pi-openai-toolkit 0.14.3 → 0.14.4

package/src/context-management/codex-provider.ts

@@ -3,16 +3,15 @@ import type { ExtensionContext } from "@earendil-works/pi-coding-agent";
 import { mergeProviderHeaders } from "../provider-headers";
 import { CODEX_CLIENT_VERSION } from "../responses-headers";
 import { normalizeBaseUrl } from "../runtime";
+import { isExactModelAllowed } from "../model-scope";
 import {
 	isNonEmptyString,
 	type CodexContextProvider,
 	type CodexContextProviderResolution,
 } from "./types";
-import { ASTRA_MODEL_ID } from "../types";
 
 const CODEX_PROVIDER = "openai-codex";
 const CODEX_API = "openai-codex-responses";
-const GATEWAY_PROVIDER = "uwoacrimson";
 const GATEWAY_API = "openai-responses";
 const CODEX_TOOL_ORIGINATOR = "codex_cli_rs";
 const CODEX_GATEWAY_ORIGINATOR = "codex_cli_rs";
@@ -32,12 +31,14 @@ export function isNativeCodexModel(
 
 export function isCodexGatewayModel(
 	model: ExtensionContext["model"] | undefined,
+	gatewayModels: readonly string[] = [],
 ): boolean {
-	// Only the Astra SKU has a verified Codex-compatible gateway relay path
-	// (NEWapi + CLIProxyAPI passthrough); other gateway models stay on remote
-	// compaction v2. Matching is on the bare model id, by design.
-	if (!model || model.provider !== GATEWAY_PROVIDER || model.api !== GATEWAY_API) return false;
-	return model.id === ASTRA_MODEL_ID;
+	// Gateway coverage is an operator allowlist of exact "provider/model" keys
+	// (compaction.gatewayContextModels). The backend relay must actually pass
+	// hosted window markers through; models outside the list stay on remote
+	// compaction v2.
+	if (!model || model.api !== GATEWAY_API) return false;
+	return isExactModelAllowed(model, gatewayModels);
 }
 
 export function normalizeCodexBackendBaseUrl(baseUrl: string | undefined | null): string | undefined {
@@ -172,6 +173,7 @@ export function codexContextProviderHeaders(
 export async function resolveCodexContextProvider(
 	ctx: ExtensionContext,
 	modelOverride: ExtensionContext["model"] = ctx.model,
+	gatewayModels: readonly string[] = [],
 ): Promise<CodexContextProviderResolution> {
 	const model = modelOverride;
 	const descriptor = {
@@ -183,13 +185,9 @@ export async function resolveCodexContextProvider(
 	if (!model) return { ok: false, reason: "unsupported-model" };
 
 	const isNative = isNativeCodexModel(model);
-	const isGateway = isCodexGatewayModel(model);
+	const isGateway = isCodexGatewayModel(model, gatewayModels);
 	if (!isNative && !isGateway) {
-		return {
-			ok: false,
-			reason: model.provider === GATEWAY_PROVIDER ? "unsupported-model" : "unsupported-model",
-			...descriptor,
-		};
+		return { ok: false, reason: "unsupported-model", ...descriptor };
 	}
 	if (!isNative && model.api !== GATEWAY_API) {
 		return { ok: false, reason: "unsupported-api", ...descriptor };
@@ -217,7 +215,7 @@ export async function resolveCodexContextProvider(
 			provider: {
 				kind: "codex-gateway",
 				route: "codex-gateway",
-				provider: GATEWAY_PROVIDER,
+				provider: model.provider,
 				api: GATEWAY_API,
 				model: model.id,
 				baseUrl: rawBaseUrl,

