expiresAt)
+				.map(c => ({ date: c.expiresAt!, ms: Date.parse(c.expiresAt!) }))
+				.filter(c => !Number.isNaN(c.ms))
+				.sort((a, b) => a.ms - b.ms);
+			const upcoming = expiries.find(c => c.ms > nowMs);
+			if (upcoming) {
+				header += chalk.dim(
+					` · soonest expires in ${formatDuration(upcoming.ms - nowMs)} (${upcoming.date.slice(0, 10)})`,
+				);
+			} else {
+				const lastExpired = expiries.at(-1);
+				if (lastExpired) header += chalk.dim(` · expired (${lastExpired.date.slice(0, 10)})`);
+			}
+		}
+	}
 	if (report.fetchedAt && nowMs - report.fetchedAt > 90_000) {
 		header += chalk.dim(` · fetched ${formatDuration(nowMs - report.fetchedAt)} ago`);
 	}

package/src/cli/web-search-cli.ts

@@ -120,7 +120,7 @@ ${chalk.bold("Arguments:")}
 
 ${chalk.bold("Options:")}
   --provider <name>   Provider: ${PROVIDERS.join(", ")}
-  --recency <value>   Recency filter (Brave/Perplexity): ${RECENCY_OPTIONS.join(", ")}
+  --recency <value>   Recency filter (when supported): ${RECENCY_OPTIONS.join(", ")}
   -l, --limit <n>     Max results to return
   --compact           Render condensed output
   -h, --help          Show this help

package/src/cli-commands.ts

@@ -23,6 +23,7 @@ export const commands: CommandEntry[] = [
 	{ name: "__complete", load: () => import("./commands/complete").then(m => m.default) },
 	{ name: "config", load: () => import("./commands/config").then(m => m.default) },
 	{ name: "dry-balance", load: () => import("./commands/dry-balance").then(m => m.default) },
+	{ name: "gc", load: () => import("./commands/gc").then(m => m.default) },
 	{ name: "grep", load: () => import("./commands/grep").then(m => m.default) },
 	{ name: "gallery", load: () => import("./commands/gallery").then(m => m.default) },
 	{ name: "grievances", load: () => import("./commands/grievances").then(m => m.default) },

package/src/collab/display-name.ts

@@ -0,0 +1,13 @@
+import * as os from "node:os";
+import type { InteractiveModeContext } from "../modes/types";
+
+/** Display name for this process's user in collab sessions. */
+export function collabDisplayName(ctx: InteractiveModeContext): string {
+	const configured = (ctx.settings.get("collab.displayName") ?? "").trim();
+	if (configured) return configured;
+	try {
+		return os.userInfo().username;
+	} catch {
+		return "anonymous";
+	}
+}

package/src/collab/guest.ts

@@ -24,7 +24,7 @@ import type { SessionEntry } from "../session/session-entries";
 import { shouldDisableReasoning, toReasoningEffort } from "../thinking";
 import { setSessionTerminalTitle } from "../utils/title-generator";
 import { importRoomKey } from "./crypto";
-import { collabDisplayName } from "./host";
+import { collabDisplayName } from "./display-name";
 import {
 	type AgentSnapshot,
 	COLLAB_PROTO,
@@ -276,6 +276,7 @@ export class CollabGuestLink {
 		}
 
