Configuration

RepoDNA works without any configuration. When you want to tune it, settings come from several layers, applied in this order (later layers win):

Built-in defaults.
Your user configuration, config.toml in the RepoDNA configuration directory (repodna config path prints where it is).
The analyzed repository's repodna.toml, at its root.
A file passed with --config FILE.
Command-line options, such as --profile deep or --max-commits 5000.

--no-user-config and --no-project-config skip layers 2 and 3. Suppressions and ignore patterns from every layer are combined; every other setting is replaced by later layers.

What a repository can and cannot set#

A repository's repodna.toml is written by whoever controls the repository, which may not be you. It can therefore only tune what is analyzed: profiles, thresholds, ignore patterns, classification, suppressions, and report defaults. The following are honored only from your user configuration, --config, or the command line, and are ignored (with a warning) when they appear in a repository's repodna.toml:

[ai]: AI providers
[plugins]: which plugins run
[execution]: running detected build and test commands
privacy.remote_ai and privacy.telemetry

Getting started#

repodna init                  # writes a commented repodna.toml into the current repository
repodna config init           # creates your user configuration from a commented template
repodna config show           # the effective configuration and where each value came from
repodna config validate       # checks every layer for mistakes

Unknown keys are errors, so a typo is reported instead of silently ignored. The JSON Schema of the file is published at schemas/repodna-config.schema.json (or repodna schema config); editors that understand JSON Schema for TOML can use it for completion and validation.

Example `repodna.toml`#

[analysis]
profile = "deep"
max_commits = 20000

[ignore]
patterns = ["fixtures/large/", "*.snap"]

[classification]
generated = ["src/generated/**"]
vendor = ["third_party/**"]

[thresholds]
large_file_lines = 1500
high_complexity = 20

[[suppress]]
rule = "security.secret"
path = "tests/fixtures/**"
reason = "Intentional fake credentials used by tests"

[report]
theme = "technical"

Example user configuration#

[analysis]
profile = "standard"

[privacy]
anonymize_contributors = true

[performance]
parallelism = 8

[plugins]
enabled = ["license-headers"]

[ai]
provider = "command"
command = ["ollama", "run", "MODEL_ID"]  # a model you have installed

Reference#

`[analysis]`: what to analyze#

Key	Default	Meaning
`profile`	`"standard"`	`quick`, `standard`, `deep`, `history-only`, `architecture-only`, `dependencies-only`, or `security-only`. See profiles.
`include_git`	`true`	Analyze Git history.
`include_history`	`true`	Reconstruct evolution: epochs, events, and Time Machine snapshots.
`include_architecture`	`true`	Infer architecture.
`include_dependencies`	`true`	Analyze manifests and lockfiles.
`include_quality`	`true`	Compute quality signals.
`include_security`	`true`	Scan for security signals.
`include_tests`	`true`	Detect tests.
`include_build`	`true`	Detect build systems and environment requirements.
`include_docs`	`true`	Analyze documentation.
`respect_gitignore`	`true`	Respect `.gitignore`, `.ignore`, and Git exclude files.
`max_file_bytes`	`2097152`	Files larger than this are classified but not read.
`max_commits`	`50000`	Most commits to analyze (`0` = unlimited).
`artifact_commits`	`1000`	Most commits stored in the artifact.
`max_file_edges`	`25000`	Most file-level dependency edges stored.
`max_symbols`	`25000`	Most symbols stored.
`snapshots`	`12`	Number of Time Machine snapshots to sample.

The include_* switches turn stages off in any profile; they never turn on a stage that the profile does not run.

`[ignore]`: paths to exclude#

Key	Default	Meaning
`patterns`	`[]`	Additional gitignore-style patterns.
`use_default_patterns`	`true`	Also exclude `node_modules/`, `bower_components/`, `.venv/`, `__pycache__/`, `.tox/`, `.mypy_cache/`, `.pytest_cache/`, `.gradle/`, `.next/`, `.nuxt/`, `.turbo/`, and `.parcel-cache/`.

`[classification]`: override how files are classified#

Each key takes glob patterns.

Key	Meaning
`generated`	Treat matching paths as generated code.
`vendor`	Treat matching paths as vendored third-party code.
`tests`	Treat matching paths as tests.
`docs`	Treat matching paths as documentation.
`source`	Treat matching paths as first-party source even if they look generated or vendored.

RepoDNA also reads linguist-generated, linguist-vendored, and linguist-documentation attributes from .gitattributes files.