package/src/context-management/history-notes.ts

@@ -78,11 +78,12 @@ export async function callHistoryNotesBackend(
 		mode: "tokens",
 		limit: TOOL_OUTPUT_TOKEN_LIMIT,
 	},
+	gatewayModels: readonly string[] = [],
 ): Promise<HistoryNotesResult> {
 	const timeoutSignal = AbortSignal.timeout(BACKEND_TIMEOUT_MS);
 	if (signal?.aborted) return { ok: false, reason: "aborted" };
 	const requestSignal = signal ? AbortSignal.any([signal, timeoutSignal]) : timeoutSignal;
-	const providerResult = await resolveProviderWithDeadline(ctx, requestSignal, timeoutSignal);
+	const providerResult = await resolveProviderWithDeadline(ctx, requestSignal, timeoutSignal, gatewayModels);
 	if (providerResult === "aborted") return { ok: false, reason: "aborted" };
 	if (providerResult === "timeout") return { ok: false, reason: "backend-timeout" };
 	const provider = providerResult;
@@ -134,6 +135,7 @@ async function resolveProviderWithDeadline(
 	ctx: ExtensionContext,
 	requestSignal: AbortSignal,
 	timeoutSignal: AbortSignal,
+	gatewayModels: readonly string[] = [],
 ): Promise<Awaited<ReturnType<typeof resolveCodexContextProvider>> | "aborted" | "timeout"> {
 	if (requestSignal.aborted) return timeoutSignal.aborted ? "timeout" : "aborted";
 	return new Promise((resolve) => {
@@ -146,7 +148,7 @@ async function resolveProviderWithDeadline(
 		};
 		const onAbort = () => finish(timeoutSignal.aborted ? "timeout" : "aborted");
 		requestSignal.addEventListener("abort", onAbort, { once: true });
-		resolveCodexContextProvider(ctx, ctx.model).then(
+		resolveCodexContextProvider(ctx, ctx.model, gatewayModels).then(
 			(value) => finish(value),
 			() => finish({ ok: false, reason: "auth-resolution-failed" }),
 		);
@@ -156,6 +158,7 @@ async function resolveProviderWithDeadline(
 export async function loadHistoryNotesThreadHint(
 	ctx: ExtensionContext,
 	signal?: AbortSignal,
+	gatewayModels: readonly string[] = [],
 ): Promise<string | undefined> {
 	const result = await callHistoryNotesBackend(
 		THREAD_HINT_ENDPOINT,
@@ -163,6 +166,7 @@ export async function loadHistoryNotesThreadHint(
 		ctx,
 		signal,
 		{ mode: "bytes", limit: THREAD_HINT_MAX_BYTES },
+		gatewayModels,
 	);
 	if (!result.ok) {
 		if (result.reason === "aborted") throw new Error("Remote context hint request was aborted");
@@ -178,6 +182,7 @@ export async function executeHistoryNotesTool(
 	params: Record<string, unknown>,
 	ctx: ExtensionContext,
 	signal?: AbortSignal,
+	gatewayModels: readonly string[] = [],
 ): Promise<AgentToolResult<CodexHistoryNotesDetails>> {
 	const endpoint = namespace === "history"
 		? HISTORY_ENDPOINTS[action as HistoryAction]
@@ -190,6 +195,7 @@ export async function executeHistoryNotesTool(
 		ctx,
 		signal,
 		undefined,
+		gatewayModels,
 	);
 	if (!result.ok) throw new Error(formatHistoryNotesFailure(result.reason, result.status));
 	const modelResult = { ...result.value };

