Sub8
Computer-use agent on its own Linux VM.
Isolation so the model can act without owning the host. Model-agnostic sandbox for coding agents.
Daniel Farina
I build the layer around the model — sandbox, tools, evals, control plane — and publish what actually breaks.
- Grokathon 3rd + Most Users SpaceXAI · ThinkVoice · Aug 2026
- DC3 Researcher of the Month DoD VDP · March 2026
- 144 public repos GitHub · daniel-farina
- US work authorized No sponsorship needed
Selected work
All work → 
Xplor
Chromium fork with an agent gateway.
CDP + API so agents browse and build in-tab — not a chat overlay bolted onto Chrome.
GrokTerm
PTY host for coding agents with voice.
Multi-tab terminal, two-way Grok Voice, notarized Mac app — drive agents without living on a keyboard.
grok-remote
ACP control plane over a tailnet.
Remote agent sessions with a real UI — supervisor/worker shaped, not a single chat pane.
grok-bench
Prompt diffs, compaction, proxy observability.
Measure where tool-call loops and long-context workflows actually break.
ThinkVoice
Speak without moving your mouth — Grokathon.
Grok Voice + sensing; blink/nod select. SpaceXAI Grokathon 2026: 3rd place and Most Users.
Writing
All writing →Sub8 — isolated computer-use for coding agents
Sandbox layer so the agent that talks to you isn't the process holding the write lock.
Xplor — Grok's own browser
Chromium fork: Grok default search, "Grok it" on every page, Grok Build in a tab.
3,500 lines of 3D game logic from a single Grok 3 session
Long-horizon tool use on XForce — where Grok 3 held and where it didn't.
First 3D single and multiplayer game with Grok 3
Raycasted worlds, WebSocket sync — Claude and o3-mini choked past ~1,000 lines.
Unity space combat — Grok 4.5, zero hand-written C#
Full Unity game via CLI agents and MCP into the editor — no prior Unity experience.
Public proof
Demos and reach live on @Daniel_Farinax. Architecture, evals, and failure modes live here. Browse the build log →