Plan 323 P2 — Qwen3.5 9B on this computer

docs/plans/323-models-that-run-on-this-computer.md

@@ -330,6 +330,8 @@ Each lineup entry (§4) is qualified on its own: both probe tiers, the build ide
 
 _Exit for each entry:_ its usefulness results, repeated runs included, are in the entry, and its recommended roles follow from them.
 
+_Entries so far:_ `ollama-qwen3.5-9b` (`qwen-local`) is in the catalog from its compatibility tier alone — native single, parallel and streamed tool calls, vision, `off`/`high` reasoning, no `format` on MLX, 7.7 GB of weights and about 69 KB of cache per token at `q8_0`, so about 10 GB at the agents' 32K window. Like every local entry it is assignable under D18 and recommended for no role; the usefulness tier, which is not built, decides that. Both entries' probes run through `localRuntimeFetch`: Gemma's took 8½ minutes, past the transport's own five.
+
 ### P3a — The window fits the machine (D8)
 
 **Build:**

packages/ai-client/src/catalogModels.ts

@@ -98,6 +98,41 @@ export const builtInModels: ModelDefinition[] = [
       downloadBytes: 16_146_867_483,
     },
   },
+  {
+    id: 'ollama-qwen3.5-9b',
+    provider: 'ollama',
+    providerModelId: 'qwen3.5:9b',
+    displayName: 'Qwen3.5 9B (this computer)',
+    description:
+      "Alibaba's Qwen3.5 9B, dense, served by Ollama on this computer. Vision, native tool calls, on/off thinking; fits a 16 GB machine at the agents' window. Apache 2.0.",
+    contextWindow: 262_144,
+    maxOutputTokens: 8192,
+    capabilities: {
+      ...defaultCapabilities,
+      vision: true,
+      reasoning: true,
+      // The MLX runner refuses `format`; JSON rides a tool call instead.
+      structuredOutputs: false,
+      // The runtime has no tool_choice.
+      forcedToolChoice: false,
+    },
+    pricing: { promptPer1M: 0, completionPer1M: 0, currency: 'USD' },
+    traits: { speed: 2, cost: 1, intelligence: 2, outputType: 'text' },
+    aliases: ['qwen-local'],
+    reasoning: { supported: ['off', 'high'] },
+    local: {
+      runtimeTag: 'qwen3.5:9b',
+      digest: 'd63f830f28d515fefea546f213fca7a75e9dd5176ef51a6deabc952d1c99a911',
+      runner: 'mlx',
+      quantization: 'nvfp4',
+      runtimeVersion: '0.40.5',
+      probeVersion: 2,
+      measuredAt: '2026-10-17',
+      // Measured at OLLAMA_KV_CACHE_TYPE=q8_0 (§9 G6).
+      footprint: { weightsBytes: 7_735_695_816, kvBytesPerToken: 68_742 },
+      downloadBytes: 7_639_229_431,
+    },
+  },
 
   // ============================================================================
   // OpenAI Text Models

packages/ai-client/src/index.ts

@@ -205,3 +205,4 @@ export {
   type LocalRuntimeClient,
   type LoadedLocalModel,
 } from './providers/ollamaRuntime.js';
+export { localRuntimeFetch } from './providers/localRuntimeFetch.js';

packages/ai-client/src/providers/localRuntimeFetch.ts

@@ -14,6 +14,6 @@ import { Agent } from 'undici';
 
 const NO_TRANSPORT_LIMIT = new Agent({ headersTimeout: 0, bodyTimeout: 0 });
 