 		this.#ctx.collabGuest = this;
+		this.#ctx.syncRunningSubagentBadge();
 	}
 
 	/** User-initiated leave (or post-disconnect cleanup): restore the previous session. */
@@ -357,6 +358,7 @@ export class CollabGuestLink {
 		this.#applyHostState(pending.state);
 		this.#ctx.resetObserverRegistry();
 		this.#applyAgentSnapshots(pending.agents);
+		this.#ctx.syncRunningSubagentBadge();
 		this.#assistantStreamSynced = false;
 		setSessionTerminalTitle(pending.state.sessionName ?? pending.header.title, pending.state.cwd);
 		this.#ctx.chatContainer.clear();
@@ -440,6 +442,7 @@ export class CollabGuestLink {
 				break;
 			case "agents":
 				this.#applyAgentSnapshots(frame.agents);
+				this.#ctx.syncRunningSubagentBadge();
 				break;
 			case "transcript": {
 				const resolve = this.#pendingTranscripts.get(frame.reqId);
@@ -573,6 +576,7 @@ export class CollabGuestLink {
 		this.#ctx.statusLine.setCollabStatus(null);
 		this.#flushPendingTranscripts();
 		this.#clearAgentMirror();
+		this.#ctx.syncRunningSubagentBadge();
 		this.#ctx.resetObserverRegistry();
 		this.#clearTransientUi();
 		// Replica file stays on disk: it is a valid session file outside the

package/src/collab/host.ts

@@ -11,7 +11,6 @@
 
 import { timingSafeEqual } from "node:crypto";
 import * as fs from "node:fs/promises";
-import * as os from "node:os";
 import type { ImageContent, TextContent } from "@oh-my-pi/pi-ai";
 import { logger } from "@oh-my-pi/pi-utils";
 import type { BusChannel, AgentEvent as WireAgentEvent, SessionEntry as WireSessionEntry } from "@oh-my-pi/pi-wire";
@@ -21,8 +20,9 @@ import { type AgentRef, AgentRegistry } from "../registry/agent-registry";
 import type { AgentSessionEvent } from "../session/agent-session";
 import { stripImagesFromMessage, USER_INTERRUPT_LABEL } from "../session/messages";
 import type { SessionEntry as StoredSessionEntry } from "../session/session-entries";
-import { TASK_SUBAGENT_LIFECYCLE_CHANNEL, TASK_SUBAGENT_PROGRESS_CHANNEL } from "../task";
+import { TASK_SUBAGENT_LIFECYCLE_CHANNEL, TASK_SUBAGENT_PROGRESS_CHANNEL } from "../task/types";
 import { generateRoomKey, generateWriteToken, importRoomKey } from "./crypto";
+import { collabDisplayName } from "./display-name";
 import {
 	type AgentSnapshot,
 	COLLAB_PROMPT_MESSAGE_TYPE,
@@ -102,17 +102,6 @@ const TRANSCRIPT_READ_CAP = 4 * 1024 * 1024;
  */
 const SNAPSHOT_CHUNK_BYTES = 512 * 1024;
 
-/** Display name for this process's user in collab sessions. */
-export function collabDisplayName(ctx: InteractiveModeContext): string {
-	const configured = (ctx.settings.get("collab.displayName") ?? "").trim();
-	if (configured) return configured;
-	try {
-		return os.userInfo().username;
-	} catch {
-		return "anonymous";
-	}
-}
-
 export class CollabHost {
 	#ctx: InteractiveModeContext;
 	#socket: CollabSocket | null = null;

