diff --git a/packages/ai/CHANGELOG.md b/packages/ai/CHANGELOG.md
index 2eb0679d4..6fb701e27 100644
--- a/packages/ai/CHANGELOG.md
+++ b/packages/ai/CHANGELOG.md
@@ -2,6 +2,10 @@
 
 ## [Unreleased]
 
+### Added
+
+- Added a per-provider in-flight request limiter for LLM streams, shared across local OMP processes and configurable by callers with `maxInFlightRequests`.
+
 ## [16.1.11] - 2026-06-21
 
 ### Fixed
diff --git a/packages/ai/src/stream.ts b/packages/ai/src/stream.ts
index 3012b51e4..82d22dd29 100644
--- a/packages/ai/src/stream.ts
+++ b/packages/ai/src/stream.ts
@@ -1,3 +1,7 @@
+import * as crypto from "node:crypto";
+import * as fsSync from "node:fs";
+import * as fs from "node:fs/promises";
+import * as path from "node:path";
 import type { Effort } from "@oh-my-pi/pi-catalog/effort";
 import { isVertexExpressOpenAIUrl, isVertexRawPredictUrl } from "@oh-my-pi/pi-catalog/hosts";
 import {
@@ -8,7 +12,7 @@ import {
 	resolveWireModelId,
 } from "@oh-my-pi/pi-catalog/model-thinking";
 import { CATALOG_PROVIDERS, type ProviderCatalogEntry } from "@oh-my-pi/pi-catalog/provider-models";
-import { $env, $pickenv, extractHttpStatusFromError } from "@oh-my-pi/pi-utils";
+import { $env, $pickenv, extractHttpStatusFromError, getConfigRootDir, isEnoent, logger } from "@oh-my-pi/pi-utils";
 import { getCustomApi } from "./api-registry";
 import { AUTH_RETRY_STEPS, isApiKeyResolver, resolveRetryKey } from "./auth-retry";
 import { ProviderHttpError } from "./errors";
@@ -73,6 +77,421 @@ function isGoogleVertexAuthenticatedModel(model: Model<Api>): boolean {
 	);
 }
 
