whygraph.example.toml

# WhyGraph configuration — whygraph.toml
#
# Scaffolded by `whygraph init`. Every field is optional and the values shown
# are the built-in defaults, so an unedited file behaves exactly as if no
# config were present. Uncomment and edit to override.

log_level = "INFO"            # DEBUG | INFO | WARN | ERROR | CRITICAL

[scan]
max_workers = 2               # parallel LLM calls in the diff-analyzer crawler
provider = "off"              # source-control backend for the PR/issue crawl:
                              #   "off"    — skip remote crawl (default)
                              #   "github" — pull PRs/issues from the GitHub remote
                              #   "auto"   — detect from the remote URL (github only, for now)
remote = "origin"             # git remote whose URL is inspected for provider/auto
# token = "ghp_..."           # GitHub token for the gh CLI during the remote crawl.
                              # Default: read GH_TOKEN / GITHUB_TOKEN from env (or an
                              # existing `gh auth login`). Set here to pin a per-project
                              # token — handy when one shared container scans repos
                              # across different orgs. whygraph.toml is gitignored, so a
                              # token here is never committed.

[analyze]
# LLM that writes a per-commit "git diff" description during `whygraph scan`.
provider = "anthropic"        # which [llm.*] adapter to use
# model = "claude-haiku-4-5"  # override the provider's model for analysis only
# max_diff_chars = 50000      # diff truncated past this length before prompting
# large_commit_file_count = 30  # commits touching more files than this are described
                              # per-file on demand (not whole-diff) — keeps a bulk
                              # import / squash commit from costing a repo-wide pass
# pr_origin_min_commits = 5   # recover a squash-merged PR's original commits once it
                              # collapsed >= this many (file-bulk squashes recover too)
# timeout_sec = 60            # per-call timeout; default: the adapter's own

[rationale]
# LLM that writes a structured "why this code exists" rationale card.
provider = "anthropic"        # which [llm.*] adapter to use
# model = "claude-haiku-4-5"  # override the provider's model for rationale only
# timeout_sec = 60            # per-call timeout; default: the adapter's own
# Caps on how much of a squash-merged PR is rendered into the rationale prompt
# (the raw whygraph_evidence_for tool output is uncapped):
# pr_roster_max_commits = 30      # squashed-commit headlines shown per PR
# pr_discussion_max_comments = 20 # PR comments shown per PR
# pr_comment_max_chars = 500      # each PR comment clipped to this length

# Uncomment to override default DB locations (resolved relative to this file):
# whygraph_db  = ".whygraph/whygraph.db"
# codegraph_db = ".codegraph/codegraph.db"

# Optional rotating file log. Console (stderr) logging is always on. Uncomment
# this section to add an *additional* file handler; leave it out (or leave
# `file` unset) to keep file logging off.
# [logging]
# file         = ".whygraph/logs/whygraph.log"   # resolved relative to this file
# level        = "DEBUG"                          # default: inherit top-level log_level
# max_bytes    = 5_000_000
# backup_count = 3

# Per-provider LLM client settings. Each table is consumed by the matching
# adapter in whygraph.services.llm via Adapter.from_config. Omit api_key to
# read the standard env var (ANTHROPIC_API_KEY, OPENAI_API_KEY,
# DEEPSEEK_API_KEY); omit a section entirely to use the defaults shown.

[llm.anthropic]
model = "claude-opus-4-7"
# api_key = "sk-ant-..."      # default: read ANTHROPIC_API_KEY from env
timeout_sec = 60

[llm.openai]
model = "gpt-4o"
# api_key = "sk-..."          # default: read OPENAI_API_KEY from env
# base_url = "..."            # default: https://api.openai.com/v1
timeout_sec = 60

[llm.deepseek]
model = "deepseek-chat"
# api_key = "sk-..."          # default: read DEEPSEEK_API_KEY from env
timeout_sec = 60

[llm.ollama]
model = "llama3"
# host = "http://localhost:11434"
timeout_sec = 120

# `claude_cli` (Python attribute) and `claude-cli` (TOML idiom) both parse.
[llm.claude_cli]
model = "claude-opus-4-7"
# api_key = "sk-ant-..."      # default: subscription billing (strips env var)
timeout_sec = 120