Qwen3.8-Flash-Next on any consumer hardware: one-click install for Windows / Linux. Strata inference engine, OpenAI/Anthropic API on localhost, optional image input.
C++
Tests follow the new reply room; the queued-message test waits for the goal to be running
Niko1221f37abb13654a
diff --git a/apps/host/tests/agent-plan-run.test.mjs b/apps/host/tests/agent-plan-run.test.mjsindex 147eb117b..29f176705 100644--- a/apps/host/tests/agent-plan-run.test.mjs+++ b/apps/host/tests/agent-plan-run.test.mjs@@ -92,8 +92,11 @@ test('a message typed while a goal without points runs is queued, can be taken bassert.throws(() => r.bridge.queueMessage({ chatId: 'q1', text: 'hello' }), /No goal is running/);const out = [];const run = r.bridge.runTurn({ ...r.base('q1', f => out.push(f)), goal: { action: 'create', objective: 'keep going', maxRounds: 50 } });- await wait(3000);- const a = r.bridge.queueMessage({ chatId: 'q1', text: 'first thing' });+ // the goal turn may take a while to start on a busy machine: queue as soon as it runs+ let a = null;+ for (let i = 0; i < 150 && !a; i++) { try { a = r.bridge.queueMessage({ chatId: 'q1', text: 'first thing' }); } catch { await wait(100); } }+ assert.ok(a, 'the goal was running');+ await wait(500);const b = r.bridge.queueMessage({ chatId: 'q1', text: 'second thing' });assert.equal(a.message.deliverAt, 'end'); assert.deepEqual(b.queue.map(m => m.text), ['first thing', 'second thing']);const back = r.bridge.cancelQueued({ chatId: 'q1', id: b.message.id });diff --git a/apps/host/tests/agent-tools.test.mjs b/apps/host/tests/agent-tools.test.mjsindex a35986fbe..baffeb311 100644--- a/apps/host/tests/agent-tools.test.mjs+++ b/apps/host/tests/agent-tools.test.mjs@@ -237,7 +237,7 @@ test('the output budget is not the old 8192, and the Reasoning menu reaches theconst off = await ask('think-off', 'Off');assert.equal(ctk(off).enable_thinking, false); assert.equal(ctk(off).reasoning_effort, undefined);assert.equal(off.enable_thinking, undefined, 'a top-level enable_thinking is the field the Strata server ignores');- assert.ok(off.max_completion_tokens > 40000, `budget ${off.max_completion_tokens} (the window minus the prompt, never the old 8192)`);+ assert.ok(off.max_completion_tokens >= 32000 && off.max_completion_tokens <= 32768, `budget ${off.max_completion_tokens} (half the window, at most 32K: never the old 8192, and room is left for the conversation to be compacted)`);const high = await ask('think-high', 'High');assert.equal(ctk(high).enable_thinking, true); assert.equal(ctk(high).reasoning_effort, 'high');const low = await ask('think-low', 'Low');