`[thresholds]`: when signals are reported#

Key	Default	Meaning
`large_file_lines`	`1000`	Code lines above which a file is reported as large.
`large_function_lines`	`80`	Lines above which a function is reported as long.
`high_complexity`	`15`	Approximate cyclomatic complexity above which a function is reported.
`deep_nesting`	`5`	Block nesting depth above which a function is reported.
`duplicate_min_tokens`	`70`	Minimum duplicated block size, in normalized tokens.
`similarity`	`0.8`	Minimum estimated similarity (0–1) for similar-file pairs.
`hotspot_min_commits`	`3`	Minimum commits for a file to be ranked as a hotspot.
`high_churn_share`	`0.3`	Share of recent churn (0–1) above which a directory is a high-churn area.
`dependency_concentration`	`0.3`	Share of internal module edges (0–1) pointing at one module above which concentration is reported.
`dormant_days`	`90`	Minimum gap in days to report a dormant period.
`recent_days`	`90`	Size of the "recent" window, in days before the latest commit.
`stale_manifest_days`	`730`	Days a manifest must be unchanged before a stale-declarations signal is reported.
`large_binary_bytes`	`5242880`	Size above which a committed binary file is reported.

`[[suppress]]`: accept known findings#

Each [[suppress]] table hides matching findings from CI failures and marks them as suppressed in reports (they stay visible, with the reason).

Key	Required	Meaning
`rule`	yes	Rule identifier or wildcard, for example `security.secret` or `quality.*`.
`reason`	yes	Why the finding is accepted.
`path`	no	Glob; the rule then applies only to findings about matching paths.
`id`	no	One exact finding identifier, as shown by `repodna findings`.

Rule identifiers are listed in the rules reference.

`[report]`: report defaults#

Key	Default	Meaning
`theme`	`"professional"`	`professional`, `minimal`, `technical`, or `dark`.
`privacy`	`"local"`	Privacy preset applied to exported reports: `local`, `share`, or `public`.
`branding`	`true`	Show the "RepoDNA · Made by the Sanskar" credit in generated visuals.
`sections`	`[]`	Report sections to include (empty = all); see `repodna show --list`.

`[privacy]`#

Key	Default	Meaning
`telemetry`	`false`	Always `false`: RepoDNA collects no telemetry. The key exists to make that explicit.
`remote_ai`	`false`	Allow AI providers that send data to other machines (user configuration only).
`anonymize_contributors`	`false`	Replace contributor names with pseudonyms in artifacts.
`include_commit_messages`	`true`	Store commit subject lines in artifacts.
`redact_paths_in_logs`	`false`	Replace repository paths in log output with hashes.

`[performance]`#

Key	Default	Meaning
`cache`	`true`	Reuse per-file results for unchanged files. Entries are keyed by file content, language definition, and RepoDNA version.
`parallelism`	`"auto"`	Worker threads: `"auto"` (one per CPU) or a number.

`[plugins]` (user configuration only)#

Key	Default	Meaning
`enabled`	`[]`	Names of plugins to run.
`directories`	`[]`	Additional directories to search (relative paths are relative to the configuration directory).
`timeout_seconds`	`60`	Time limit per plugin.
`max_output_bytes`	`8388608`	Most bytes a plugin may write to standard output.

See plugins.

`[execution]` (user configuration only)#

RepoDNA detects build and test commands but never runs them unless you allow it here.

Key	Default	Meaning
`allow_build_commands`	`false`	Run detected build commands.
`allow_test_commands`	`false`	Run detected test commands.
`timeout_seconds`	`600`	Time limit per command.
`max_output_lines`	`200`	Lines of output kept per command.

See tests and build.

`[ai]` (user configuration only)#

Key	Default	Meaning
`provider`	`"none"`	`none`, `command`, `openai-compatible`, or `anthropic`.
`command`	`[]`	Command line for the `command` provider.
`endpoint`	none	Base URL for HTTP providers (required for `openai-compatible`; `anthropic` defaults to `https://api.anthropic.com`).
`model`	none	Model identifier (required for `openai-compatible` and `anthropic`).
`api_key_env`	none	Name of the environment variable that holds the API key. Keys are never stored in files. When unset, `anthropic` reads `ANTHROPIC_API_KEY`, but only for Anthropic's own endpoint.
`max_context_tokens`	`6000`	Most estimated tokens of repository context per request.
`max_output_tokens`	none	Most tokens the model may generate (default 16000 for `anthropic`, 2000 for the others).
`timeout_seconds`	`300`	Request time limit. Models that reason before answering can need minutes.
`input_cost_per_million`, `output_cost_per_million`	none	Prices you set yourself, used only for cost estimates.
`include_source_excerpts`	`false`	Allow short source excerpts in prompts.

See AI.

Configuration

What a repository can and cannot set#

Getting started#

Example repodna.toml#

Example user configuration#

Reference#

[analysis]: what to analyze#

[ignore]: paths to exclude#

[classification]: override how files are classified#

[thresholds]: when signals are reported#

[[suppress]]: accept known findings#

[report]: report defaults#

[privacy]#

[performance]#

[plugins] (user configuration only)#

[execution] (user configuration only)#

[ai] (user configuration only)#