package/src/commands/gc.ts

@@ -0,0 +1,46 @@
+/**
+ * Run on-disk storage maintenance.
+ */
+import { Command, Flags } from "@oh-my-pi/pi-utils/cli";
+import { collectGcErrors, type GcCommandArgs, runGcCommand } from "../cli/gc-cli";
+
+export default class Gc extends Command {
+	static description = "Run storage garbage collection";
+
+	static flags = {
+		apply: Flags.boolean({ description: "Apply changes (default is dry-run)" }),
+		json: Flags.boolean({ description: "Output JSON" }),
+		"agent-dir": Flags.string({ description: "Agent directory to maintain" }),
+		blobs: Flags.boolean({ description: "Sweep unreferenced blobs" }),
+		archive: Flags.boolean({ description: "Archive cold sessions" }),
+		wal: Flags.boolean({ description: "Checkpoint history/model database WAL files" }),
+		"cold-archive-after-days": Flags.integer({ description: "Minimum session age before archiving" }),
+		"retain-newest-global": Flags.integer({ description: "Always keep this many newest sessions active" }),
+		"retain-newest-per-cwd": Flags.integer({ description: "Always keep this many newest sessions per cwd active" }),
+	};
+
+	async run(): Promise<void> {
+		const { flags } = await this.parse(Gc);
+		const cmd: GcCommandArgs = {
+			flags: {
+				apply: flags.apply,
+				json: flags.json,
+				agentDir: flags["agent-dir"],
+				blobs: flags.blobs,
+				archive: flags.archive,
+				wal: flags.wal,
+				coldArchiveAfterDays: flags["cold-archive-after-days"],
+				retainNewestGlobal: flags["retain-newest-global"],
+				retainNewestPerCwd: flags["retain-newest-per-cwd"],
+			},
+		};
+		const result = await runGcCommand(cmd);
+		const errors = collectGcErrors(result);
+		if (errors.length > 0) {
+			process.stderr.write(
+				`GC completed with ${errors.length} error${errors.length === 1 ? "" : "s"}:\n${errors.map(error => `- ${error}`).join("\n")}\n`,
+			);
+			process.exitCode = 1;
+		}
+	}
+}

package/src/commands/worktree.ts

@@ -1,8 +1,10 @@
 /**
  * List and clean up agent-managed git worktrees under `~/.omp/wt`.
  */
+import { getProjectDir } from "@oh-my-pi/pi-utils";
 import { Args, Command, Flags } from "@oh-my-pi/pi-utils/cli";
 import { clearWorktrees, listWorktrees } from "../cli/worktree-cli";
+import { Settings } from "../config/settings";
 
 export default class Worktree extends Command {
 	static description = "List or clear agent-managed git worktrees (~/.omp/wt)";
@@ -43,6 +45,10 @@ export default class Worktree extends Command {
 
 	async run(): Promise<void> {
 		const { args, flags } = await this.parse(Worktree);
+		// Load settings so the `worktree.base` override is applied before we scan
+		// — otherwise this command would inspect ~/.omp/wt while the agent created
+		// its worktrees under the configured base.
+		await Settings.init({ cwd: getProjectDir() });
 		if (args.action === "clear") {
 			await clearWorktrees({
 				all: flags.all ?? false,

package/src/config/inline-tool-descriptors-mode.ts

@@ -0,0 +1,27 @@
+import { modelFamilyToken } from "@oh-my-pi/pi-catalog/identity";
+
+/**
+ * Resolves whether full tool descriptors should be inlined into the system
+ * prompt (and stripped from provider tool schemas) for a given model and
+ * setting.
+ *
+ * `auto` enforces a per-model policy: inline for Gemini models, off otherwise.
+ * Gemini benefits from descriptors in-prompt; other providers keep them in the
+ * tool schemas. `on`/`off` are explicit user overrides.
+ *
+ * @param modelId Canonical model id (e.g. `gemini-3-pro`); resolve aliases via
+ *   `ModelRegistry.getCanonicalId` before calling so `auto` classifies correctly.
+ */
+export function shouldInlineToolDescriptors(
+	setting: "auto" | "on" | "off" | undefined,
+	modelId: string | undefined,
+): boolean {
+	switch (setting ?? "auto") {
+		case "on":
+			return true;
+		case "off":
+			return false;
+		default:
+			return modelId !== undefined && modelFamilyToken(modelId) === "gemini";
+	}
+}