+type ProviderInFlightLease = {
+	path: string;
+	heartbeat: NodeJS.Timeout;
+	flushHeartbeat: () => Promise<void>;
+};
+
+type ProviderInFlightLeaseInfo = {
+	pid: number;
+	timestamp: number;
+	token: string;
+};
+type ProviderInFlightStaleLock = { token: string } | { mtimeMs: number };
+
+const PROVIDER_INFLIGHT_LOCK_STALE_MS = 10_000;
+const PROVIDER_INFLIGHT_LEASE_STALE_MS = 30_000;
+const PROVIDER_INFLIGHT_HEARTBEAT_MS = 5_000;
+const PROVIDER_INFLIGHT_SIGNAL_FALLBACK_MS = 250;
+
+let configuredProviderMaxInFlightRequests: Record<string, number> = {};
+let providerInFlightRootOverride: string | undefined;
+
+export function configureProviderMaxInFlightRequests(limits: Record<string, number> | undefined): void {
+	configuredProviderMaxInFlightRequests = limits ?? {};
+}
+
+function resolveProviderInFlightLimit(
+	provider: string,
+	options?: Pick<StreamOptions, "maxInFlightRequests">,
+): number | undefined {
+	const limits = options?.maxInFlightRequests ?? configuredProviderMaxInFlightRequests;
+	const value = limits[provider];
+	if (typeof value !== "number" || !Number.isFinite(value) || value <= 0) return undefined;
+	return Math.max(1, Math.floor(value));
+}
+
+function providerInFlightRoot(): string {
+	if (providerInFlightRootOverride) return providerInFlightRootOverride;
+	return path.join(getConfigRootDir(), "run", "provider-inflight");
+}
+
+function providerInFlightSegment(provider: string): string {
+	return crypto.createHash("sha256").update(provider).digest("base64url");
+}
+
+function providerInFlightDir(provider: string): string {
+	return path.join(providerInFlightRoot(), providerInFlightSegment(provider));
+}
+
+function providerInFlightSignalPath(provider: string): string {
+	return path.join(providerInFlightDir(provider), ".wakeup");
+}
+
+function providerInFlightLockDir(provider: string): string {
+	return `${providerInFlightDir(provider)}.lock`;
+}
+
+// `process.kill(pid, 0)` may throw for permission/sandbox reasons even when a
+// process exists. Treat non-ESRCH failures as alive; timestamp expiry still
+// reaps leases whose heartbeat stopped.
+function isProcessAlive(pid: number): boolean {
+	try {
+		process.kill(pid, 0);
+		return true;
+	} catch (error) {
+		return (error as NodeJS.ErrnoException).code !== "ESRCH";
+	}
+}
+
+async function readProviderInFlightInfo(infoPath: string): Promise<ProviderInFlightLeaseInfo | null> {
+	try {
+		const content = await fs.readFile(infoPath, "utf-8");
+		const parsed = JSON.parse(content) as Partial<ProviderInFlightLeaseInfo>;
+		if (typeof parsed.pid !== "number" || typeof parsed.timestamp !== "number" || typeof parsed.token !== "string") {
+			return null;
+		}
+		return { pid: parsed.pid, timestamp: parsed.timestamp, token: parsed.token };
+	} catch {
+		return null;
+	}
+}
+
+async function writeProviderInFlightInfo(dir: string, token: string): Promise<void> {
+	const info: ProviderInFlightLeaseInfo = { pid: process.pid, timestamp: Date.now(), token };
+	const infoPath = path.join(dir, "info.json");
+	const tempPath = path.join(dir, `.info-${process.pid}-${crypto.randomUUID()}.tmp`);
+	try {
+		await Bun.write(tempPath, JSON.stringify(info));
+		await fs.rename(tempPath, infoPath);
+	} catch (error) {
+		await fs.rm(tempPath, { force: true }).catch(() => {});
+		throw error;
+	}
+}
+
+async function isProviderInFlightDirStale(dir: string, staleMs: number): Promise<boolean> {
+	const info = await readProviderInFlightInfo(path.join(dir, "info.json"));
+	if (info) {
+		if (!isProcessAlive(info.pid)) return true;
+		return Date.now() - info.timestamp > staleMs;
+	}
+
+	try {
+		const stat = await fs.stat(path.join(dir, "info.json"));
+		return Date.now() - stat.mtimeMs > staleMs;
+	} catch (error) {
+		if (!isEnoent(error)) throw error;
+	}
+
+	try {
+		const stat = await fs.stat(dir);
+		return Date.now() - stat.mtimeMs > staleMs;
+	} catch (error) {
+		if (isEnoent(error)) return false;
+		throw error;
+	}
+}
+
+async function readProviderInFlightStaleLock(lockDir: string): Promise<ProviderInFlightStaleLock | null> {
+	const infoPath = path.join(lockDir, "info.json");
+	const info = await readProviderInFlightInfo(infoPath);
+	if (info) return isProcessAlive(info.pid) ? null : { token: info.token };
+
+	try {
+		const stat = await fs.stat(lockDir);
+		return Date.now() - stat.mtimeMs > PROVIDER_INFLIGHT_LOCK_STALE_MS ? { mtimeMs: stat.mtimeMs } : null;
+	} catch (error) {
+		if (isEnoent(error)) return null;
+		throw error;
+	}
+}
+
+async function releaseProviderInFlightStaleLock(lockDir: string, stale: ProviderInFlightStaleLock): Promise<void> {
+	if ("token" in stale) {
+		await releaseProviderInFlightLock(lockDir, stale.token);
+		return;
+	}
+
+	const infoPath = path.join(lockDir, "info.json");
+	if (await readProviderInFlightInfo(infoPath)) return;
+	try {
+		const stat = await fs.stat(lockDir);
+		if (stat.mtimeMs !== stale.mtimeMs || Date.now() - stat.mtimeMs <= PROVIDER_INFLIGHT_LOCK_STALE_MS) return;
+		await fs.rm(lockDir, { recursive: true, force: true });
+	} catch {}
+}
+
+// Best-effort token-checked release. Untokened calls are used only after stale
+// detection concludes the owner is dead or its heartbeat expired.
+async function releaseProviderInFlightLock(lockDir: string, token?: string): Promise<void> {
+	try {
+		if (token !== undefined) {
+			const info = await readProviderInFlightInfo(path.join(lockDir, "info.json"));
+			if (!info || info.token !== token) return;
+		}
+		await fs.rm(lockDir, { recursive: true, force: true });
+	} catch {}
+}
+
+async function acquireProviderInFlightLock(provider: string, signal?: AbortSignal): Promise<() => Promise<void>> {
+	const lockDir = providerInFlightLockDir(provider);
+	await fs.mkdir(path.dirname(lockDir), { recursive: true });
+
+	while (true) {
+		if (signal?.aborted) throw signal.reason ?? new Error("Provider request aborted before dispatch");
+		try {
+			await fs.mkdir(lockDir);
+			const token = crypto.randomUUID();
+			try {
+				await writeProviderInFlightInfo(lockDir, token);
+			} catch (error) {
+				await releaseProviderInFlightLock(lockDir);
+				throw error;
+			}
+			return async () => {
+				await releaseProviderInFlightLock(lockDir, token);
+			};
+		} catch (error) {
+			if ((error as NodeJS.ErrnoException).code !== "EEXIST") throw error;
+		}
+
+		const staleLock = await readProviderInFlightStaleLock(lockDir);
+		if (staleLock) {
+			await releaseProviderInFlightStaleLock(lockDir, staleLock);
+			await signalProviderInFlightWaiters(provider);
+			continue;
+		}
+
+		await waitForProviderInFlightSignal(provider, signal);
+	}
+}
+
+async function cleanupProviderInFlightLeases(providerDir: string): Promise<number> {
+	let active = 0;
+	let entries: string[];
+	try {
+		entries = await fs.readdir(providerDir);
+	} catch (error) {
+		if (isEnoent(error)) return 0;
+		throw error;
+	}
+
+	for (const entry of entries) {
+		const leaseDir = path.join(providerDir, entry);
+		let isDirectory = false;
+		try {
+			isDirectory = (await fs.stat(leaseDir)).isDirectory();
+		} catch (error) {
+			if (isEnoent(error)) continue;
+			throw error;
+		}
+		if (!isDirectory) continue;
+		if (await isProviderInFlightDirStale(leaseDir, PROVIDER_INFLIGHT_LEASE_STALE_MS)) {
+			await fs.rm(leaseDir, { recursive: true, force: true });
+			continue;
+		}
+		active++;
+	}
+	return active;
+}
+
+async function tryAcquireProviderInFlightLease(
+	provider: string,
+	limit: number,
+	signal?: AbortSignal,
+): Promise<ProviderInFlightLease | null> {
+	const releaseLock = await acquireProviderInFlightLock(provider, signal);
+	let leaseCreated = false;
+	try {
+		const dir = providerInFlightDir(provider);
+		await fs.mkdir(dir, { recursive: true });
+		const active = await cleanupProviderInFlightLeases(dir);
+		if (active >= limit) return null;
+
+		const leaseDir = path.join(dir, `${process.pid}-${Date.now()}-${crypto.randomUUID()}`);
+		const token = crypto.randomUUID();
+		try {
+			await fs.mkdir(leaseDir);
+			await writeProviderInFlightInfo(leaseDir, token);
+			leaseCreated = true;
+		} catch (error) {
+			await removeProviderInFlightLeaseDir(leaseDir).catch(() => {});
+			throw error;
+		}
+		let heartbeatFlush = Promise.resolve();
+		const touchHeartbeat = () => {
+			heartbeatFlush = heartbeatFlush
+				.then(
+					() => writeProviderInFlightInfo(leaseDir, token),
+					() => writeProviderInFlightInfo(leaseDir, token),
+				)
+				.catch(() => {});
+		};
+		const heartbeat = setInterval(touchHeartbeat, PROVIDER_INFLIGHT_HEARTBEAT_MS);
+		heartbeat.unref?.();
+		return { path: leaseDir, heartbeat, flushHeartbeat: () => heartbeatFlush };
+	} finally {
+		await releaseLock();
+		if (leaseCreated) await signalProviderInFlightWaiters(provider);
+	}
+}
+
+async function signalProviderInFlightWaiters(provider: string): Promise<void> {
+	try {
+		const dir = providerInFlightDir(provider);
+		await fs.mkdir(dir, { recursive: true });
+		await Bun.write(providerInFlightSignalPath(provider), String(Date.now()));
+	} catch {}
+}
+
+function waitForProviderInFlightSignal(provider: string, signal?: AbortSignal): Promise<void> {
+	if (signal?.aborted) return Promise.reject(signal.reason ?? new Error("Provider request aborted before dispatch"));
+	const signalPath = providerInFlightSignalPath(provider);
+	const waitStarted = Date.now();
+	const { promise, resolve, reject } = Promise.withResolvers<void>();
+	let settled = false;
+	let watcher: fsSync.FSWatcher | undefined;
+	const timer = setTimeout(() => finish(resolve), PROVIDER_INFLIGHT_SIGNAL_FALLBACK_MS);
+	const finish = (settle: () => void) => {
+		if (settled) return;
+		settled = true;
+		clearTimeout(timer);
+		watcher?.close();
+		signal?.removeEventListener("abort", onAbort);
+		settle();
+	};
+	const onAbort = () => {
+		finish(() => reject(signal?.reason ?? new Error("Provider request aborted before dispatch")));
+	};
+	signal?.addEventListener("abort", onAbort, { once: true });
+	try {
+		watcher = fsSync.watch(providerInFlightDir(provider), (_event, filename) => {
+			if (filename === ".wakeup" || filename === null) {
+				finish(resolve);
+			}
+		});
+		void fs.stat(signalPath).then(
+			stat => {
+				if (stat.mtimeMs >= waitStarted) finish(resolve);
+			},
+			error => {
+				if (!isEnoent(error)) finish(resolve);
+			},
+		);
+	} catch {
+		// Filesystem notifications are best-effort across platforms; the fallback
+		// timer keeps stale-lock/lease cleanup progressing if an event is dropped.
+	}
+	return promise;
+}
+
+async function removeProviderInFlightLeaseDir(leasePath: string): Promise<void> {
+	for (let attempt = 0; attempt < 3; attempt++) {
+		try {
+			await fs.rm(leasePath, { recursive: true, force: true });
+			return;
+		} catch (error) {
+			if (isEnoent(error)) return;
+			const code = (error as NodeJS.ErrnoException).code;
+			if (attempt < 2 && (code === "EBUSY" || code === "ENOTEMPTY" || code === "EPERM")) {
+				await Bun.sleep(25);
+				continue;
+			}
+			throw error;
+		}
+	}
+}
+
+async function releaseProviderInFlightLease(provider: string, lease: ProviderInFlightLease): Promise<void> {
+	clearInterval(lease.heartbeat);
+	await lease.flushHeartbeat();
+	await removeProviderInFlightLeaseDir(lease.path);
+	await signalProviderInFlightWaiters(provider);
+}
+
+async function acquireProviderInFlightSlot(
+	provider: string,
+	limit: number | undefined,
+	signal?: AbortSignal,
+): Promise<() => Promise<void>> {
+	if (limit === undefined) return async () => {};
+	let loggedWait = false;
+	while (true) {
+		if (signal?.aborted) throw signal.reason ?? new Error("Provider request aborted before dispatch");
+		const lease = await tryAcquireProviderInFlightLease(provider, limit, signal);
+		if (lease) return () => releaseProviderInFlightLease(provider, lease);
+		if (!loggedWait) {
+			loggedWait = true;
+			logger.debug("Provider in-flight limit blocked request", { provider, limit });
+		}
+		await waitForProviderInFlightSignal(provider, signal);
+	}
+}
+
+export const __providerInFlightForTesting = {
+	setRoot(root: string | undefined): void {
+		providerInFlightRootOverride = root;
+	},
+	providerDir(provider: string): string {
+		return providerInFlightDir(provider);
+	},
+	lockDir(provider: string): string {
+		return providerInFlightLockDir(provider);
+	},
+	async captureStaleLockRelease(provider: string): Promise<(() => Promise<void>) | null> {
+		const lockDir = providerInFlightLockDir(provider);
+		const stale = await readProviderInFlightStaleLock(lockDir);
+		if (!stale) return null;
+		return () => releaseProviderInFlightStaleLock(lockDir, stale);
+	},
+};
+
+function withProviderInFlightLimit<TOptions extends Pick<StreamOptions, "signal" | "maxInFlightRequests">>(
+	model: Model<Api>,
+	options: TOptions | undefined,
+	dispatch: () => AssistantMessageEventStream,
+): AssistantMessageEventStream {
+	const limit = resolveProviderInFlightLimit(model.provider, options);
+	if (limit === undefined) return dispatch();
+
+	const outer = new AssistantMessageEventStream();
+	void (async () => {
+		let release: (() => Promise<void>) | undefined;
+		let released = false;
+		const releaseOnce = async () => {
+			if (!release || released) return;
+			released = true;
+			await release();
+		};
+		try {
+			const startedWaitingAt = Date.now();
+			release = await acquireProviderInFlightSlot(model.provider, limit, options?.signal);
+			if (Date.now() - startedWaitingAt >= PROVIDER_INFLIGHT_SIGNAL_FALLBACK_MS) {
+				logger.debug("Provider in-flight limit wait completed", { provider: model.provider, limit });
+			}
+			if (options?.signal?.aborted) {
+				throw options.signal.reason ?? new Error("Provider request aborted before dispatch");
+			}
+			const inner = dispatch();
+			try {
+				for await (const event of inner) {
+					outer.push(event);
+					if (outer.done) return;
+				}
+				if (!outer.done) outer.end(await inner.result());
+			} finally {
+				await releaseOnce();
+			}
+		} catch (error) {
+			await releaseOnce();
+			if (!outer.done) outer.fail(error);
+		}
+	})();
+	return outer;
+}
+
 function createVertexAuthenticatedFetch(options: StreamOptions | undefined): FetchImpl {
 	const baseFetch = options?.fetch ?? fetch;
 	const vertexFetch = async (input: string | URL | Request, init?: RequestInit): Promise<Response> => {
@@ -227,7 +646,9 @@ export function stream<TApi extends Api>(
 	context: Context,
 	options?: OptionsForApi<TApi>,
 ): AssistantMessageEventStream {
-	return withGeminiThinkingLoopGuard(model, options, opts => streamDispatch(model, context, opts));
+	return withGeminiThinkingLoopGuard(model, options, opts =>
+		withProviderInFlightLimit(model, opts, () => streamDispatch(model, context, opts)),
+	);
 }
 
