aallam · aallam · May 1, 2026 · May 1, 2026 · May 1, 2026
diff --git a/.changeset/drop-quickjs-process-host.md b/.changeset/drop-quickjs-process-host.md
@@ -0,0 +1,5 @@
+---
+"@execbox/quickjs": minor
+---
+
+Remove `QuickJsExecutor({ host: "process" })` and its process-hosted runner entrypoint. Use `host: "worker"` for local hosted QuickJS execution, or `@execbox/remote` for app-owned process, container, or VM boundaries.
diff --git a/README.md b/README.md
@@ -10,14 +10,14 @@ Portable code execution for [Model Context Protocol](https://modelcontextprotoco
 
 </div>
 
-Execbox turns host tool catalogs into callable guest namespaces, supports MCP wrapping on both sides of the boundary, and lets you place guest JavaScript where it fits your deployment: inline, behind a worker or child-process host, or across your own remote transport.
+Execbox turns host tool catalogs into callable guest namespaces, supports MCP wrapping on both sides of the boundary, and lets you place guest JavaScript where it fits your deployment: inline, behind a worker host, or across your own remote transport.
 
 ## Package Map
 
 | Package                                   | npm                                                                                                                           | What it is for                                                                            |
 | ----------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------- |
 | [`@execbox/core`](./packages/core/)       | [![npm](https://img.shields.io/npm/v/%40execbox%2Fcore?style=flat-square)](https://www.npmjs.com/package/@execbox/core)       | Core execution contract, provider resolution, MCP adapters, and runtime/protocol subpaths |
-| [`@execbox/quickjs`](./packages/quickjs/) | [![npm](https://img.shields.io/npm/v/%40execbox%2Fquickjs?style=flat-square)](https://www.npmjs.com/package/@execbox/quickjs) | QuickJS executor for inline, worker, and process hosts                                    |
+| [`@execbox/quickjs`](./packages/quickjs/) | [![npm](https://img.shields.io/npm/v/%40execbox%2Fquickjs?style=flat-square)](https://www.npmjs.com/package/@execbox/quickjs) | QuickJS executor for inline and worker hosts                                              |
 | [`@execbox/remote`](./packages/remote/)   | [![npm](https://img.shields.io/npm/v/%40execbox%2Fremote?style=flat-square)](https://www.npmjs.com/package/@execbox/remote)   | Advanced transport-backed executor for app-owned runtime boundaries                       |
 
 ## Examples

diff --git a/benchmarks/benchmark.ts b/benchmarks/benchmark.ts
@@ -2,7 +2,7 @@
  * Comprehensive execbox performance benchmark suite.
  *
  * Covers:
- *   1. Single-execution latency across all in-process & out-of-process executors
+ *   1. Single-execution latency across inline and worker-hosted executors
  *   2. Cold-start vs warm (pooled/prewarmed) first-execution cost
  *   3. Tool-call overhead scaling (0, 1, 5, 10 tool calls)
  *   4. Schema validation overhead (with vs without input/output schemas)
@@ -349,7 +349,7 @@ async function suiteToolCalls(): Promise<void> {
     ).join(";\n");
   }
 
-  // Use quickjs (fastest) and worker-pooled (representative out-of-process)
+  // Use quickjs (fastest) and worker-pooled (representative hosted local mode)
   const testFactories = factories.filter(
     (f) => f.name === "quickjs (in-process)" || f.name === "worker (pooled)",
   );
@@ -507,31 +507,29 @@ async function suiteContention(): Promise<void> {
   const poolSizes = [1, 2, 4];
   const concurrency = 8;
 
-  for (const executor_type of ["worker", "process"] as const) {
-    console.log(`\n  [${executor_type}]`);
+  console.log("\n  [worker]");
 
-    for (const poolSize of poolSizes) {
-      const executor = createContentionExecutor(executor_type, poolSize);
-      try {
-        await executor.prewarm?.(poolSize);
-        const { totalMs, perRequestMs } = await benchConcurrent(
-          executor,
-          code,
-          [simpleProvider],
-          concurrency,
-          totalRuns,
-        );
-        const throughput = (totalRuns / totalMs) * 1000;
-        console.log(
-          `    pool_size=${String(poolSize).padEnd(2)} concurrency=${concurrency} ` +
-            `throughput=${throughput.toFixed(1)} exec/s ` +
-            `median=${fmt(median(perRequestMs))} ` +
-            `p95=${fmt(percentile(perRequestMs, 95))} ` +
-            `max=${fmt(max(perRequestMs))}`,
-        );
-      } finally {
-        await executor.dispose?.();
-      }
+  for (const poolSize of poolSizes) {
+    const executor = createContentionExecutor(poolSize);
+    try {
+      await executor.prewarm?.(poolSize);
+      const { totalMs, perRequestMs } = await benchConcurrent(
+        executor,
+        code,
+        [simpleProvider],
+        concurrency,
+        totalRuns,
+      );
+      const throughput = (totalRuns / totalMs) * 1000;
+      console.log(
+        `    pool_size=${String(poolSize).padEnd(2)} concurrency=${concurrency} ` +
+          `throughput=${throughput.toFixed(1)} exec/s ` +
+          `median=${fmt(median(perRequestMs))} ` +
+          `p95=${fmt(percentile(perRequestMs, 95))} ` +
+          `max=${fmt(max(perRequestMs))}`,
+      );
+    } finally {
+      await executor.dispose?.();
     }
   }
 }

