Setup

Configuration

Dojo works out of the box with no config file. If you want to customize things, create a dojo.toml in your project root:

[skills]
dir = ["skills", ".agents/skills"]

[model]
provider = "anthropic"
evaluator = "claude-haiku-4-5"

[reporting]
per-skill = true
consolidated = false

See the Config Reference for the full schema.

Global CLI flags

These flags work on all commands and override anything in dojo.toml:

Flag	Description
-s, --skills-dir <dir>	Override skills directory (repeatable)
-c, --config <path>	Path to config file (default: auto-detect dojo.toml)
-d, --cwd <dir>	Working directory for config and skill discovery

Model-related flags (--model-provider, -m, -j) live on dojo run specifically. See the Commands Reference for all options.

Model providers

Dojo needs a model provider to run evals. Four are supported:

Provider	Config value	Auth	Default model
Anthropic	anthropic	ANTHROPIC_API_KEY env var	claude-haiku-4-5
OpenAI	openai	OPENAI_API_KEY env var	gpt-4o-mini
GitHub Copilot	copilot	Active Copilot subscription	SDK default
Vercel AI SDK	vercel	Key for the underlying provider	openai/gpt-4o-mini

Anthropic is the default. Switch via dojo.toml:

[model]
provider = "anthropic"
evaluator = "claude-sonnet-4-6"
judge = "claude-opus-4-6"

Or at runtime:

dojo run --model-provider anthropic -m claude-sonnet-4-6

For the Vercel provider, model strings use the format <provider>/<model-id> (e.g. openai/gpt-4o-mini or anthropic/claude-sonnet-4-6). This routes through the Vercel AI SDK to whichever underlying provider you specify.

See the Config Reference for all model options.