Braingent

Reference

Configuration

Optional configuration knobs — most users never need them. Defaults are designed to be right.

Braingent’s defaults are intentional: the smallest configuration that makes search, capture, and validation work out of the box. Most users never touch any of these knobs. If you do, here’s what’s available.

Configuration sources, in priority order

Higher in this list wins.

  1. CLI flags (--path, --strict, etc.).
  2. Environment variables (BRAINGENT_PATH, BRAINGENT_NO_COLOR).
  3. Repo-local config (.braingent/config.toml inside the memory repo).
  4. User-level config (~/.braingent/config.toml).
  5. Built-in defaults.

Repo-local config

Optional. Lives at .braingent/config.toml inside the memory repo.

# .braingent/config.toml
version = 1

[memory]
# Override the default record kind directories.
# Most users leave this alone.
decisions_dir = "decisions"
reviews_dir   = "reviews"
learnings_dir = "learnings"
tasks_dir     = "tasks"

[task_ids]
# Format for task IDs. Default: BGT-NNNN, zero-padded to 4.
prefix = "BGT"
pad    = 4

[capture]
# Trigger phrases that mean "capture this".
phrases = [
  "capture this",
  "save to braingent",
  "write to braingent",
  "task done",
  "done thanks",
  "ok done",
]

# Whether to auto-capture on PR open / merge / ticket close.
auto_capture_on_pr     = true
auto_capture_on_ticket = true

[indexes]
# Which indexes to regenerate on `braingent reindex`.
generate = ["by-topic", "by-repo", "by-tool", "decisions-index", "recent"]
recent_count = 50

[validate]
# Status vocabularies per record kind. Override only if you really mean it.
[validate.task]
allowed_status = ["planned", "in_progress", "done", "abandoned", "blocked"]

[mcp]
# Read-only mode. Currently every tool is read-only; this flag is
# future-proof for write tools.
read_only = false

braingent doctor validates this file at startup.

User-level config

~/.braingent/config.toml. Same shape as the repo-local file. Useful for personal preferences (clipboard tool, editor) you don’t want to commit.

[ui]
editor    = "code"           # what `braingent open <id>` launches
clipboard = "pbcopy"          # for `--copy` flags

[print_prompts]
default_agent = "claude"

Privacy / safety knobs

[safety]
# Paths that should never appear inside committed Markdown.
# `braingent doctor` flags violations.
forbid_paths = ["~/private", "/Users/secrets"]

# Patterns that should never appear in record bodies.
forbid_patterns = [
  "AKIA[0-9A-Z]{16}",            # AWS access key ids
  "ghp_[A-Za-z0-9]{36}",         # GitHub PATs
  "sk-[A-Za-z0-9]{40,}",         # OpenAI / Anthropic-style keys
]

forbid_patterns is a defense-in-depth — the manifesto ships with the most common patterns enabled. Add yours.

Bootstrap hooks

.braingent/hooks/ (optional). Each script runs at a known lifecycle moment, if it exists.

HookWhen it runs
pre-capture.shBefore a record is written. Receives the draft frontmatter on stdin. Exit non-zero to abort.
post-capture.shAfter capture commits. Receives the commit SHA.
pre-update.shBefore braingent update applies changes.
post-doctor.shAfter doctor runs. Receives the report on stdin.

Hooks are plain shell scripts. Keep them small. If you find yourself writing a long hook, you probably want a workflow under workflows/ instead.

MCP server

The MCP server reads its config from CLI args, not a file. See Installation → MCP Server for the JSON config blocks per agent.

Where to go next