diff --git a/benchmarks/config.ts b/benchmarks/config.ts
@@ -70,24 +70,6 @@ export function createBenchmarkFactories(): ExecutorFactory[] {
         }) as DisposableExecutor,
       supportsExplicitPrewarm: true,
     },
-    {
-      name: "process (ephemeral)",
-      create: () =>
-        new QuickJsExecutor({
-          host: "process",
-          mode: "ephemeral",
-        }) as DisposableExecutor,
-      supportsExplicitPrewarm: false,
-    },
-    {
-      name: "process (pooled)",
-      create: () =>
-        new QuickJsExecutor({
-          host: "process",
-          ...createPooledBenchmarkOptions(),
-        }) as DisposableExecutor,
-      supportsExplicitPrewarm: true,
-    },
   ];
 }
 
@@ -98,12 +80,7 @@ export function createMemoryBenchmarkFactories(
   return factories.filter((factory) => allowedNames.has(factory.name));
 }
 
-export function createContentionExecutor(
-  executorType: "worker" | "process",
-  poolSize: number,
-): DisposableExecutor {
+export function createContentionExecutor(poolSize: number): DisposableExecutor {
   const pool = createContentionPoolOptions(poolSize);
-  return executorType === "worker"
-    ? (new QuickJsExecutor({ host: "worker", pool }) as DisposableExecutor)
-    : (new QuickJsExecutor({ host: "process", pool }) as DisposableExecutor);
+  return new QuickJsExecutor({ host: "worker", pool }) as DisposableExecutor;
 }
diff --git a/benchmarks/results.md b/benchmarks/results.md
@@ -16,14 +16,11 @@ This file records the current local benchmark run. Treat the tables as measured
 | QuickJS (in-process) | **1.47ms**        | **2.71ms**           | **3.94ms**            |
 | Worker (ephemeral)   | 62.40ms           | 64.16ms              | 65.67ms               |
 | Worker (pooled)      | 1.53ms            | 2.81ms               | 4.10ms                |
-| Process (ephemeral)  | 202.37ms          | 204.10ms             | 207.70ms              |
-| Process (pooled)     | 1.61ms            | 2.92ms               | 4.18ms                |
 
 ### Notes
 
 - On this machine, warmed pooled executors stayed close to QuickJS for trivial scripts.
-- Ephemeral executors remained far slower than pooled executors because each execution still pays worker or process startup cost.
-- Process pooled stayed low-latency in median terms, but its process startup path remained much more expensive than worker startup.
+- Ephemeral executors remained far slower than pooled executors because each execution still pays worker startup cost.
 
 ---
 
@@ -36,12 +33,10 @@ Only pooled executors expose explicit `prewarm()`. QuickJS and ephemeral executo
 | QuickJS (in-process) | 1.46ms                  | N/A                     | N/A     |
 | Worker (ephemeral)   | 62.70ms                 | N/A                     | N/A     |
 | Worker (pooled)      | 62.30ms                 | 1.81ms                  | 97.1%   |
-| Process (ephemeral)  | 202.51ms                | N/A                     | N/A     |
-| Process (pooled)     | 202.31ms                | 2.32ms                  | 98.9%   |
 
 ### Notes
 
-- True `prewarm()` delivered the intended first-request behavior in this run: pooled worker and pooled process executors dropped from shell-plus-guest startup latency to low-single-digit warm execution latency.
+- True `prewarm()` delivered the intended first-request behavior in this run: pooled worker executors dropped from shell-plus-guest startup latency to low-single-digit warm execution latency.
 - `prewarm()` pays the host-shell and guest-startup path before live traffic arrives.
 
 ---
@@ -92,13 +87,11 @@ The pooled benchmark factories in this suite use a fixed `pool.maxSize: 2`.
 | -------------------- | --------------- | ------ | ------ | ---------- |
 | QuickJS (in-process) | **395.3**       | 684.6  | 1078.8 | **1635.1** |
 | Worker (pooled)      | 356.5           | 722.9  | 706.4  | 749.8      |
-| Process (pooled)     | 352.3           | 653.2  | 634.5  | 675.1      |
 
 ### Notes
 
 - QuickJS stayed the strongest path for trusted, in-process workloads at low and high concurrency in this run, while worker pooled edged ahead at concurrency 2.
-- Worker and process pooled executors paid visible queueing once demand moved past the benchmark pool size.
-- Process pooled stayed competitive, but it still trailed worker pooled at every tested concurrency level.
+- Worker pooled executors paid visible queueing once demand moved past the benchmark pool size.
 