 function streamDispatch<TApi extends Api>(
@@ -500,14 +921,16 @@ export function streamSimple<TApi extends Api>(
 	// extension-registered APIs can't accidentally override a configured
 	// pi-native transport.
 	if (model.transport === "pi-native") {
-		return withGeminiThinkingLoopGuard(model, requestOptions, opts => streamPiNative(model, context, opts));
+		return withGeminiThinkingLoopGuard(model, requestOptions, opts =>
+			withProviderInFlightLimit(model, opts, () => streamPiNative(model, context, opts)),
+		);
 	}
 
 	// Check custom API registry (extension-provided APIs)
 	const customApiProvider = getCustomApi(model.api);
 	if (customApiProvider) {
 		return withGeminiThinkingLoopGuard(model, requestOptions, opts =>
-			customApiProvider.streamSimple(model, context, opts),
+			withProviderInFlightLimit(model, opts, () => customApiProvider.streamSimple(model, context, opts)),
 		);
 	}
 
@@ -531,30 +954,36 @@ export function streamSimple<TApi extends Api>(
 
 	// GitLab Duo - wraps Anthropic/OpenAI behind GitLab AI Gateway direct access tokens
 	if (isGitLabDuoModel(model)) {
-		return streamGitLabDuo(model, context, {
-			...requestOptions,
-			apiKey,
-		});
+		return withProviderInFlightLimit(model, requestOptions, () =>
+			streamGitLabDuo(model, context, {
+				...requestOptions,
+				apiKey,
+			}),
+		);
 	}
 
 	// Kimi Code - route to dedicated handler that wraps OpenAI or Anthropic API
 	if (isKimiModel(model)) {
 		// Pass raw SimpleStreamOptions - streamKimi handles mapping internally
-		return streamKimi(model as Model<"openai-completions">, context, {
-			...requestOptions,
-			apiKey,
-			format: requestOptions?.kimiApiFormat ?? "anthropic",
-		});
+		return withProviderInFlightLimit(model, requestOptions, () =>
+			streamKimi(model as Model<"openai-completions">, context, {
+				...requestOptions,
+				apiKey,
+				format: requestOptions?.kimiApiFormat ?? "anthropic",
+			}),
+		);
 	}
 