package/src/config/model-discovery.ts

@@ -6,14 +6,19 @@
  * discovery lives in pi-catalog's provider-models.
  */
 import { type ApiKey, type FetchImpl, withAuth } from "@oh-my-pi/pi-ai";
-import type { Api, Model } from "@oh-my-pi/pi-ai/types";
+import type { Api, Model, RemoteCompactionConfig } from "@oh-my-pi/pi-ai/types";
 import { buildModel } from "@oh-my-pi/pi-catalog/build";
 import {
 	getBundledModelReferenceIndex,
 	resolveModelReference,
 	stripBracketedModelIdAffixes,
 } from "@oh-my-pi/pi-catalog/identity";
-import { fetchLmStudioNativeModelMetadata } from "@oh-my-pi/pi-catalog/provider-models/openai-compat";
+import {
+	fetchLiteLLMRichModels,
+	fetchLmStudioNativeModelMetadata,
+	OPENAI_COMPAT_DISCOVERY_DEFAULT_CONTEXT_WINDOW,
+	OPENAI_COMPAT_DISCOVERY_DEFAULT_MAX_TOKENS,
+} from "@oh-my-pi/pi-catalog/provider-models/openai-compat";
 import type { ModelSpec } from "@oh-my-pi/pi-catalog/types";
 import { isRecord } from "@oh-my-pi/pi-utils";
 import type { ProviderDiscovery } from "./models-config-schema";
@@ -27,7 +32,8 @@ import type { ProviderDiscovery } from "./models-config-schema";
 // mid-stream when models hit the cap on legitimate large tool calls (see
 // issue #1528: `write` payloads >~5KB on deepseek-v4-pro surfaced as
 // "socket connection was closed unexpectedly").
-export const DISCOVERY_DEFAULT_MAX_TOKENS = 32_768;
+export const DISCOVERY_DEFAULT_CONTEXT_WINDOW = OPENAI_COMPAT_DISCOVERY_DEFAULT_CONTEXT_WINDOW;
+export const DISCOVERY_DEFAULT_MAX_TOKENS = OPENAI_COMPAT_DISCOVERY_DEFAULT_MAX_TOKENS;
 
 const DEFAULT_OLLAMA_BASE_URL = "http://127.0.0.1:11434";
 const OLLAMA_HOST_DEFAULT_PORT = "11434";
@@ -89,6 +95,7 @@ export interface DiscoveryProviderConfig {
 	baseUrl?: string;
 	headers?: Record<string, string>;
 	compat?: ModelSpec<Api>["compat"];
+	remoteCompaction?: RemoteCompactionConfig<Api>;
 	discovery: ProviderDiscovery;
 	optional?: boolean;
 }
@@ -209,6 +216,8 @@ export function discoverModelsByProviderType(
 			return discoverOpenAIModelsList(providerConfig, ctx);
 		case "proxy":
 			return discoverProxyModels(providerConfig, ctx);
+		case "litellm":
+			return discoverLiteLLMModels(providerConfig, ctx);
 	}
 }
 
@@ -302,7 +311,7 @@ export async function discoverOllamaModels(
 			input: metadata?.input ?? ["text"],
 			imageInputDecoder: "stb",
 			cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
-			contextWindow: metadata?.contextWindow ?? 128000,
+			contextWindow: metadata?.contextWindow ?? DISCOVERY_DEFAULT_CONTEXT_WINDOW,
 			maxTokens: Math.min(metadata?.contextWindow ?? Number.POSITIVE_INFINITY, DISCOVERY_DEFAULT_MAX_TOKENS),
 			headers: providerConfig.headers,
 		} as ModelSpec<Api>);
@@ -369,7 +378,7 @@ export async function discoverLlamaCppModels(
 	for (const item of models) {
 		const { id } = item;
 		if (!id) continue;
-		const contextWindow = item.contextWindow ?? serverMetadata?.contextWindow ?? 128000;
+		const contextWindow = item.contextWindow ?? serverMetadata?.contextWindow ?? DISCOVERY_DEFAULT_CONTEXT_WINDOW;
 		discovered.push(
 			buildModel({
 				id,
@@ -467,7 +476,7 @@ export async function discoverOpenAIModelsList(
 			toPositiveNumberOrUndefined(item.max_model_len) ??
 			toPositiveNumberOrUndefined(item.context_length) ??
 			nativeMetadataForModel?.contextWindow ??
-			128000;
+			DISCOVERY_DEFAULT_CONTEXT_WINDOW;
 		discovered.push(
 			buildModel({
 				id,
@@ -493,6 +502,59 @@ export async function discoverOpenAIModelsList(
 	return discovered;
 }
 