package/src/context-management/tools.ts

@@ -85,6 +85,7 @@ export function createContextManagementTools(
 	pi: ExtensionAPI,
 	manager: CodexContextWindowManager,
 	isActive: (ctx: ExtensionContext) => Promise<boolean> | boolean,
+	getGatewayModels: () => readonly string[] = () => [],
 ): ContextManagementTools {
 	const assertActive = async (ctx: ExtensionContext): Promise<void> => {
 		if (!(await isActive(ctx))) throw new Error("remote-context-inactive");
@@ -142,7 +143,7 @@ export function createContextManagementTools(
 		],
 		execute: async (_id, params, _signal, _update, ctx) => {
 			await assertActive(ctx);
-			return executeHistoryNotesTool("history", params.action, params as Record<string, unknown>, ctx, _signal);
+			return executeHistoryNotesTool("history", params.action, params as Record<string, unknown>, ctx, _signal, getGatewayModels());
 		},
 	};
 	const notes: ToolDefinition<typeof NOTES_PARAMETERS, CodexHistoryNotesDetails> = {
@@ -161,7 +162,7 @@ export function createContextManagementTools(
 		executionMode: "sequential",
 		execute: async (_id, params, _signal, _update, ctx) => {
 			await assertActive(ctx);
-			return executeHistoryNotesTool("notes", params.action, params as Record<string, unknown>, ctx, _signal);
+			return executeHistoryNotesTool("notes", params.action, params as Record<string, unknown>, ctx, _signal, getGatewayModels());
 		},
 	};
 	return { newContext, getContextRemaining, history, notes };
@@ -293,9 +294,10 @@ export function registerContextManagementTools(
 	pi: ExtensionAPI,
 	manager: CodexContextWindowManager,
 	isActive: (ctx: ExtensionContext) => Promise<boolean> | boolean,
+	getGatewayModels?: () => readonly string[],
 ): ContextManagementToolController {
 	const controller = new ContextManagementToolController(pi);
-	controller.register(createContextManagementTools(pi, manager, isActive));
+	controller.register(createContextManagementTools(pi, manager, isActive, getGatewayModels));
 	return controller;
 }
 

package/src/context-management/types.ts

@@ -67,7 +67,8 @@ export type NativeCodexContextProvider = {
 export type GatewayCodexContextProvider = {
 	kind: "codex-gateway";
 	route: "codex-gateway";
-	provider: "uwoacrimson";
+	/** Operator-defined provider name; matched via compaction.gatewayContextModels. */
+	provider: string;
 	api: "openai-responses";
 	model: string;
 	baseUrl: string;

package/src/extension-runtime.ts

@@ -14,6 +14,7 @@ import {
 	isNativeCodexModel,
 } from "./context-management/codex-provider";
 import { routeContextNamespaceToolMessage } from "./context-management/namespace-tools";
+import { loadHistoryNotesThreadHint } from "./context-management/history-notes";
 import { CodexContextWindowManager } from "./context-management/window-manager";
 import { registerContextManagementTools } from "./context-management/tools";
 import { writeDebugArtifact, writeReplayFailureArtifact } from "./debug";
@@ -109,7 +110,7 @@ async function isRemoteContextActive(
 	model: ExtensionContext["model"] = ctx.model,
 ): Promise<boolean> {
 	if (!config.enabled || config.contextManagement !== "remote") return false;
-	const resolution = await resolveCodexContextProvider(ctx, model);
+	const resolution = await resolveCodexContextProvider(ctx, model, config.gatewayContextModels);
 	return resolution.ok;
 }
 
@@ -118,7 +119,7 @@ function isCodexContextModel(
 	config: CompactionConfig,
 ): boolean {
 	return config.contextManagement === "remote"
-		&& (isNativeCodexModel(model) || isCodexGatewayModel(model));
+		&& (isNativeCodexModel(model) || isCodexGatewayModel(model, config.gatewayContextModels));
 }
 