 	// Synthetic - route to dedicated handler that wraps OpenAI or Anthropic API
 	if (isSyntheticModel(model)) {
 		// Pass raw SimpleStreamOptions - streamSynthetic handles mapping internally
-		return streamSynthetic(model as Model<"openai-completions">, context, {
-			...requestOptions,
-			apiKey,
-			format: requestOptions?.syntheticApiFormat ?? "openai", // Default to OpenAI format
-		});
+		return withProviderInFlightLimit(model, requestOptions, () =>
+			streamSynthetic(model as Model<"openai-completions">, context, {
+				...requestOptions,
+				apiKey,
+				format: requestOptions?.syntheticApiFormat ?? "openai", // Default to OpenAI format
+			}),
+		);
 	}
 	const providerOptions = mapOptionsForApi(model, requestOptions, apiKey);
 	return stream(model, context, providerOptions);
@@ -774,6 +1203,7 @@ function mapOptionsForApi<TApi extends Api>(
 		streamFirstEventTimeoutMs: options?.streamFirstEventTimeoutMs,
 		streamIdleTimeoutMs: options?.streamIdleTimeoutMs,
 		providerSessionState: options?.providerSessionState,
+		maxInFlightRequests: options?.maxInFlightRequests,
 		onPayload: options?.onPayload,
 		onResponse: options?.onResponse,
 		onSseEvent: options?.onSseEvent,
diff --git a/packages/ai/src/types.ts b/packages/ai/src/types.ts
index 80d1a48a3..2efbe4e3f 100644
--- a/packages/ai/src/types.ts
+++ b/packages/ai/src/types.ts
@@ -272,6 +272,14 @@ export interface StreamOptions {
 	 * Providers can use this to persist transport/session state between turns.
 	 */
 	providerSessionState?: Map<string, ProviderSessionState>;
+	/**
+	 * Optional per-provider concurrent request cap for LLM stream calls. Keys are
+	 * provider ids (`model.provider`); positive numeric values cap in-flight
+	 * requests across local OMP processes that share the same config root. Omitted
+	 * providers are unlimited. Non-chat provider APIs that bypass stream helpers
+	 * are not covered.
+	 */
+	maxInFlightRequests?: Record<string, number>;
 	/**
 	 * Optional callback for inspecting or replacing provider payloads before sending.
 	 * Return undefined to keep the payload unchanged.
diff --git a/packages/ai/test/provider-inflight.test.ts b/packages/ai/test/provider-inflight.test.ts
new file mode 100644
index 000000000..23eb667b4
--- /dev/null
+++ b/packages/ai/test/provider-inflight.test.ts
@@ -0,0 +1,270 @@
+import { afterEach, beforeEach, describe, expect, test } from "bun:test";
+import * as fs from "node:fs/promises";
+import * as os from "node:os";
+import * as path from "node:path";
+import { clearCustomApis } from "@oh-my-pi/pi-ai/api-registry";
+import { createMockModel, registerMockApi } from "@oh-my-pi/pi-ai/providers/mock";
+import {
+	__providerInFlightForTesting,
+	configureProviderMaxInFlightRequests,
+	streamSimple,
+} from "@oh-my-pi/pi-ai/stream";
+import type { Context } from "@oh-my-pi/pi-ai/types";
+
+function context(): Context {
+	return {
+		systemPrompt: [],
+		messages: [{ role: "user", content: "hi", timestamp: 0 }],
+	};
+}
+
+let limiterRoot: string | undefined;
+
+afterEach(async () => {
+	clearCustomApis();
+	configureProviderMaxInFlightRequests(undefined);
+	__providerInFlightForTesting.setRoot(undefined);
+	if (limiterRoot !== undefined) {
+		await fs.rm(limiterRoot, { recursive: true, force: true });
+		limiterRoot = undefined;
+	}
+});
+
+async function useIsolatedLimiterRoot(): Promise<void> {
+	limiterRoot = await fs.mkdtemp(path.join(os.tmpdir(), "omp-provider-inflight-test-"));
+	__providerInFlightForTesting.setRoot(limiterRoot);
+}
+
+function limiterDir(provider: string): string {
+	return __providerInFlightForTesting.providerDir(provider);
+}
+
+describe("provider in-flight request limits", () => {
+	beforeEach(async () => {
+		await useIsolatedLimiterRoot();
+	});
+	test("serializes concurrent streamSimple calls for the same provider", async () => {
+		registerMockApi();
+		const firstStarted = Promise.withResolvers<void>();
+		const releaseFirst = Promise.withResolvers<void>();
+		let active = 0;
+		let maxActive = 0;
+		let callIndex = 0;
+		const mock = createMockModel({
+			provider: "tests",
+			handler: async () => {
+				callIndex++;
+				active++;
+				maxActive = Math.max(maxActive, active);
+				try {
+					if (callIndex === 1) {
+						firstStarted.resolve();
+						await releaseFirst.promise;
+					}
+					return { content: [`reply ${callIndex}`] };
+				} finally {
+					active--;
+				}
+			},
+		});
+
+		const first = streamSimple(mock.model, context(), { maxInFlightRequests: { tests: 1 } });
+		const firstResult = first.result();
+		await firstStarted.promise;
+
+		const second = streamSimple(mock.model, context(), { maxInFlightRequests: { tests: 1 } });
+		await Bun.sleep(20);
+		expect(mock.calls).toHaveLength(1);
+
+		releaseFirst.resolve();
+		const [firstMessage, secondMessage] = await Promise.all([firstResult, second.result()]);
+
+		expect(firstMessage.content).toEqual([{ type: "text", text: "reply 1" }]);
+		expect(secondMessage.content).toEqual([{ type: "text", text: "reply 2" }]);
+		expect(maxActive).toBe(1);
+		expect(mock.calls).toHaveLength(2);
+	});
+
+	test("removes an aborted queued request without dispatching it", async () => {
+		registerMockApi();
+		const firstStarted = Promise.withResolvers<void>();
+		const releaseFirst = Promise.withResolvers<void>();
+		let callIndex = 0;
+		const mock = createMockModel({
+			provider: "tests",
+			handler: async () => {
+				callIndex++;
+				if (callIndex === 1) {
+					firstStarted.resolve();
+					await releaseFirst.promise;
+				}
+				return { content: [`reply ${callIndex}`] };
+			},
+		});
+
+		const first = streamSimple(mock.model, context(), { maxInFlightRequests: { tests: 1 } });
+		const firstResult = first.result();
+		await firstStarted.promise;
+
+		const controller = new AbortController();
+		const second = streamSimple(mock.model, context(), {
+			maxInFlightRequests: { tests: 1 },
+			signal: controller.signal,
+		});
+		controller.abort(new Error("cancel queued request"));
+
+		await expect(second.result()).rejects.toThrow("cancel queued request");
+		expect(mock.calls).toHaveLength(1);
+
+		releaseFirst.resolve();
+		await firstResult;
+		expect(mock.calls).toHaveLength(1);
+	});
+
+	test("shares limits with leases created by another process", async () => {
+		registerMockApi();
+		const providerDir = limiterDir("tests");
+		const externalLease = path.join(providerDir, "external");
+		await fs.mkdir(externalLease, { recursive: true });
+		await Bun.write(
+			path.join(externalLease, "info.json"),
+			JSON.stringify({ pid: process.pid, timestamp: Date.now(), token: "external" }),
+		);
+
+		const controller = new AbortController();
+		const mock = createMockModel({ provider: "tests", responses: [{ content: ["reply"] }] });
+		const stream = streamSimple(mock.model, context(), {
+			maxInFlightRequests: { tests: 1 },
+			signal: controller.signal,
+		});
+
+		await Bun.sleep(150);
+		expect(mock.calls).toHaveLength(0);
+
+		await fs.rm(externalLease, { recursive: true, force: true });
+		await Bun.write(path.join(providerDir, ".wakeup"), String(Date.now()));
+		const result = await stream.result();
+		expect(result.content).toEqual([{ type: "text", text: "reply" }]);
+		expect(mock.calls).toHaveLength(1);
+	});
+
+	test("does not signal waiters when no slot was freed", async () => {
+		registerMockApi();
+		const providerDir = limiterDir("tests");
+		const externalLease = path.join(providerDir, "external");
+		await fs.mkdir(externalLease, { recursive: true });
+		await Bun.write(
+			path.join(externalLease, "info.json"),
+			JSON.stringify({ pid: process.pid, timestamp: Date.now(), token: "external" }),
+		);
+
+		const controller = new AbortController();
+		const mock = createMockModel({ provider: "tests", responses: [{ content: ["reply"] }] });
+		const stream = streamSimple(mock.model, context(), {
+			maxInFlightRequests: { tests: 1 },
+			signal: controller.signal,
+		});
+
+		await Bun.sleep(50);
+		expect(await Bun.file(path.join(providerDir, ".wakeup")).exists()).toBe(false);
+		expect(mock.calls).toHaveLength(0);
+
+		controller.abort(new Error("cancel saturated waiter"));
+		await expect(stream.result()).rejects.toThrow("cancel saturated waiter");
+	});
+
+	test("does not reap a live lock just because its timestamp is old", async () => {
+		registerMockApi();
+		const lockDir = __providerInFlightForTesting.lockDir("tests");
+		await fs.mkdir(lockDir, { recursive: true });
+		await Bun.write(
+			path.join(lockDir, "info.json"),
+			JSON.stringify({ pid: process.pid, timestamp: Date.now() - 60_000, token: "live-lock" }),
+		);
+
+		const controller = new AbortController();
+		const mock = createMockModel({ provider: "tests", responses: [{ content: ["reply"] }] });
+		const stream = streamSimple(mock.model, context(), {
+			maxInFlightRequests: { tests: 1 },
+			signal: controller.signal,
+		});
+
+		await Bun.sleep(150);
+		expect(mock.calls).toHaveLength(0);
+
+		controller.abort(new Error("cancel lock waiter"));
+		await expect(stream.result()).rejects.toThrow("cancel lock waiter");
+		expect(mock.calls).toHaveLength(0);
+	});
+
+	test("treats unreadable fresh lease info as active", async () => {
+		registerMockApi();
+		const providerDir = limiterDir("tests");
+		const externalLease = path.join(providerDir, "partial-info");
+		await fs.mkdir(externalLease, { recursive: true });
+		const old = new Date(Date.now() - 60_000);
+		await fs.utimes(externalLease, old, old);
+		await Bun.write(path.join(externalLease, "info.json"), "{");
+
+		const controller = new AbortController();
+		const mock = createMockModel({ provider: "tests", responses: [{ content: ["reply"] }] });
+		const stream = streamSimple(mock.model, context(), {
+			maxInFlightRequests: { tests: 1 },
+			signal: controller.signal,
+		});
+
+		await Bun.sleep(150);
+		expect(mock.calls).toHaveLength(0);
+
+		controller.abort(new Error("cancel partial-info waiter"));
+		await expect(stream.result()).rejects.toThrow("cancel partial-info waiter");
+		expect(mock.calls).toHaveLength(0);
+	});
+
+	test("does not delete a fresh lock after observing a stale lock", async () => {
+		const lockDir = __providerInFlightForTesting.lockDir("tests");
+		await fs.mkdir(lockDir, { recursive: true });
+		await Bun.write(
+			path.join(lockDir, "info.json"),
+			JSON.stringify({ pid: 999999, timestamp: Date.now() - 60_000, token: "stale-lock" }),
+		);
+		const staleRelease = await __providerInFlightForTesting.captureStaleLockRelease("tests");
+		expect(staleRelease).not.toBeNull();
+
+		await fs.rm(lockDir, { recursive: true, force: true });
+		await fs.mkdir(lockDir, { recursive: true });
+		await Bun.write(
+			path.join(lockDir, "info.json"),
+			JSON.stringify({ pid: process.pid, timestamp: Date.now(), token: "fresh-lock" }),
+		);
+
+		await staleRelease?.();
+
+		const remaining = JSON.parse(await Bun.file(path.join(lockDir, "info.json")).text()) as { token: string };
+		expect(remaining.token).toBe("fresh-lock");
+	});
+
+	test("does not dispatch when aborted immediately after slot acquisition", async () => {
+		registerMockApi();
+		const controller = new AbortController();
+		const mock = createMockModel({ provider: "tests", responses: [{ content: ["reply"] }] });
+		const stream = streamSimple(mock.model, context(), {
+			maxInFlightRequests: { tests: 1 },
+			signal: controller.signal,
+		});
+
+		controller.abort(new Error("cancel acquired request"));
+
+		await expect(stream.result()).rejects.toThrow("cancel acquired request");
+		expect(mock.calls).toHaveLength(0);
+	});
+
+	test("uses opaque path segments for provider ids", async () => {
+		const dir = limiterDir("..");
+		const relative = path.relative(limiterRoot!, dir);
+
+		expect(relative).not.toBe("");
+		expect(relative.startsWith("..")).toBe(false);
+		expect(path.isAbsolute(relative)).toBe(false);
+	});
+});
diff --git a/packages/coding-agent/CHANGELOG.md b/packages/coding-agent/CHANGELOG.md
index 1fc767cea..d2b5a61a8 100644
--- a/packages/coding-agent/CHANGELOG.md
+++ b/packages/coding-agent/CHANGELOG.md
@@ -2,6 +2,10 @@
 