 ---
 
@@ -109,21 +102,16 @@ The pooled benchmark factories in this suite use a fixed `pool.maxSize: 2`.
 | Worker   | 1         | 362.6               | 22.18ms        | 22.51ms     | 22.53ms     |
 | Worker   | 2         | 704.2               | 10.97ms        | 11.64ms     | 12.39ms     |
 | Worker   | 4         | **1327.2**          | 5.60ms         | 6.95ms      | 6.99ms      |
-| Process  | 1         | 351.0               | 22.54ms        | 23.66ms     | 23.97ms     |
-| Process  | 2         | 693.3               | 11.02ms        | 12.95ms     | 12.99ms     |
-| Process  | 4         | 1111.3              | 6.50ms         | 8.73ms      | 8.75ms      |
 
 ### Notes
 
-- Increasing pool size improved both worker and process executors in this run.
-- Worker pooled still kept the better mix of throughput and tail latency at every pool size tested.
-- Process pooled also scaled up well here, but its tails remained wider than worker pooled at the same pool size.
+- Increasing pool size improved worker throughput and tail latency in this run.
 
 ---
 
 ## 7. Host-Process Memory Delta
 
-This suite only measures the parent Node process. It does not attempt to attribute child-process RSS back to `QuickJsExecutor({ host: "process" })`.
+This suite only measures the parent Node process.
 
 | Executor             | Heap Delta | RSS Delta | External Delta |
 | -------------------- | ---------- | --------- | -------------- |
@@ -144,13 +132,11 @@ This suite only measures the parent Node process. It does not attempt to attribu
 ### High-value takeaways from this snapshot
 
 - QuickJS remained the lowest-latency option for trusted, in-process workloads on this machine.
-- True `prewarm()` delivered the intended first-request benefit for pooled worker and pooled process executors.
+- True `prewarm()` delivered the intended first-request benefit for pooled worker executors.
 - Worker pooled remained the strongest general-purpose local trade-off between isolation, throughput, and tail latency.
-- Process pooled stayed viable when process isolation matters, but it still trailed worker pooled on throughput and tail latency.
 - Ephemeral modes remained dramatically slower than pooled modes and are best reserved for cases that need a fresh host boundary per execution.
 
 ### What this snapshot does not prove
 
 - It does not prove exact throughput rankings for every workload or host. The concurrency and tool-call suites are still sensitive to local scheduler noise.
-- It does not prove memory behavior for `QuickJsExecutor({ host: "process" })`, because the memory suite intentionally avoids reporting child-process RSS as if it were host-process memory.
 - It does not measure `RemoteExecutor`, because remote performance depends on the caller-owned transport and remote runtime deployment.
diff --git a/docs/architecture/README.md b/docs/architecture/README.md
@@ -40,11 +40,11 @@ flowchart LR
 
 ### Package Roles
 
-| Package            | Role                                                                                                         |
-| ------------------ | ------------------------------------------------------------------------------------------------------------ |
-| `@execbox/core`    | App-facing core types, provider resolution, MCP adapters, plus runtime and protocol subpaths                 |
-| `@execbox/quickjs` | Default QuickJS executor package with inline, worker-hosted, and process-hosted modes plus a reusable runner |
-| `@execbox/remote`  | Transport-backed executor that runs against an app-defined runner boundary                                   |
+| Package            | Role                                                                                         |
+| ------------------ | -------------------------------------------------------------------------------------------- |
+| `@execbox/core`    | App-facing core types, provider resolution, MCP adapters, plus runtime and protocol subpaths |
+| `@execbox/quickjs` | Default QuickJS executor package with inline and worker-hosted modes plus a reusable runner  |
+| `@execbox/remote`  | Transport-backed executor that runs against an app-defined runner boundary                   |
 
 ## End-to-End Execution Model
 

diff --git a/docs/architecture/execbox-core.md b/docs/architecture/execbox-core.md
@@ -189,12 +189,12 @@ Executors are responsible for their own runtime-specific classification rules, b
 
 ## Why the Core Stays Small
 
-The core package does not own QuickJS, worker threads, child processes, or transport mechanics. That separation keeps the core useful for:
+The core package does not own QuickJS, worker threads, process boundaries, or transport mechanics. That separation keeps the core useful for:
 
 - direct in-process runtimes
 - worker-backed runtimes
 - MCP wrapper servers
-- process or remote execution models
+- remote execution models
 
 The consequence is deliberate separation between: