Configuration

Everything works out of the box with defaults. If you want to customize, create a dynamo.yaml in your project root.

dynamo.yaml

project:
  name: "MyGame"
  github_repo: "myorg/mygame"
  base_ref: "main"

ai:
  provider: "anthropic"
  personality: "full"        # full | light | minimal
  sandbox: "full"            # project | full | unrestricted
  memory: true               # false to disable memory
  auto_checkpoint: true      # false to disable auto-checkpoints
  auto_consolidate: false    # true to auto-consolidate memory on session start
  slim_mode: false           # compact mode for local providers (Ollama etc.)
  show_costs: false          # show per-turn cost in the status bar

  models:
    interactive: "anthropic/<model-id>"          # model for interactive chat

  budget:
    session_limit: 5.0       # warn (non-blocking) at 80% / 100% of this spend

  shell:
    defaultTimeout: 120000   # 2 min (ms)
    maxTimeout: 600000       # 10 min max
    maxOutput: 50000         # 50K chars before truncation

  providers:
    deepseek:
      type: "openai-compatible"
      base_url: "https://api.deepseek.com/v1"
      api_key_env: "DEEPSEEK_API_KEY"

engines:
  unity:
    project_path: "./MyUnityProject"

Once a dynamo.yaml exists, project.name is the one required key — everything else has sensible defaults.

Run dynamo config to see your full resolved configuration.

Config namespaces

These top-level namespaces are available (see dynamo config for the full resolved tree):

Getting more out of newer OpenAI models

Newer OpenAI models accept a few things older ones don't. All three are off by default — turn one on only if you want it. original image detail and the prompt cache are each checked against your model first: on a model that can't take them, Dynamo sends high instead and leaves the cache as it was. The prompt cache and persisted reasoning only work through OpenAI's newer API (the one current models use), and persisted reasoning only on a model that reasons at all.

These settings reach models on the built-in OpenAI provider (openai/…) only. A custom OpenAI-compatible provider — DeepSeek, Groq and the like — ignores them, even written under its own ai.<name> block.

ai:
  openai:
    image_detail: "original"        # send images untouched instead of resized
    explicit_prompt_cache: true     # mark the reusable part of the prompt yourself
    persisted_reasoning: true       # let the model keep its own train of thought

Turning these on doesn't change Dynamo's privacy posture: every OpenAI request still asks OpenAI not to keep it for retrieval (store: false).

Personality Levels

Override with DYNAMO_PERSONALITY env var.

Environment Variables

API Keys

For Ollama, no API key is needed — Dynamo auto-detects a running Ollama instance.

For custom providers (DeepSeek, Groq, etc.), set the key name in api_key_env in your provider config.

Overrides

Auth Priority

Dynamo resolves an API key in this order:

1. 1.Environment variable for the provider — ANTHROPIC_API_KEY, OPENAI_API_KEY, OPENROUTER_API_KEY, GOOGLE_API_KEY, XAI_API_KEY.
2. 2.Stored credentials — credentials.json in your config dir, written by dynamo setup (mode 0600).

That's the whole chain: keys are never read from preferences.json or dynamo.yaml. OpenClaw doesn't sit in this order at all — if you have an OpenClaw install, its .env is loaded at startup and fills in variables that aren't already set, so it feeds step 1 rather than overriding it. Ollama needs no key.

If no key is found, the provider is simply unavailable — dynamo setup is how you add one.

OpenClaw Integration

If you use OpenClaw, Dynamo detects it automatically:

• ●API key import — your OpenClaw API keys are imported automatically (explicit environment variables always take priority)
• ●Model aliases — an alias you've set in OpenClaw (e.g. /model my-fast) resolves here too, if OpenClaw knows the model
• ●Context window sizes — imported so Dynamo knows each model's limits
• ●Config watching — Dynamo watches for changes to your OpenClaw config and picks them up live
• ●Manual sync — run /sync to re-sync OpenClaw settings on demand

No OpenClaw? No problem. Dynamo works fine standalone.

File Locations

Paths under ~/.config/dynamo/ are %APPDATA%\dynamo\ on Windows.