 ## [Unreleased]
 
+### Added
+
+- Added `providers.maxInFlightRequests` to cap concurrent LLM requests per provider across local OMP processes from settings.
+
 ## [16.1.11] - 2026-06-21
 
 ### Added
diff --git a/packages/coding-agent/src/cli/config-cli.ts b/packages/coding-agent/src/cli/config-cli.ts
index c2fd6025e..b9ecf3ca9 100644
--- a/packages/coding-agent/src/cli/config-cli.ts
+++ b/packages/coding-agent/src/cli/config-cli.ts
@@ -16,6 +16,7 @@ import {
 	Settings,
 	type SettingValue,
 	settings,
+	validateProviderMaxInFlightRequests,
 } from "../config/settings";
 import { SETTINGS_SCHEMA } from "../config/settings-schema";
 import { theme } from "../modes/theme/theme";
@@ -218,6 +219,9 @@ function parseAndSetValue(path: SettingPath, rawValue: string): void {
 			if (parsed === null || typeof parsed !== "object" || Array.isArray(parsed)) {
 				throw new Error(`Invalid record JSON: ${rawValue}`);
 			}
+			if (path === "providers.maxInFlightRequests") {
+				parsed = validateProviderMaxInFlightRequests(parsed);
+			}
 			parsedValue = parsed;
 			break;
 		}
diff --git a/packages/coding-agent/src/config/settings-schema.ts b/packages/coding-agent/src/config/settings-schema.ts
index 86103e0af..72bb9398d 100644
--- a/packages/coding-agent/src/config/settings-schema.ts
+++ b/packages/coding-agent/src/config/settings-schema.ts
@@ -272,6 +272,7 @@ export interface ModelTagsSettings {
 // under `as const` while still letting SettingValue infer the correct element type.
 const EMPTY_STRING_ARRAY: string[] = [];
 const EMPTY_STRING_RECORD: Record<string, string> = {};
+const EMPTY_NUMBER_RECORD: Record<string, number> = {};
 const DEFAULT_CYCLE_ORDER: string[] = ["smol", "default", "slow"];
 const EMPTY_MODEL_TAGS_RECORD: ModelTagsSettings = {};
 const HINDSIGHT_RECALL_TYPES_DEFAULT: string[] = ["world", "experience"];
@@ -449,6 +450,18 @@ export const SETTINGS_SCHEMA = {
 
 	disabledProviders: { type: "array", default: EMPTY_STRING_ARRAY },
 
+	"providers.maxInFlightRequests": {
+		type: "record",
+		default: EMPTY_NUMBER_RECORD,
+		ui: {
+			tab: "providers",
+			group: "Services",
+			label: "Max In-Flight Requests",
+			description:
+				'Maximum concurrent LLM requests per provider id (for example "openai" or "anthropic"), shared across local OMP processes with this config root. Omitted providers are unlimited.',
+		},
+	},
+
 	disabledExtensions: { type: "array", default: EMPTY_STRING_ARRAY },
 
 	modelRoles: { type: "record", default: EMPTY_STRING_RECORD },
diff --git a/packages/coding-agent/src/config/settings.ts b/packages/coding-agent/src/config/settings.ts
index 58f0398f8..a0972f5e6 100644
--- a/packages/coding-agent/src/config/settings.ts
+++ b/packages/coding-agent/src/config/settings.ts
@@ -14,6 +14,7 @@
 import * as fs from "node:fs";
 import * as os from "node:os";
 import * as path from "node:path";
+import { configureProviderMaxInFlightRequests } from "@oh-my-pi/pi-ai/stream";
 import {
 	getAgentDbPath,
 	getAgentDir,
@@ -105,6 +106,33 @@ function setByPath(obj: RawSettings, segments: string[], value: unknown): void {
 	current[segments[segments.length - 1]] = value;
 }
 
+export function normalizeProviderMaxInFlightRequests(value: unknown): Record<string, number> {
+	if (!value || typeof value !== "object" || Array.isArray(value)) return {};
+	const normalized: Record<string, number> = {};
+	for (const [provider, rawLimit] of Object.entries(value)) {
+		if (typeof rawLimit !== "number" || !Number.isFinite(rawLimit) || rawLimit <= 0) continue;
+		normalized[provider] = Math.max(1, Math.floor(rawLimit));
+	}
+	return normalized;
+}
+
+export function validateProviderMaxInFlightRequests(value: unknown): Record<string, number> {
+	if (!value || typeof value !== "object" || Array.isArray(value)) return {};
+	const invalidProviders: string[] = [];
+	const normalized: Record<string, number> = {};
+	for (const [provider, rawLimit] of Object.entries(value)) {
+		if (typeof rawLimit !== "number" || !Number.isFinite(rawLimit) || rawLimit <= 0) {
+			invalidProviders.push(provider);
+			continue;
+		}
+		normalized[provider] = Math.max(1, Math.floor(rawLimit));
+	}
+	if (invalidProviders.length > 0) {
+		throw new Error(`Provider request limits must be positive numbers: ${invalidProviders.join(", ")}`);
+	}
+	return normalized;
+}
+
 const PATH_SCOPED_ARRAY_SETTINGS = new Set<SettingPath>(["enabledModels", "disabledProviders"]);
 type PathScopedStringArrayEntry = {
 	path?: unknown;
@@ -336,7 +364,7 @@ export class Settings {
 		// Trigger hook if exists
 		const hook = SETTING_HOOKS[path];
 		if (hook) {
-			hook(value, prev);
+			hook(next, prev);
 		}
 		this.#fireEffectiveSettingChanged(path, next, prev);
 	}
@@ -1147,6 +1175,9 @@ const SETTING_HOOKS: Partial<Record<SettingPath, SettingHook<any>>> = {
 			appendOnlyModeSignal.fire(value);
 		}
 	},
+	"providers.maxInFlightRequests": value => {
+		configureProviderMaxInFlightRequests(validateProviderMaxInFlightRequests(value));
+	},
 	"hindsight.bankId": () => hindsightScopeSignal.fire(),
 	"hindsight.bankIdPrefix": () => hindsightScopeSignal.fire(),
 	"hindsight.scoping": () => hindsightScopeSignal.fire(),
@@ -1211,6 +1242,7 @@ export function resetSettingsForTest(): void {
 	globalInstance = null;
 	globalInstancePromise = null;
 	clearBoundSettingsMethods();
+	configureProviderMaxInFlightRequests(undefined);
 }
 
 /**
diff --git a/packages/coding-agent/src/modes/components/settings-defs.ts b/packages/coding-agent/src/modes/components/settings-defs.ts
index 3c9d54577..582877281 100644
--- a/packages/coding-agent/src/modes/components/settings-defs.ts
+++ b/packages/coding-agent/src/modes/components/settings-defs.ts
@@ -68,7 +68,16 @@ export interface TextInputSettingDef extends BaseSettingDef {
 	type: "text";
 }
 
-export type SettingDef = BooleanSettingDef | EnumSettingDef | SubmenuSettingDef | TextInputSettingDef;
+export interface ProviderLimitsSettingDef extends BaseSettingDef {
+	type: "providerLimits";
+}
+
+export type SettingDef =
+	| BooleanSettingDef
+	| EnumSettingDef
+	| SubmenuSettingDef
+	| TextInputSettingDef
+	| ProviderLimitsSettingDef;
 
