Configuration

erode is configured through environment variables. There are no configuration files.

For GitHub Actions-specific inputs (model-repo, fail-on-violations, etc.), see GitHub Actions.

AI provider

Variable	Description	Default
`AI_PROVIDER`	AI provider to use (`gemini`, `openai`, or `anthropic`)	`gemini`
`GEMINI_API_KEY`	Google Gemini API key	—
`OPENAI_API_KEY`	OpenAI API key	—
`ANTHROPIC_API_KEY`	Anthropic API key (experimental)	—

Architecture model

Variable	Description	Default
`MODEL_FORMAT`	Architecture model format	`likec4`
`LIKEC4_EXCLUDE_PATHS`	Comma-separated paths to exclude from model loading	—
`LIKEC4_EXCLUDE_TAGS`	Comma-separated tags to exclude from model loading	—

Diff limits

Variable	Description	Default
`MAX_FILES_PER_DIFF`	Maximum number of files to include in the diff	`50`
`MAX_LINES_PER_DIFF`	Maximum number of lines to include in the diff	`5000`
`MAX_CONTEXT_CHARS`	Maximum characters of architectural context	`10000`

Large diffs are truncated to stay within these limits. If a PR exceeds them, erode processes the most relevant files first based on the architecture model context.

Provider model overrides

Each AI provider uses two model tiers: a fast model for extraction stages and an advanced model for analysis. Override the defaults with these variables:

Variable	Description
`GEMINI_FAST_MODEL`	Gemini model for Stages 1–2 (extraction)
`GEMINI_ADVANCED_MODEL`	Gemini model for Stages 3–4 (analysis)
`OPENAI_FAST_MODEL`	OpenAI model for Stages 1–2 (extraction)
`OPENAI_ADVANCED_MODEL`	OpenAI model for Stages 3–4 (analysis)
`ANTHROPIC_FAST_MODEL`	Anthropic model for Stages 1–2 (extraction)
`ANTHROPIC_ADVANCED_MODEL`	Anthropic model for Stages 3–4 (analysis)

See AI Providers for default model names and guidance on choosing a provider.

GitHub

Variable	Description	Default
`GITHUB_TOKEN`	GitHub token for API access	—
`GITHUB_TIMEOUT`	Request timeout for GitHub API (ms)	`30000`
`MODEL_REPO_PR_TOKEN`	Separate token for the model repository	Uses `GITHUB_TOKEN`

Token permissions

GITHUB_TOKEN is used to read the source PR and post analysis comments. MODEL_REPO_PR_TOKEN is used to create model update PRs (branches, commits, pull requests) on the model repository and falls back to GITHUB_TOKEN when not set.

Same repository — source code and architecture model live in one repo, so a single token covers everything:

Feature	Permissions
Read PR and diff	Contents: Read, Pull requests: Read
Post analysis comments	Issues: Read and write
Create model update PR (`--open-pr`)	Contents: Read and write, Pull requests: Read and write

External model repository — source and model are in separate repos, each with its own token:

Token	Repository	Permissions
`GITHUB_TOKEN`	Source repo	Contents: Read, Pull requests: Read, Issues: Read and write
`MODEL_REPO_PR_TOKEN`	Model repo	Contents: Read and write, Pull requests: Read and write

Fine-grained PATs

Select these Repository permissions when creating a fine-grained personal access token:

Contents — Read-only (or Read and write if using --open-pr on that repo)
Pull requests — Read-only (or Read and write if using --open-pr on that repo)
Issues — Read and write (source repo only)

Classic PATs

The repo scope covers all required permissions. If the model repository is public, public_repo is sufficient for MODEL_REPO_PR_TOKEN.

GitHub Apps (recommended for organizations)

GitHub Apps are the recommended token strategy for organizations:

Short-lived tokens — automatically generated and rotated on every workflow run, eliminating long-lived secrets
Repository-scoped — access is limited to specific repositories, not broad user-level access
Not tied to user accounts — tokens keep working when people leave the organization or change roles
Centralized permissions — managed through the App’s installation settings, not individual developer tokens

Use the same Repository permissions as fine-grained PATs above. See GitHub App Token for a complete workflow example.

GitLab (experimental)

Variable	Description	Default
`GITLAB_TOKEN`	GitLab token with `api` scope	—
`GITLAB_BASE_URL`	GitLab instance URL	`https://gitlab.com`

Token permissions

GITLAB_TOKEN is used for all operations on the source project — reading MR diffs, posting notes, and (with --open-pr) creating branches, commits, and merge requests. The api scope is required; read_api is not sufficient.

For external model projects, the CI entrypoint accepts LIKEC4_MODEL_REPO_TOKEN (see GitLab CI).

Type	Scope	Minimum role
Personal Access Token	`api`	—
Project Access Token	`api`	Developer
Group Access Token	`api`	Developer

Bitbucket (experimental)

Variable	Description	Default
`BITBUCKET_TOKEN`	Bitbucket app password or token	—
`BITBUCKET_BASE_URL`	Bitbucket API base URL	`https://api.bitbucket.org/2.0`

Token permissions

BITBUCKET_TOKEN handles all operations — there is no separate model-repo token. If the token contains : (e.g. username:app_password), erode uses HTTP Basic auth; otherwise it uses Bearer auth.

Feature	App password scopes
Read PRs and diffs	Repositories: Read
Post PR comments	Pull requests: Write
Create model update PR (`--open-pr`)	Repositories: Write, Pull requests: Write

Minimum scopes (no --open-pr): Repositories: Read + Pull requests: Write. Full scopes (with --open-pr): Repositories: Write + Pull requests: Write.

Repository access tokens and workspace access tokens use the same permission categories but authenticate with Bearer auth.

Timeouts

Variable	Description	Default
`GEMINI_TIMEOUT`	Request timeout for Gemini API calls (ms)	`60000`
`OPENAI_TIMEOUT`	Request timeout for OpenAI API calls (ms)	`60000`
`ANTHROPIC_TIMEOUT`	Request timeout for Anthropic API calls (ms)	`60000`
`GITHUB_TIMEOUT`	Request timeout for GitHub API calls (ms)	`30000`

Debug

Variable	Description	Default
`DEBUG_MODE`	Enable debug output	`false`
`VERBOSE`	Enable verbose logging	`false`