+export async function discoverLiteLLMModels(
+	providerConfig: DiscoveryProviderConfig,
+	ctx: DiscoveryContext,
+): Promise<Model<Api>[]> {
+	const baseUrl = normalizeLiteLLMDiscoveryBaseUrl(providerConfig.baseUrl);
+	const references = getBundledModelReferenceIndex();
+	const resolveReference = (id: string) => resolveModelReference(id, references) as ModelSpec<Api> | undefined;
+	const baseHeaders: Record<string, string> = { ...(providerConfig.headers ?? {}) };
+	let headers = baseHeaders;
+	const attempt = async (h: Record<string, string>) => {
+		headers = h;
+		let authError: (Error & { status: number }) | undefined;
+		const authAwareFetch: FetchImpl = async (input, init) => {
+			const response = await ctx.fetch(input, init);
+			if (response.status === 401) {
+				authError = new Error(`HTTP ${response.status} from ${String(input)}`) as Error & { status: number };
+				authError.status = response.status;
+			}
+			return response;
+		};
+		const models = await fetchLiteLLMRichModels({
+			api: providerConfig.api,
+			provider: providerConfig.provider,
+			baseUrl,
+			headers: h,
+			fetch: authAwareFetch,
+			referenceResolver: resolveReference,
+			signal: AbortSignal.timeout(10_000),
+		});
+		if (authError && models === null) {
+			throw authError;
+		}
+		return models;
+	};
+	const apiKey = await ctx.getBearerApiKeyResolver(providerConfig.provider);
+	let richModels: ModelSpec<Api>[] | null;
+	try {
+		richModels = apiKey
+			? await withAuth(apiKey, key => attempt({ ...baseHeaders, Authorization: `Bearer ${key}` }))
+			: await attempt(baseHeaders);
+	} catch (error) {
+		const status = typeof error === "object" && error !== null && "status" in error ? error.status : undefined;
+		if (status !== 401) {
+			throw error;
+		}
+		richModels = null;
+	}
+	if (!richModels || richModels.length === 0) {
+		return discoverOpenAIModelsList({ ...providerConfig, baseUrl }, ctx);
+	}
+	return richModels.map(spec => buildModel({ ...spec, headers }));
+}
+
 /**
  * Discover models from an Anthropic+OpenAI-compatible reseller proxy that
  * exposes both `/v1/messages` and `/v1/chat/completions`, advertising each
@@ -571,7 +633,10 @@ export async function discoverProxyModels(
 				cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
 				// Prefer the context_length the API reports for this model; fall
 				// back to the bundled reference, then a sane default.
-				contextWindow: toPositiveNumberOrUndefined(item.context_length) ?? reference?.contextWindow ?? 128000,
+				contextWindow:
+					toPositiveNumberOrUndefined(item.context_length) ??
+					reference?.contextWindow ??
+					DISCOVERY_DEFAULT_CONTEXT_WINDOW,
 				maxTokens: reference?.maxTokens ?? discoveryDefaultMaxTokens(api),
 				headers,
 				// OpenAI-compat fields are no-ops on anthropic models; the
@@ -616,7 +681,11 @@ function toLlamaCppNativeBaseUrl(baseUrl: string): string {
 	}
 }
 
-function normalizeOpenAIModelsListBaseUrl(baseUrl?: string): string {
+export function normalizeLiteLLMDiscoveryBaseUrl(baseUrl?: string): string {
+	return normalizeOpenAIModelsListBaseUrl(baseUrl ?? "http://localhost:4000/v1");
+}
+
+export function normalizeOpenAIModelsListBaseUrl(baseUrl?: string): string {
 	const defaultBaseUrl = "http://127.0.0.1:1234/v1";
 	const raw = baseUrl || defaultBaseUrl;
 	try {