Frontier is a Rust + Bevy prototype pivoting into an AI-lab strategy game about running a frontier AI company, outcompeting rival labs, and managing the tradeoffs between model capability, users, cash, trust, safety, compute, pricing, and talent.
Historical design reference: https://outline.gregagi.com/doc/garden-of-minds-game-design-brief-1dRK1Wt5yn
Current pivot backlog: see AGENTS.md.
The pivot slice answers one question:
Is it fun to make CEO-style AI-lab decisions and watch capability, users, cash, trust, safety, and competitors react?
- Deterministic AI-lab simulation seed
- One player lab and three competitor labs
- Core resources: capability, cash, compute, users, trust, safety, price, and talent
- CEO decisions with costs, benefits, and tradeoffs
- Competitor reactions each quarter
- Market model for price, adoption, revenue, and rank
- Decision forecasts before commit
- Event log with concise cause/effect feedback
- Short-run win and loss conditions
- Strategic Bevy dashboard
1: preview Train Model2: preview Cut API Prices3: preview Buy Compute4: preview Ship Product Features5: preview Invest in Safety/Reliability6: preview Hire ResearchersEnter: commit the selected CEO decisionSpace: advance one quarter after committingR: reset the run
See docs/install.md for full macOS and Windows setup instructions.
cargo run
cargo test
cargo fmtThe prototype includes a headless playtest mode for build-play-observe loops:
cargo run -- --script "train-model,advance-quarter:2,cut-prices,invest-safety,advance-quarter:3" --prettyThis runs the same simulation actions as the keyboard controls, prints a JSON state report, and exits without opening a Bevy window. Use it after gameplay changes to compare the planned consequence against quarter, market rank, competitors, status, events, deltas, and lab stats.
Useful options:
--seed 42: run a deterministic alternate seed.--window-size 1000x650: launch the visual game at a stable size.--dump-state: print the initial state as JSON and exit.--strategy aggressive-growth: run a built-in CEO strategy script.--pretty: pretty-print JSON output.
--window-size only affects visual windowed runs; headless --script and
--dump-state runs do not open a window.
Script actions:
advance-quarter,step, orspace: advance one quarter.1ortrain-model: Train Model.2orcut-prices: Cut API Prices.3orbuy-compute: Buy Compute.4orship-features: Ship Product Features.5orinvest-safety: Invest in Safety/Reliability.6orhire-researchers: Hire Researchers.rorreset: reset the run to the configured seed.
Built-in strategies:
aggressive-growthquality-first-researchtrust-safety-first
Repeat an action with action:count, for example advance-quarter:4. Repeats
are capped at 10000 to catch accidental runaway scripts.
Prototype acceptance criteria are mapped in docs/prototype-acceptance.md.
The original prototype explored a god-game about artificial minds, nodes, values, and interventions. That intent is preserved as historical context, but the current playable slice is now scoped around the clearer AI-lab strategy fantasy. See docs/pivot-model.md for the keep/remap/cut decisions behind the pivot.
This prototype can be run from source on macOS and Windows with Rust installed,
or downloaded from the Desktop Artifacts workflow.
- macOS: install Xcode Command Line Tools, install Rust with
rustup, then runcargo run --release. - Windows: install Visual Studio C++ Build Tools with
Desktop development with C++, install Rust withrustup-init.exe, then runcargo run --release.
The Desktop Artifacts GitHub Actions workflow can build downloadable
prototype packages for macOS and Windows. The latest manually built artifacts
are linked from the Desktop Artifacts workflow page.
See docs/install.md.
Do not add these before the first prototype loop works:
- real AI API calls
- combat
- multiplayer
- save/load
- complex procedural worlds
- production art pipeline
- fake desktop UI