CapybaraHow Capybara turns your idea into a playable AI-built game
Understand the Capybara workflow: upload style, prompt an idea, shape the plan, approve the first build, then let the chat agent coordinate assets and code.
Capybara works best when you treat it like a small game team: art direction first, design planning second, build execution third. Each step has a different job.
Step 1: Bring the look before the build
If you already know the visual direction, upload an art style reference first. This gives the system a concrete target for the asset generation phase. It is especially useful for games where mood matters: cozy farming sims, eerie mysteries, anime adventure RPGs, or polished commercial demos.
You can still describe the style in a few words, but do not make the prompt carry the entire look. Save your prompt budget for the game experience.
- Upload style for visual consistency.
- Prompt gameplay for design clarity.
- Keep the first build visually coherent before expanding scope.
Step 2: The planning agent turns the idea into a decision path
After your initial prompt, the planning agent may ask clickable questions. These are not busywork. They are there when a choice changes the core loop, asset list, map shape, or audience goal.
Once the important forks are clear, the planner creates a structured game plan: title, pitch, design sections, first build summary, gameplay notes, asset plan, pipeline notes, and a builder brief. The user-facing plan teaches why the design works; the builder brief explains how the first playable slice should behave.
- The plan keeps the game fantasy intact while narrowing scope.
- The first build focuses on a complete playable loop.
- The builder brief avoids raw asset-bucket language and describes player-facing mechanics.
Step 3: The build chat coordinates assets and code
When you approve a plan, the build chat treats it as the source of truth. The chat workflow streams the agent turn, chooses which tools are available, and lets the durable chat agent call specialized helpers. With asset credits, it can use the assets agent and code agent; for later guidance it can also use an asset advisor.
This separation is important. The planning agent decides what should be built. The chat agent coordinates the actual build work: assets, implementation, workspace cleanup, message persistence, and finalization of the turn.
- Assets agent: creates maps, characters, props, HUD, and music when available.
- Code agent: wires gameplay, UI, interactions, persistence, and rules.
- Asset advisor: helps with follow-up asset decisions after the first build path.
Step 4: Iterate like a designer, not a bug reporter only
After the first build, your best follow-up prompts describe what the player should feel or understand next. Instead of only saying “add more,” ask for a stronger objective tracker, a clearer clue reveal, a more satisfying success screen, a new NPC memory consequence, or a visible blocked goal that opens after the loop.
Capybara’s platform can support AI NPCs, TTS narration, stateful quests, save/load, multiplayer patterns, and persistent progression. The strongest projects add these when they improve the loop, not because the feature list looks impressive.
- For AI NPCs, anchor conversations to game state, evidence, inventory, reputation, and role limits.
- For TTS, use short narrated beats and skippable dramatic reveals.
- For persistence, decide what carries forward: reputation, discoveries, unlocked areas, inventory, or case history.
Before you build
- Upload an art style if visuals are important.
- Prompt the fantasy, goal, loop, and first objective.
- Use planning questions to make the important design forks explicit.
- Approve a plan with a clear first five minutes.
- Iterate by asking for better player feedback, progression, and return hooks.
Keep reading
More practical guides for building better first slices.