 // ═══════════════════════════════════════════════════════════════════════════
 // Condition Functions
@@ -170,6 +179,10 @@ function pathToSettingDef(path: SettingPath): SettingDef | null {
 		return { ...base, type: "text" };
 	}
 
+	if (schemaType === "record") {
+		return path === "providers.maxInFlightRequests" ? { ...base, type: "providerLimits" } : null;
+	}
+
 	return null;
 }
 
diff --git a/packages/coding-agent/src/modes/components/settings-selector.ts b/packages/coding-agent/src/modes/components/settings-selector.ts
index dbc60a64e..f9ea39a63 100644
--- a/packages/coding-agent/src/modes/components/settings-selector.ts
+++ b/packages/coding-agent/src/modes/components/settings-selector.ts
@@ -11,6 +11,7 @@ import {
 	Input,
 	matchesKey,
 	parseSgrMouse,
+	replaceTabs,
 	type SelectItem,
 	SelectList,
 	type SettingItem,
@@ -24,7 +25,14 @@ import {
 	visibleWidth,
 } from "@oh-my-pi/pi-tui";
 import type { ShapeTarget } from "@oh-my-pi/snapcompact";
-import { getDefault, type SettingPath, settings } from "../../config/settings";
+import {
+	getDefault,
+	getType,
+	normalizeProviderMaxInFlightRequests,
+	type SettingPath,
+	settings,
+	validateProviderMaxInFlightRequests,
+} from "../../config/settings";
 import type {
 	SettingTab,
 	StatusLinePreset,
@@ -50,6 +58,7 @@ import { getPreset } from "./status-line/presets";
  */
 class TextInputSubmenu extends Container {
 	#input: Input;
+	#error: Text;
 
 	constructor(
 		label: string,
@@ -71,11 +80,18 @@ class TextInputSubmenu extends Container {
 		if (currentValue) {
 			this.#input.setValue(currentValue);
 		}
+		this.#error = new Text("", 0, 0);
 		this.#input.onSubmit = value => {
-			this.onSubmit(value); // empty string clears the setting
+			try {
+				this.onSubmit(value); // empty string clears the setting
+			} catch (error) {
+				const message = error instanceof Error ? error.message : String(error);
+				this.#error.setText(theme.fg("error", truncateToWidth(replaceTabs(message).replace(/[\r\n]+/g, " "), 100)));
+			}
 		};
 		this.addChild(this.#input);
 		this.addChild(new Spacer(1));
+		this.addChild(this.#error);
 		this.addChild(new Text(theme.fg("dim", "  Enter to save · Esc to cancel · Clear field to unset"), 0, 0));
 	}
 
@@ -217,6 +233,113 @@ class SelectSubmenu extends Container {
 	}
 }
 
+class ProviderLimitsSubmenu extends Container {
+	#selectList: SelectList | undefined;
+
+	constructor(
+		private readonly providers: readonly string[],
+		private readonly onChange: (value: Record<string, number>) => void,
+		private readonly onCancel: () => void,
+		private readonly requestRender?: () => void,
+	) {
+		super();
+		this.#showProviderList();
+	}
+
+	#providerIds(): string[] {
+		const limits = normalizeProviderMaxInFlightRequests(settings.get("providers.maxInFlightRequests"));
+		return [...new Set([...this.providers, ...Object.keys(limits)])].sort((a, b) => a.localeCompare(b));
+	}
+
+	#showProviderList(): void {
+		this.clear();
+		this.addChild(new Text(theme.bold(theme.fg("accent", "Max In-Flight Requests")), 0, 0));
+		this.addChild(new Spacer(1));
+		this.addChild(
+			new Text(
+				theme.fg(
+					"muted",
+					"Select a provider, enter a positive number to cap concurrent LLM requests, or clear it for unlimited.",
+				),
+				0,
+				0,
+			),
+		);
+		this.addChild(new Spacer(1));
+
+		const limits = normalizeProviderMaxInFlightRequests(settings.get("providers.maxInFlightRequests"));
+		const providerItems = this.#providerIds().map((provider): SelectItem => {
+			const limit = limits[provider];
+			return {
+				value: provider,
+				label: provider,
+				description: limit === undefined ? "Unlimited" : `Limit: ${limit}`,
+			};
+		});
+		const clearItem: SelectItem[] =
+			Object.keys(limits).length === 0
+				? []
+				: [{ value: "__clear_all", label: "Clear all limits", description: "Make every provider unlimited" }];
+		const items = [...providerItems, ...clearItem];
+		this.#selectList = new SelectList(items, Math.min(Math.max(items.length, 1), 12), getSelectListTheme());
+		this.#selectList.onSelect = item => {
+			if (item.value === "__clear_all") {
+				settings.set("providers.maxInFlightRequests", {});
+				this.onChange({});
+				this.#showProviderList();
+				this.requestRender?.();
+				return;
+			}
+			this.#showProviderEditor(item.value);
+		};
+		this.#selectList.onCancel = this.onCancel;
+		this.addChild(this.#selectList);
+		this.addChild(new Spacer(1));
+		this.addChild(new Text(theme.fg("dim", "  Enter to edit provider · Esc to go back"), 0, 0));
+	}
+
+	#showProviderEditor(provider: string): void {
+		const limits = normalizeProviderMaxInFlightRequests(settings.get("providers.maxInFlightRequests"));
+		this.clear();
+		this.#selectList = undefined;
+		this.addChild(
+			new TextInputSubmenu(
+				`Max In-Flight Requests: ${provider}`,
+				"Enter a positive number. Decimals round down. Clear the field to make this provider unlimited.",
+				limits[provider]?.toString() ?? "",
+				value => {
+					const next = { ...limits };
+					const trimmed = value.trim();
+					if (trimmed === "") {
+						delete next[provider];
+					} else {
+						const limit = Number(trimmed);
+						if (!Number.isFinite(limit) || limit <= 0) throw new Error("Limit must be a positive number.");
+						next[provider] = Math.max(1, Math.floor(limit));
+					}
+					const normalized = validateProviderMaxInFlightRequests(next);
+					settings.set("providers.maxInFlightRequests", normalized);
+					this.onChange(normalized);
+					this.#showProviderList();
+					this.requestRender?.();
+				},
+				() => {
+					this.#showProviderList();
+					this.requestRender?.();
+				},
+			),
+		);
+	}
+
+	handleInput(data: string): void {
+		if (this.#selectList) {
+			this.#selectList.handleInput(data);
+			return;
+		}
+		this.children[0]?.handleInput?.(data);
+	}
+}
+
 let cachedSidebarWidth: number | undefined;
 /**
  * Split-sidebar width derived from every group name in the schema (not just
@@ -258,6 +381,8 @@ export interface SettingsRuntimeContext {
 	thinkingLevel: ThinkingLevel | undefined;
 	/** Available themes */
 	availableThemes: string[];
+	/** Provider/source ids shown in /model. */
+	providers: string[];
 	/** Working directory for plugins tab */
 	cwd: string;
 	/** Active model (api + id); resolves what the snapcompact `auto` shape maps to. */
@@ -721,10 +846,20 @@ export class SettingsSelectorComponent implements Component {
 					id: def.path,
 					label: def.label,
 					description: def.description,
-					currentValue: (currentValue as string) ?? "",
+					currentValue: this.#formatTextInputValue(def.path, currentValue),
 					submenu: (cv, done) => this.#createTextInput(def, cv, done),
 					changed,
 				};
+
+			case "providerLimits":
+				return {
+					id: def.path,
+					label: def.label,
+					description: def.description,
+					currentValue: this.#formatProviderLimitsValue(currentValue),
+					submenu: (_cv, done) => this.#createProviderLimitsInput(done),
+					changed,
+				};
 		}
 	}
 
@@ -854,7 +989,7 @@ export class SettingsSelectorComponent implements Component {
 	 */
 	#createTextInput(
 		def: SettingDef & { type: "text" },
-		currentValue: string,
+		_currentValue: string,
 		done: (value?: string) => void,
 	): Container {
 		this.#textInputActive = true;
@@ -865,28 +1000,72 @@ export class SettingsSelectorComponent implements Component {
 		return new TextInputSubmenu(
 			def.label,
 			def.description,
-			currentValue,
+			this.#formatTextInputEditValue(def.path, settings.get(def.path)),
 			value => {
 				// Empty string clears the setting; undefined-typed string settings
 				// store "" which the browser.ts expandPath ignores (no-op fallback).
 				this.#setSettingValue(def.path, value);
-				this.callbacks.onChange(def.path, value);
-				wrappedDone(value);
+				this.callbacks.onChange(def.path, settings.get(def.path));
+				wrappedDone(this.#formatTextInputValue(def.path, settings.get(def.path)));
 			},
 			() => wrappedDone(),
 		);
 	}
 
