Releases: flowork-os/Flowork-OS
Release list
Flowork 1.0.1
Flowork OS 1.0.1 — production audit release
A full GUI→logic→agent audit of all 46 features (27 router + 19 agent). Every feature verified wired end-to-end (no mocks, no dead wires); third-party providers (tunnels, models, TTS/STT/image/embedding, OAuth) confirmed plug-and-play — remove the vendor file and the core still builds.
Highlights
- Console Log (router): self-scrolling terminal panel with a hacker theme + live request stream (provider, tokens, latency, cost).
- Code Progress (agent): self-scrolling terminal view with a streaming line-reveal effect.
- Settings: removed dead code; every control is a real, wired save/load.
- Brain seed shipped in English (personal data removed) so first-boot doctrine/instincts are ready out of the box.
- Stability lock: brain-seed core + the stable GUI surface are frozen (SHA256 + immutable) to prevent accidental drift.
Fully local, white-label, multi-OS. Bring your own keys via the GUI (Settings → Credentials) — nothing is baked in.
🤖 Generated with Claude Code
Flowork 1.0
Flowork OS 1.0 — an AI-colony operating system designed to last: a frozen core that grows only through plug-in seams, white-label, multi-OS.
What's in this release
Three ways to run it, all public & secret-free (fill in your own keys on first boot):
| Artifact | What it is | How to run |
|---|---|---|
flowork-portable-1.0.0.zip |
Run-from-folder, no install (Windows / Linux / macOS Intel+Apple) | Unzip → Start-Flowork |
flowork-os-1.0.0.iso |
Bootable ISO (VM or CD/USB) | Boot it, or dd to a stick |
flowork-os-1.0.0-public.usb.img.gz |
Bootable USB appliance image | gunzip → dd if=… of=/dev/sdX bs=4M |
Slim by design: no bundled LLM weights (the local model is pulled on demand; cloud router works out of the box) and the knowledge brain grows as you use it.
Highlights
- Multimodal — paste/send images (Ctrl+V / 📎) to chat and LLM vision; read documents; fixed Telegram photo handling (real image blocks, far fewer tokens).
- AI Studio × Mr.Flow — chat directly with the owner agent (Mr.Flow) inside AI Studio: target dropdown, send docs + images, a live "thinking…" indicator.
- Mesh Lockbox — per-peer knowledge provenance + revoke, verify-on-promote (integrity + signature), a mesh↔evolution wall (up by default), public-vs-private knowledge visibility (private by default), and a consent gate for peer tool manifests.
- Cross-turn prompt cache — conversation history is cached (~90% cheaper reads).
- Wake-push supervisor — the "mandor" agent wakes by event when a worker finishes.
- Sound release tooling — the appliance now ships its sidecar tools + docs/apps/skills, a single in-repo sidecar manifest, and a hardened public sanitizer.
Architecture: nano-modular, no hard-coded paths, GUI is the source of truth. Third-party dependencies (models/tunnels) are pluggable and removable.
🤖 Generated with Claude Code