Package v0.2.12→v0.2.124: manifest, actions, version graph

startos/versions/v0.2.62.ts

@@ -0,0 +1,13 @@
+import { VersionInfo } from '@start9labs/start-sdk'
+
+export const v_0_2_62 = VersionInfo.of({
+  version: '0.2.62:0',
+  releaseNotes: {
+    en_US:
+      'Operator-tunable output token caps + section-count targets. Until now four of the most performance- and quality-impacting knobs were hardcoded: Gemini transcribe maxOutputTokens (65536), Gemini analyze maxOutputTokens (implicit Google default ~8192), hardware analyze max_tokens (16000), and the section-count target string used to populate {{targetSections}} in the analyze prompt (1/1-2/2-3 based on window length). All four are now first-class Settings tab fields with live-reload. (1) Gemini Backend panel: TWO new sliders "TX max output tokens" (1024-65536, default 65536) and "AN max output tokens" (1024-65536, default 8192). Lower TX = cheaper + faster but more truncation risk on dense audio. AN was previously implicit Google default; making it explicit lets operators with reasoning-heavy models cap unbounded thinking preambles. (2) Operator Hardware panel: new "AN max output tokens" slider (1024-32768, default 16000). Pass-through to the vLLM/Ollama chat-completion `max_tokens` parameter. (3) New "Section-count targets" panel below the LLM prompts: three free-form text inputs (short ≤10min / medium ≤20min / long >20min). Each is spliced into the analyze prompt as {{targetSections}} based on each window\'s actual transcript duration. Defaults match the historical hardcoded values ("1 section" / "1-2 sections" / "2-3 sections") so behavior is unchanged unless the operator overrides. Empty string = use default (future code-side default changes flow through). Net effect: an operator who wants denser segmentation can set medium to "3-4 sections" without rewriting the whole analyze prompt template; one who wants to constrain Gemini AN output token spend can cap maxOutputTokens at 4096 in the new slider. All knobs validated server-side (1024 floor on token caps to prevent unusable values).',
+  },
+  migrations: {
+    up: async ({ effects }) => {},
+    down: async ({ effects }) => {},
+  },
+})