+	#createProviderLimitsInput(done: (value?: string) => void): Container {
+		return new ProviderLimitsSubmenu(
+			this.context.providers,
+			value => {
+				this.callbacks.onChange("providers.maxInFlightRequests", value);
+				done(this.#formatProviderLimitsValue(value));
+			},
+			() => done(),
+			this.context.requestRender,
+		);
+	}
+
+	#formatProviderLimitsValue(value: unknown): string {
+		const limits = normalizeProviderMaxInFlightRequests(value);
+		const entries = Object.entries(limits).sort(([a], [b]) => a.localeCompare(b));
+		if (entries.length === 0) return "Unlimited";
+		return entries.map(([provider, limit]) => `${provider}: ${limit}`).join(", ");
+	}
+
+	#formatTextInputValue(path: SettingPath, value: unknown): string {
+		if (path === "providers.maxInFlightRequests") return this.#formatProviderLimitsValue(value);
+		return this.#formatTextInputEditValue(path, value);
+	}
+
+	#formatTextInputEditValue(_path: SettingPath, value: unknown): string {
+		if (value === undefined || value === null) return "";
+		if (typeof value === "object") return JSON.stringify(value);
+		return String(value);
+	}
+
 	/**
 	 * Set a setting value, handling type conversion.
 	 */
 	#setSettingValue(path: SettingPath, value: string): void {
-		// Handle number conversions
 		const currentValue = settings.get(path);
+		const schemaType = getType(path);
 		if (path === "compaction.thresholdPercent" && value === "default") {
 			settings.set(path, -1 as never);
 		} else if (path === "compaction.thresholdTokens" && value === "default") {
 			settings.set(path, -1 as never);
+		} else if (schemaType === "record") {
+			let parsed: unknown;
+			try {
+				parsed = JSON.parse(value || "{}");
+			} catch {
+				throw new Error(`Invalid record JSON for ${path}`);
+			}
+			if (!parsed || typeof parsed !== "object" || Array.isArray(parsed)) {
+				throw new Error(`Invalid record JSON for ${path}`);
+			}
+			if (path === "providers.maxInFlightRequests") {
+				parsed = validateProviderMaxInFlightRequests(parsed);
+			}
+			settings.set(path, parsed as never);
 		} else if (typeof currentValue === "number") {
 			settings.set(path, Number(value) as never);
 		} else if (typeof currentValue === "boolean") {
diff --git a/packages/coding-agent/src/modes/controllers/selector-controller.ts b/packages/coding-agent/src/modes/controllers/selector-controller.ts
index 7ecf68e90..5ea909e17 100644
--- a/packages/coding-agent/src/modes/controllers/selector-controller.ts
+++ b/packages/coding-agent/src/modes/controllers/selector-controller.ts
@@ -117,6 +117,9 @@ export class SelectorController {
 					availableThinkingLevels: [...this.ctx.session.getAvailableThinkingLevels()],
 					thinkingLevel: this.ctx.session.thinkingLevel,
 					availableThemes,
+					providers: [...new Set(this.ctx.session.getAvailableModels().map(model => model.provider))].sort(
+						(a, b) => a.localeCompare(b),
+					),
 					cwd: getProjectDir(),
 					model: this.ctx.session.model,
 					imageBudget: this.ctx.ui.imageBudget,
diff --git a/packages/coding-agent/src/sdk.ts b/packages/coding-agent/src/sdk.ts
index a2640858a..05819c48f 100644
--- a/packages/coding-agent/src/sdk.ts
+++ b/packages/coding-agent/src/sdk.ts
@@ -42,7 +42,7 @@ import {
 	resolveModelRoleValue,
 } from "./config/model-resolver";
 import { loadPromptTemplates as loadPromptTemplatesInternal, type PromptTemplate } from "./config/prompt-templates";
-import { Settings, type SkillsSettings } from "./config/settings";
+import { Settings, type SkillsSettings, validateProviderMaxInFlightRequests } from "./config/settings";
 import { CursorExecHandlers } from "./cursor";
 import "./discovery";
 import { initializeWithSettings } from "./discovery";
@@ -2520,6 +2520,9 @@ export async function createAgentSession(options: CreateAgentSessionOptions = {}
 					...streamOptions,
 					openrouterVariant: streamOptions?.openrouterVariant ?? openrouterVariant,
 					antigravityEndpointMode: streamOptions?.antigravityEndpointMode ?? antigravityEndpointMode,
+					maxInFlightRequests: validateProviderMaxInFlightRequests(
+						streamOptions?.maxInFlightRequests ?? settings.get("providers.maxInFlightRequests"),
+					),
 					loopGuard: {
 						enabled: settings.get("model.loopGuard.enabled"),
 						checkAssistantContent: settings.get("model.loopGuard.checkAssistantContent"),
diff --git a/packages/coding-agent/src/session/agent-session.ts b/packages/coding-agent/src/session/agent-session.ts
index a6b91853c..0ef5ff2a5 100644
--- a/packages/coding-agent/src/session/agent-session.ts
+++ b/packages/coding-agent/src/session/agent-session.ts
@@ -159,7 +159,7 @@ import {
 import { MODEL_ROLE_IDS, MODEL_ROLES } from "../config/model-roles";
 import { expandPromptTemplate, type PromptTemplate } from "../config/prompt-templates";
 import type { Settings, SkillsSettings } from "../config/settings";
-import { getDefault, onAppendOnlyModeChanged } from "../config/settings";
+import { getDefault, onAppendOnlyModeChanged, validateProviderMaxInFlightRequests } from "../config/settings";
 import { RawSseDebugBuffer } from "../debug/raw-sse-buffer";
 import { loadCapability } from "../discovery";
 import { expandApplyPatchToEntries, normalizeDiff, normalizeToLF, ParseError, previewPatch, stripBom } from "../edit";
@@ -5298,6 +5298,9 @@ export class AgentSession {
 			...options,
 			...(openrouterVariant !== undefined && { openrouterVariant }),
 			...(antigravityEndpointMode !== undefined && { antigravityEndpointMode }),
+			maxInFlightRequests: validateProviderMaxInFlightRequests(
+				options.maxInFlightRequests ?? this.settings.get("providers.maxInFlightRequests"),
+			),
 			loopGuard: {
 				enabled: this.settings.get("model.loopGuard.enabled"),
 				checkAssistantContent: this.settings.get("model.loopGuard.checkAssistantContent"),
diff --git a/packages/coding-agent/test/config-cli.test.ts b/packages/coding-agent/test/config-cli.test.ts
index 76715aa74..f102fac72 100644
--- a/packages/coding-agent/test/config-cli.test.ts
+++ b/packages/coding-agent/test/config-cli.test.ts
@@ -66,6 +66,46 @@ describe("config CLI schema coverage", () => {
 		expect(parsed.value).toEqual({ default: "claude-opus-4-6" });
 	});
 