-export function localRuntimeFetch(url: string, init: RequestInit): Promise<Response> {
+export function localRuntimeFetch(url: string, init: RequestInit = {}): Promise<Response> {
   return fetch(url, { ...init, dispatcher: NO_TRANSPORT_LIMIT } as RequestInit);
 }

scripts/local-model-gates.ts

@@ -19,6 +19,7 @@
  * `lifecycle` and `embeddings` pull small models (about 1–3 GB) and delete
  * what they pulled. `machine` loads every model it is given, one at a time.
  */
+import { localRuntimeFetch } from '@aflow/ai-client';
 import { writeFileSync } from 'node:fs';
 import { execFileSync } from 'node:child_process';
 import os from 'node:os';
@@ -51,7 +52,7 @@ const describe = (e: unknown): string =>
 const seconds = (ns: unknown): number => (typeof ns === 'number' ? Math.round(ns / 1e7) / 100 : 0);
 
 async function api(path: string, body?: unknown, signal?: AbortSignal): Promise<Response> {
-  return fetch(`${BASE}${path}`, {
+  return localRuntimeFetch(`${BASE}${path}`, {
     method: body === undefined ? 'GET' : 'POST',
     headers: { 'content-type': 'application/json' },
     ...(body === undefined ? {} : { body: JSON.stringify(body) }),
@@ -161,7 +162,7 @@ async function pull(
 }
 
 async function remove(model: string): Promise<void> {
-  const res = await fetch(`${BASE}/api/delete`, {
+  const res = await localRuntimeFetch(`${BASE}/api/delete`, {
     method: 'DELETE',
     headers: { 'content-type': 'application/json' },
     body: JSON.stringify({ model }),

scripts/verify-local-models.ts

@@ -17,6 +17,7 @@
  */
 import {
   createDefaultModelCatalog,
+  localRuntimeFetch,
   type LocalModelBuild,
   type LocalModelFootprint,
   type ModelDefinition,
@@ -69,7 +70,7 @@ async function post(
   path: string,
   body: unknown,
 ): Promise<{ status: number; json: Record<string, unknown> }> {
-  const res = await fetch(`${BASE}${path}`, {
+  const res = await localRuntimeFetch(`${BASE}${path}`, {
     method: 'POST',
     headers: { 'content-type': 'application/json' },
     body: JSON.stringify(body),
@@ -103,7 +104,7 @@ function toolCalls(
 }
 
 async function streamedToolCall(tag: string): Promise<boolean> {
-  const res = await fetch(`${BASE}/api/chat`, {
+  const res = await localRuntimeFetch(`${BASE}/api/chat`, {
     method: 'POST',
     headers: { 'content-type': 'application/json' },
     body: JSON.stringify({
@@ -123,7 +124,7 @@ async function streamedToolCall(tag: string): Promise<boolean> {
 }
 
 async function loadedSize(tag: string): Promise<number | null> {
-  const ps = (await (await fetch(`${BASE}/api/ps`)).json()) as {
+  const ps = (await (await localRuntimeFetch(`${BASE}/api/ps`)).json()) as {
     models: Array<{ name: string; size: number }>;
   };
   return ps.models.find((m) => m.name === tag || m.name === `${tag}:latest`)?.size ?? null;
@@ -165,14 +166,16 @@ async function measureFootprint(tag: string): Promise<LocalModelFootprint | null
 }
 
 async function measure(tag: string): Promise<Measured | null> {
-  const tags = (await (await fetch(`${BASE}/api/tags`)).json()) as {
+  const tags = (await (await localRuntimeFetch(`${BASE}/api/tags`)).json()) as {
     models: Array<{ name: string; digest: string; remote_host?: string }>;
   };
   const installed = tags.models.find((m) => m.name === tag || m.name === `${tag}:latest`);
   if (!installed) return null;
   if (installed.remote_host)
     throw new Error(`${tag} is served by ${installed.remote_host}, not this computer`);
-  const version = (await (await fetch(`${BASE}/api/version`)).json()) as { version: string };
+  const version = (await (await localRuntimeFetch(`${BASE}/api/version`)).json()) as {
+    version: string;
+  };
   const show = (await post('/api/show', { model: tag })).json;
   const details = (show['details'] ?? {}) as Record<string, string>;
   const caps = new Set((show['capabilities'] as string[] | undefined) ?? []);