 function notifyRemoteContextFailure(ctx: ExtensionContext, reason: string): void {
@@ -759,7 +760,9 @@ export default function registerCompactionExtension(
 	overrides: Partial<CompactionDependencies> = {},
 ) {
 	const loadConfig = overrides.loadConfig ?? loadToolkitConfig;
-	const contextWindows = overrides.contextWindows ?? new CodexContextWindowManager();
+	const contextWindows = overrides.contextWindows ?? new CodexContextWindowManager((ctx, signal) =>
+		loadHistoryNotesThreadHint(ctx, signal, loadConfig().config.compaction.gatewayContextModels)
+	);
 	const dependencies: CompactionDependencies = {
 		loadConfig,
 		remoteCompact: executeRemoteV2Compaction,
@@ -775,6 +778,7 @@ export default function registerCompactionExtension(
 			const config = dependencies.loadConfig().config.compaction;
 			return tools.isRegistered && await isRemoteContextActive(ctx, config);
 		},
+		() => dependencies.loadConfig().config.compaction.gatewayContextModels,
 	);
 	const remoteContextActive: RemoteContextActive = async (ctx, config, model = ctx.model) =>
 		tools.isRegistered && await isRemoteContextActive(ctx, config, model);
@@ -807,7 +811,7 @@ export default function registerCompactionExtension(
 				activationReason = "tool-name-conflict";
 				notifyRemoteContextFailure(ctx, activationReason);
 			} else {
-				const remoteResolution = await resolveCodexContextProvider(ctx, ctx.model);
+				const remoteResolution = await resolveCodexContextProvider(ctx, ctx.model, config.gatewayContextModels);
 				activationReason = remoteResolution.ok ? "codex-context-unavailable" : remoteResolution.reason;
 				notifyRemoteContextFailure(ctx, activationReason);
 			}
@@ -874,7 +878,7 @@ export default function registerCompactionExtension(
 		const config = dependencies.loadConfig().config.compaction;
 		if (!isCodexContextModel(ctx.model, config)) return;
 		if (!(await remoteContextActive(ctx, config))) return;
-		const provider = await resolveCodexContextProvider(ctx, ctx.model);
+		const provider = await resolveCodexContextProvider(ctx, ctx.model, config.gatewayContextModels);
 		if (provider.ok && provider.provider.kind === "codex-gateway") {
 			const sessionId = getSessionId(ctx);
 			const gatewayHeaders = codexContextProviderHeaders(provider.provider, {

package/src/types.ts

@@ -80,6 +80,12 @@ export type CompactionConfig = {
 	 * Subset of RESPONSES_COMPACT_CAPABLE_APIS that should use remote compaction.
 	 */
 	responsesApis: string[];
+	/**
+	 * Exact "provider/model" keys allowed to use Codex Remote Context on the
+	 * `openai-responses` gateway wire. The native `openai-codex` route ignores
+	 * this list; gateway coverage is opt-in per model.
+	 */
+	gatewayContextModels: string[];
 	/**
 	 * Percentage of the context window that, when remaining tokens drop below it,
 	 * triggers a checkpoint/new_context reminder (0-100; 0 disables reminders).
@@ -506,6 +512,7 @@ export const DEFAULT_COMPACTION_CONFIG: CompactionConfig = {
 	remoteCompactModel: undefined,
 	nativeFallback: { ...DEFAULT_NATIVE_FALLBACK_CONFIG },
 	responsesApis: [...RESPONSES_COMPACT_CAPABLE_APIS],
+	gatewayContextModels: [],
 	contextReminderThresholdPercent: 5,
 	notifyOnLoad: false,
 	debug: false,

package/src/web-search/prompt.ts

@@ -4,7 +4,7 @@ import { isWebSearchEnabledForModel, WEB_SEARCH_PROMPT_MARKER, type WebSearchMod
 export const WEB_SEARCH_PROMPT_SECTION = `${WEB_SEARCH_PROMPT_MARKER}
 ## Web Search
 
-Native OpenAI Web Search is available for this turn. Use it for current or online information when freshness matters instead of guessing, and cite the returned sources in the answer.`;
+The hosted \`web_search\` tool is part of your tool list for this model. The API executes it server-side: it is a real callable tool even though no client-side function schema or description accompanies it. Call it whenever current or online information matters instead of guessing, cite the returned sources in the answer, and never claim that web search or internet access is unavailable while this section is present.`;
 
 export function appendWebSearchPrompt(args: {
 	model: WebSearchModel | undefined;