+	it("normalizes valid provider in-flight request limits from JSON objects", async () => {
+		const logSpy = vi.spyOn(console, "log").mockImplementation(() => {});
+
+		await runConfigCommand({
+			action: "set",
+			key: "providers.maxInFlightRequests",
+			value: '{"openai":2.8,"anthropic":1}',
+			flags: { json: true },
+		});
+		await runConfigCommand({ action: "get", key: "providers.maxInFlightRequests", flags: { json: true } });
+
+		const payload = logSpy.mock.calls.at(-1)?.[0];
+		expect(typeof payload).toBe("string");
+		const parsed = JSON.parse(String(payload)) as { key: string; value: unknown; type: string };
+		expect(parsed.key).toBe("providers.maxInFlightRequests");
+		expect(parsed.type).toBe("record");
+		expect(parsed.value).toEqual({ openai: 2, anthropic: 1 });
+	});
+
+	it("rejects invalid provider in-flight request limit entries", async () => {
+		vi.spyOn(console, "log").mockImplementation(() => {});
+		vi.spyOn(console, "error").mockImplementation(() => {});
+		const exitSpy = vi.spyOn(process, "exit").mockImplementation((() => {
+			throw new Error("process.exit");
+		}) as typeof process.exit);
+
+		await expect(
+			runConfigCommand({
+				action: "set",
+				key: "providers.maxInFlightRequests",
+				value: '{"openai":"2","anthropic":0}',
+				flags: { json: true },
+			}),
+		).rejects.toThrow("process.exit");
+		expect(exitSpy).toHaveBeenCalledWith(1);
+		expect(console.error).toHaveBeenCalledWith(
+			expect.stringContaining("Provider request limits must be positive numbers: openai, anthropic"),
+		);
+	});
+
 	it("sets and gets array settings as JSON arrays", async () => {
 		const logSpy = vi.spyOn(console, "log").mockImplementation(() => {});
 		const arrayValue = '["claude-opus-4-6","gpt-5.3-codex"]';
diff --git a/packages/coding-agent/test/modes/components/settings-layout.test.ts b/packages/coding-agent/test/modes/components/settings-layout.test.ts
index f9fec2699..a669e6d71 100644
--- a/packages/coding-agent/test/modes/components/settings-layout.test.ts
+++ b/packages/coding-agent/test/modes/components/settings-layout.test.ts
@@ -76,4 +76,15 @@ describe("settings layout", () => {
 			expect(def.condition?.()).toBe(true);
 		}
 	});
+
+	it("shows provider request limits as a providers services submenu setting", () => {
+		const [def] = getSettingsForTab("providers").filter(item => item.path === "providers.maxInFlightRequests");
+
+		expect(def).toMatchObject({
+			path: "providers.maxInFlightRequests",
+			type: "providerLimits",
+			tab: "providers",
+			group: "Services",
+		});
+	});
 });
diff --git a/packages/coding-agent/test/modes/components/settings-selector-memory-refresh.test.ts b/packages/coding-agent/test/modes/components/settings-selector-memory-refresh.test.ts
index 369a072c9..a288e481f 100644
--- a/packages/coding-agent/test/modes/components/settings-selector-memory-refresh.test.ts
+++ b/packages/coding-agent/test/modes/components/settings-selector-memory-refresh.test.ts
@@ -45,6 +45,7 @@ function createSelector(onCancel: () => void = () => {}): SettingsSelectorCompon
 			availableThinkingLevels: [],
 			thinkingLevel: undefined,
 			availableThemes: ["dark"],
+			providers: [],
 			cwd: process.cwd(),
 		},
 		{
diff --git a/packages/coding-agent/test/settings-manager.test.ts b/packages/coding-agent/test/settings-manager.test.ts
index dce19a62f..b64ea5c96 100644
--- a/packages/coding-agent/test/settings-manager.test.ts
+++ b/packages/coding-agent/test/settings-manager.test.ts
@@ -2,6 +2,10 @@ import { afterEach, beforeEach, describe, expect, it } from "bun:test";
 import * as fs from "node:fs";
 import * as path from "node:path";
 import { Effort } from "@oh-my-pi/pi-ai";
+import { clearCustomApis } from "@oh-my-pi/pi-ai/api-registry";
+import { createMockModel, registerMockApi } from "@oh-my-pi/pi-ai/providers/mock";
+import { __providerInFlightForTesting, streamSimple } from "@oh-my-pi/pi-ai/stream";
+import type { Context } from "@oh-my-pi/pi-ai/types";
 import {
 	getDefault,
 	getEnumValues,
@@ -11,10 +15,18 @@ import {
 	type SettingPath,
 	Settings,
 } from "@oh-my-pi/pi-coding-agent/config/settings";
+import { AgentStorage } from "@oh-my-pi/pi-coding-agent/session/agent-storage";
 import { getProjectAgentDir, TempDir } from "@oh-my-pi/pi-utils";
 import { YAML } from "bun";
 import { beginSettingsTest, restoreSettingsTestState, type SettingsTestState } from "./helpers/settings-test-state";
 
+function context(): Context {
+	return {
+		systemPrompt: [],
+		messages: [{ role: "user", content: "hi", timestamp: 0 }],
+	};
+}
+
 describe("Settings", () => {
 	let settingsState: SettingsTestState | undefined;
 	let tempDir: TempDir;
@@ -49,10 +61,14 @@ describe("Settings", () => {
 		return parsed as Record<string, unknown>;
 	};
 
-	afterEach(() => {
+	afterEach(async () => {
+		clearCustomApis();
+		__providerInFlightForTesting.setRoot(undefined);
+		AgentStorage.resetInstance();
 		restoreSettingsTestState(settingsState);
 		settingsState = undefined;
-		tempDir?.removeSync();
+		await Bun.sleep(0);
+		await tempDir?.remove();
 	});
 	describe("defaults", () => {
 		it("keeps eight inline images live by default", async () => {
@@ -66,6 +82,12 @@ describe("Settings", () => {
 			expect(getDefault("startup.showSplash")).toBe(false);
 		});
 
+		it("defaults provider in-flight request limits to an empty map", async () => {
+			const settings = Settings.isolated();
+			expect(settings.get("providers.maxInFlightRequests")).toEqual({});
+			expect(getDefault("providers.maxInFlightRequests")).toEqual({});
+		});
+
 		it("exposes all tool calling mode options", () => {
 			const values = getEnumValues("tools.format");
 			expect(values).toEqual([
@@ -585,5 +607,78 @@ describe("Settings", () => {
 			const settings = await Settings.init({ cwd: projectDir, agentDir });
 			expect(settings.get("power.sleepPrevention")).toBe("off");
 		});
+
+		describe("provider request limits", () => {
+			it("uses the effective merged value when configuring hooks", async () => {
+				const settings = Settings.isolated({ "providers.maxInFlightRequests": { openai: 1 } });
+				__providerInFlightForTesting.setRoot(tempDir.join("provider-inflight"));
+				registerMockApi();
+				const firstStarted = Promise.withResolvers<void>();
+				const releaseFirst = Promise.withResolvers<void>();
+				let active = 0;
+				let maxActive = 0;
+				let callIndex = 0;
+				const mock = createMockModel({
+					provider: "openai",
+					handler: async () => {
+						callIndex++;
+						active++;
+						maxActive = Math.max(maxActive, active);
+						try {
+							if (callIndex === 1) {
+								firstStarted.resolve();
+								await releaseFirst.promise;
+							}
+							return { content: [`reply ${callIndex}`] };
+						} finally {
+							active--;
+						}
+					},
+				});
+
+				settings.set("providers.maxInFlightRequests", { openai: 4 });
+
+				const first = streamSimple(mock.model, context());
+				const firstResult = first.result();
+				await firstStarted.promise;
+				const second = streamSimple(mock.model, context());
+				await Bun.sleep(20);
+
+				expect(settings.get("providers.maxInFlightRequests")).toEqual({ openai: 1 });
+				expect(mock.calls).toHaveLength(1);
+
+				releaseFirst.resolve();
+				await Promise.all([firstResult, second.result()]);
+				expect(maxActive).toBe(1);
+			});
+
+			it("rejects invalid provider limits from config.yml", async () => {
+				await writeSettings({ providers: { maxInFlightRequests: { openai: "2" } } });
+
+				await expect(Settings.init({ cwd: projectDir, agentDir })).rejects.toThrow(
+					"Provider request limits must be positive numbers: openai",
+				);
+			});
+
+			it("rejects invalid provider limits from project settings", async () => {
+				await Bun.write(
+					path.join(getProjectAgentDir(projectDir), "settings.json"),
+					JSON.stringify({ providers: { maxInFlightRequests: { anthropic: 0 } } }),
+				);
+
+				await expect(Settings.init({ cwd: projectDir, agentDir, inMemory: true })).rejects.toThrow(
+					"Provider request limits must be positive numbers: anthropic",
+				);
+			});
+
+			it("rejects invalid provider limits from config overlays", async () => {
+				const overlayPath = tempDir.join("overlay.yml");
+				await Bun.write(overlayPath, YAML.stringify({ providers: { maxInFlightRequests: { umans: -1 } } }));
+
+				await expect(
+					Settings.init({ cwd: projectDir, agentDir, inMemory: true, configFiles: [overlayPath] }),
+				).rejects.toThrow("Provider request limits must be positive numbers: umans");
+			});
+		});
 	});
 });