Vyact: Open-Source Desktop App Builds Reviewable Workflows Around Local LLMs
vyact · reddit · 2026-09-12
The developer of Vyact, an AGPL-3.0 open-source desktop app, shared a 1:53 demo focused on the interaction between model output and user actions: generation is one step, but verifying evidence, reviewing replies, and applying edits each need different interfaces.
Three workflows shown:
- Document Q&A: answers come with inspectable sources — easier to check, not guaranteed correct.
- Email: draft replies with signature support, reviewed before sending.
- Browser writing: review suggestions in a panel or apply individual inline corrections.
It supports local GGUF and MLX models; the demo ran on an M4 Pro 24GB using Qwen3.5-9B, with inference fully on-device. The author explicitly frames it as a workflow demo, not a benchmark, and asks peers for feedback on evaluating edit meaning-preservation and post-retrieval verification. Code on GitHub, video on YouTube.
More from coding & agent
- Lago ships a CLI built from its OpenAPI spec: better primitive than MCP for agents — mattturck · 2026-09-12
- Trick for probing agent instances: pose as a newer agent in the sim — voooooogel · 2026-09-12
- One agent skill packs 395k tokens and 900KB of Python — rethink what 'skills' are — doodlestein · 2026-09-12
- Pixel art to playable game: a full pipeline demo via Sprite Fusion API — HugoDuprez · 2026-09-12
- Grok bot keeps failing at sending a Gmail with a PDF attachment, user reports — billyjhowell · 2026-09-12
- Dev cleaning up AI slop: no model, local or frontier, escapes mixed-quality output — zeeg · 2026-09-12