WikiWayne
Local AIAI ToolsDigital MarketingTech NewsGamesAboutBlogContact

As an Amazon Associate I earn from qualifying purchases. This site contains affiliate links.

WikiWayne

Independent guides on open-weight AI, local inference, and the hardware that runs it.

Categories

  • Local AI Hub
  • Local AI
  • AI Tools
  • Digital Marketing
  • Tech News

Quick Links

  • About Wayne
  • Contact
  • Methodology
  • Editorial Standards
  • Disclosures
  • Privacy Policy
  • Sitemap

Follow on X

Daily AI insights, tech takes, and more.

Follow @wikiwayne
WikiWayne© 2026
PrivacyMethodologyEditorialDisclosuresTermsSitemap
Home/Games/Token Breaker

Token Breaker

Lines of source code march down the terminal. You are the shredder at the bottom. Break every token before a line reaches you, fill the bucket, and the round resets faster and meaner.

Open fullscreen ↗Click the game first so it picks up your keyboard.

Controls

← → / A D
Move the shredder
SPACE
Boost fire rate (builds heat) — also starts the game
P
Pause / resume
R
Restart after a game over
ESC
Back to the title screen

On phones and tablets, on-screen left / right / boost pads appear automatically.

How to play

  • Every character in a falling line is a token with its own hit points. Clear the whole line for a bonus.
  • Boost doubles your damage and fire rate but builds heat. Max heat forces a cooldown where you cannot fire at all.
  • Drag the tok slider at the bottom-left to raise throughput. Higher tok/s means a bigger score multiplier and faster, denser lines.
  • Minified lines (round 4+) are armored and quick. Comment lines (round 6+) are the toughest but pay double.
  • Fill the bucket to clear the round. Let a line touch the shredder and it is over.

Runs entirely in your browser. High scores are saved to local storage on your own device — nothing is uploaded.

How it was built: one box, one agent, no cloud

I built Token Breaker with Hermes Agent. Not “AI-assisted.” The agent did the heavy lifting on architecture, implementation, iteration, and tooling while I directed, reviewed, and steered. Everything ran locally — the codebase never left the machine.

For everyone who has been asking about the stack

DeepSeek-V4-Flash on a single ASUS Ascent GX10 (NVIDIA GB10, 128 GB unified memory). Not multi-Spark tensor parallel. Not vLLM.

Hardware
ASUS Ascent GX10 — NVIDIA GB10, 128 GB coherent unified memory
Model
DeepSeek-V4-Flash — 284B MoE, ~13B active
Serving stack
Entrpi ds4-on-spark, Blackwell-optimized CUDA path
Weights
antirez 0731 IQ2 imatrix GGUF (~81 GB)
Drafter
Matching DSpark drafter (~6.5 GB), lossless speculative decode
Context
1M native, text-only
API
OpenAI-compatible on :8888, full agent tool support

Best one-box local coding setup I have found for this hardware. Vision lives elsewhere.

Why this exact configuration

The GX10’s 128 GB of coherent unified memory is both the constraint and the opportunity. Higher-precision quants of DeepSeek-V4-Flash simply do not fit with usable KV cache and headroom. The asymmetric IQ2 imatrix quant from antirez keeps the dense and shared components — attention, router, shared expert, output head — at higher precision while aggressively quantizing the routed experts. That trade-off has proven surprisingly effective for tool-calling and code generation.

DSpark speculative decoding is the real multiplier. It is lossless relative to the target model and delivers a meaningful decode speedup without the quality cliff you sometimes get from aggressive drafting. Combined with Entrpi’s Blackwell-optimized CUDA path and its significant prefill and decode lifts over upstream ds4, the box stays responsive through long contexts and tool loops.

Hermes Agent sits on top of the OpenAI-compatible endpoint, so it gets the full tool surface, persistent memory, skill evolution, and the ability to work across large codebases without constantly fighting the context window. A 1M native context means the agent can usually hold current game state, recent changes, design notes, and the relevant prior code all at once.

Text-only is intentional. Vision models and multimodal pipelines live on other machines. Keeping the primary coding agent pure text reduces complexity and keeps the memory footprint predictable on a single GX10.

What this enables for game development

  • Scaffold systems and iterate on mechanics
  • Write and refactor game logic
  • Generate and tune content pipelines
  • Debug across files while retaining long conversation history
  • Maintain its own skill library from the work it does on the project

I still review everything, set direction, and make the final calls. This is not autonomous founder mode. It is a high-capability local coding partner that never sends the codebase off-machine. Token Breaker exists because that local loop is fast enough, private enough, and capable enough that building this way felt natural instead of forced.

Trade-offs and reality check

  • IQ2 is aggressive. Quality is good for coding and agentic work on this model, but it is not identical to higher-bit or full precision.
  • Single-box means no easy scaling to larger active parameter counts without another machine.
  • Speculative decode helps, but long tool-heavy trajectories still take time.
  • Vision is deliberately external.

For pure local coding agents on current single-GX10 hardware, this remains the strongest combination I have run: model capability, context length, tool fidelity, speculative speed, and full offline control.

Play Token Breaker. Break some tokens. Tell me what breaks — or what doesn’t.

The game and the stack that built it both live here. Tell me what you find, or read more on running AI locally.