Keysat
0ae59f3550
Introduces RECAP_MODE=multi alongside single-mode self-host: - Tenant auth + accounts (magic-link via System SMTP), per-tenant credit pool, anonymous trial minting with per-IP/-64 caps - Self-serve Pro/Max purchase: inline Lightning (BTCPay) + card (Zaprite), prepaid 30-day periods, expiry-reminder emails - Core-decoupling: relay owns cloud tier/expiry keyed by Recaps user-id - SQLite (better-sqlite3) schema for multi-mode; filesystem unchanged for single - StartOS actions/versions through 0.2.155
13 lines
1.3 KiB
TypeScript
13 lines
1.3 KiB
TypeScript
import { VersionInfo } from '@start9labs/start-sdk'
|
|
|
|
export const v_0_2_59 = VersionInfo.of({
|
|
version: '0.2.59:0',
|
|
releaseNotes: {
|
|
en_US: 'Topic analysis is now chunked-and-parallel on long content. Previously, a 2-hour transcript meant one giant analyze call against ~25K tokens of prompt — long prefill, single point of failure, and minutes of wall-time even on fast backends. Now, content longer than 25 minutes is split into overlapping 18-minute windows (with a 2-minute overlap on each side to handle topic boundaries that span windows) and the windows are analyzed in parallel (up to 4 concurrent calls). Results are stitched back into a single ordered list of sections with no overlap, using a "trust the second window for the overlap region" rule that prevents boundary fragmentation. Short content (≤25 min) still single-shots as before — no overhead for the common case. End-to-end speedup is typically 3-5x on 1+ hour content, and even larger when paired with the relay\'s recent JSON-mode/thinking-off change (relay 0.2.16). Works across all backends (relay, direct Gemini, OpenAI, Anthropic, Ollama) since chunking happens at the orchestrator level.',
|
|
},
|
|
migrations: {
|
|
up: async ({ effects }) => {},
|
|
down: async ({ effects }) => {},
|
|
},
|
|
})
|