Skip to content

Configuration

Dumpcat supports project-level configuration via .dumpcat.toml files. This lets you define defaults and named profiles so you don't have to repeat flags.

Config file discovery

Dumpcat looks for configuration in this order:

  1. Explicit path — --config path/to/config.toml
  2. Walk up — .dumpcat.toml in the target directory, then each parent directory
  3. Global — ~/.config/dumpcat/config.toml

The first file found is used. If no config is found, built-in defaults apply.

File format

[default]
format = "markdown"
gitignore = true
max_size = "1mb"
hidden = false
exclude = ["__pycache__", ".git", "*.pyc", "*.lock"]

[profiles.python]
include = [".py", ".pyi", ".toml", ".cfg", ".txt", ".md"]
exclude = ["__pycache__", ".git", "*.pyc", ".venv", ".mypy_cache"]

[profiles.web]
include = [".js", ".ts", ".jsx", ".tsx", ".css", ".html", ".json"]
exclude = ["node_modules", "dist", "build", ".next"]

Sections

[default]

Base settings applied to every run. Supports the same keys as profile sections.

[profiles.<name>]

Named profiles that layer on top of [default]. Activate with --profile:

dumpcat --profile python

Supported keys

Key Type Description
format string "markdown", "plain", or "json"
include list of strings File extensions to include
exclude list of strings Glob patterns to exclude
depth integer Max directory depth
max_size string Max file size (e.g. "500kb")
hidden boolean Include dotfiles/dotdirs

Precedence

Settings are merged in this order (later wins):

built-in defaults < [default] section < [profiles.X] section < CLI flags

CLI flags always take priority. This means you can set sensible defaults in your config and override them on a per-command basis.

Example workflow

Create a .dumpcat.toml in your project root:

[default]
exclude = ["__pycache__", "*.pyc", ".venv", "node_modules"]

[profiles.backend]
include = [".py", ".sql", ".toml"]
exclude = ["__pycache__", "*.pyc", ".venv", "migrations"]

[profiles.frontend]
include = [".ts", ".tsx", ".css", ".html"]
exclude = ["node_modules", "dist", ".next"]

Then use it:

# Uses [default] settings
dumpcat

# Uses [profiles.backend] layered on [default]
dumpcat --profile backend

# Override depth even with a profile
dumpcat --profile backend -d 3

Commit the .dumpcat.toml to your repo so your team shares the same profiles.

LLM configuration

LLM targets are configured in a separate file: ~/.dumpcat/dumpcat_profiles.toml. Run dumpcat init to create a starter config.

Config file format

[llm]
default_target = "ollama"

[llm.targets.ollama]
provider = "ollama"
model = "llama3"
system_prompt = "You are a helpful coding assistant."
temperature = 0.7
max_tokens = 4096

[llm.targets.openai]
url = "https://api.openai.com/v1/chat/completions"
model = "gpt-4o"
api_key_env = "OPENAI_API_KEY"
temperature = 0.7

Target keys

Key Type Description
provider string Built-in provider: ollama, vllm, or lmstudio
url string Full endpoint URL (overrides provider)
model string Model name
system_prompt string Default system prompt
api_key_env string Environment variable containing the API key
temperature float Sampling temperature
top_p float Top-p sampling
max_tokens integer Maximum response tokens
n integer Number of completions
seed integer Random seed
top_k integer Top-k sampling
frequency_penalty float Frequency penalty
presence_penalty float Presence penalty

Usage

# Use the default target
dumpcat --llm -p "Review this code"

# Use a named target
dumpcat --llm -t openai -p "Review this code"

# Override model from a target
dumpcat --llm -t ollama -m mistral

# CLI flags override target config
dumpcat --llm -t openai --set temperature=0 --system-prompt "Be brief"

Precedence

LLM settings are resolved in this order (later wins):

target config < CLI flags (--model, --system-prompt, --set, --api-key)