Configuration

Config file: ~/.the-brain/config.json

Minimal Configuration

The absolute minimum to get started:

{
  "daemon": { "pollIntervalMs": 30000 },
  "database": { "path": "~/.the-brain/global/brain.db" },
  "activeContext": "global",
  "contexts": {},
  "extensions": [],
  "plugins": [
    { "name": "@the-brain-dev/plugin-graph-memory", "enabled": true },
    { "name": "@the-brain-dev/plugin-spm-curator", "enabled": true }
  ]
}

Configuration Topics

Topic	Reference
Full JSON schema — all fields, types, defaults	Configuration Schema
Plugin config — graph memory, SPM, identity anchor	Config Schema → Plugins
MLX training — model path, schedule, parameters	Config Schema → MLX
Server mode — local vs remote, ports, auth	Config Schema → Server
Environment variables — overrides for paths, tokens	Environment Variables

Quick Tips

Config changes require daemon restart: the-brain daemon stop && the-brain daemon start
Secrets go in env vars, not in config.json (auth tokens, API keys)
Plugin loading order matters — hooks fire in registration order
Project contexts get their own DB, wiki, and LoRA checkpoints — see CLI Reference → init

Pipeline Configuration

Control which plugins load when the daemon starts. Set pipeline in config.json:

{
  "pipeline": {
    "harvesters": ["cursor", "claude"],
    "layers": {
      "instant": true,
      "selection": true,
      "deep": true
    },
    "outputs": ["auto-wiki"],
    "training": { "mlx": false },
    "llm": true
  }
}

Pipeline Fields

Field	Type	Default	Description
`harvesters`	`string[]`	`["cursor","claude"]`	Which AI tool logs to harvest. Options: `cursor`, `claude`, `hermes`, `gemini`, `windsurf`, `lm-eval`
`layers.instant`	`boolean`	`true`	Graph Memory — instant corrections
`layers.selection`	`boolean`	`true`	SPM Curator — surprise-gated filtering
`layers.deep`	`boolean`	`true`	Identity Anchor — stable self-vector
`outputs`	`string[]`	`["auto-wiki"]`	Output plugins. Options: `auto-wiki`
`training.mlx`	`boolean`	`false`	Overnight LoRA training (Apple Silicon only)
`llm`	`boolean`	`true`	LLM backend for data curator classification

Managing Pipeline

Interactive: the-brain setup — full TUI wizard
Status: the-brain setup --status
Quick toggle: the-brain setup --enable cursor --disable hermes
Layers: the-brain setup --layer-deep off
Restart required: the-brain daemon stop && the-brain daemon start

If pipeline is omitted from config, all plugins load (backward compatible).