mirror of
https://github.com/block/buzz.git
synced 2026-08-18 06:50:31 +02:00
Shared-compute serve nodes currently run whatever reasoning behaviour the model ships with. For reasoning models (e.g. Qwen3.6-35B-A3B) that means every agent turn burns hidden thinking tokens — a 1-word reply cost 161 completion tokens in testing — which is slow and expensive under the goose / Buzz agent harness. Buzz lets mesh-llm synthesise an isolated throwaway serve config that only disables the telemetry + blobstore plugins. This points the embedded serve node at a Buzz-owned config file instead, which re-asserts those no-leak disables and adds a reasoning-off request default. Clients can still opt back into thinking per request. No mesh-llm source change and no touching the user's global mesh config. DRAFT: mechanism (config_path plumbing) verified from source and compiles; the reasoning-off *default* content is NOT yet verified end-to-end on the embedded v0.73.1 runtime. See PR description.
Buzz
Desktop chat shell with:
- Tauri + React + TypeScript + Vite
- Tailwind CSS
- shadcn/ui-ready shared components
- Biome (lint/format/check)
- Feature-driven frontend structure
Scripts
pnpm dev- run the web frontendpnpm tauri dev- run the desktop apppnpm build- typecheck and build frontendpnpm typecheck- TypeScript checkspnpm lint- Biome lintpnpm format- Biome format (write)pnpm check- Biome check
Structure
src/shared- reusable app-wide code (ui,lib,styles)src/features- feature modules (vertical slices)src/app- top-level app composition