Configuration Reference

The complete config schema for pi-vault-mind. Two config files exist, and they merge (global → project).

File locations

Scope	Path	Applies
Global	~/.pi/agent/vault-mind.config.json	All projects, all directories
Project	./pi-vault-mind.config.json (or .pi/vault-mind.config.json)	This project only

Use /wiki setup to write the global config. Use /wiki init to write a project config.

Top-level shape

{
  "version": 2,
  "wiki": { ... },
  "collections": { ... },
  "injectors": [ ... ],
  "extensionCompatibility": { ... }
}

wiki — embedding, graph, vaults

{
  "wiki": {
    "dataDir": ".lancedb",
    "embedding": {
      "provider": "transformers",
      "ollamaModel": "embeddinggemma",
      "ollamaHost": "http://127.0.0.1:11434"
    },
    "graph": {
      "enabled": true,
      "canvasSync": true
    },
    "ftsEnabled": true,
    "httpPort": 11435,
    "vaults": {
      "default": {
        "path": "/home/you/Obsidian/MyVault",
        "autoSync": true,
        "autoSyncTags": ["decision", "insight", "requirement"],
        "autoSyncMinLength": 200
      }
    }
  }
}

Field	Type	Default	Description
dataDir	string	.lancedb	Where to put the LanceDB files (relative to cwd or absolute)
embedding.provider	enum	transformers	transformers (offline, downloads all-MiniLM-L6-v2) or ollama
embedding.ollamaModel	string	embeddinggemma	Model name when provider = ollama
embedding.ollamaHost	string	http://127.0.0.1:11434	Ollama API endpoint
graph.enabled	boolean	true	Extract entity-relation edges from entries
graph.canvasSync	boolean	true	Sync the entity graph to Obsidian .canvas files
ftsEnabled	boolean	true	Build a Tantivy full-text index alongside the vector index
httpPort	number	11435	Port for the HTTP server (Obsidian ↔ pi bridge). Binds 127.0.0.1.
vaults	object	{}	Map of vault name → vault config (see below)
vaults.<name>.path	string	(required)	Absolute path to the Obsidian vault
vaults.<name>.autoSync	boolean	true	Auto-write substantial entries to the vault
vaults.<name>.autoSyncTags	string[]	[]	Tags that always trigger auto-sync, regardless of length
vaults.<name>.autoSyncMinLength	number	200	Minimum character count for auto-sync

collections — JSONL storage definitions

{
  "collections": {
    "main": {
      "path": "collections/main.jsonl",
      "schema": ["id", "domain", "source", "fact", "tag", "artifact"],
      "dedupField": "fact"
    },
    "pending": {
      "path": "collections/pending.jsonl",
      "schema": "main"
    }
  }
}

Field	Type	Description
<name>.path	string	File path (relative to project root, or absolute)
<name>.schema	string[] | string	Field names. Use a string to inherit another collection's schema
<name>.dedupField	string	Field to check for duplicates in autopilot mode

Built-in collections

These are created by /wiki init and are typically always present:

• main — the primary collection for substantial knowledge
• pending — staging area for gated mode and promote_wiki
• context_events — auto-appended when pi-context integration is enabled

Example domain: research

{
  "collections": {
    "findings": {
      "path": "research/findings.jsonl",
      "schema": ["id", "paper", "claim", "evidence", "confidence", "tag"],
      "dedupField": "claim"
    }
  }
}

Example domain: decision log

{
  "collections": {
    "decisions": {
      "path": "decisions/log.jsonl",
      "schema": ["id", "date", "context", "decision", "rationale", "status", "owner"],
      "dedupField": "decision"
    }
  }
}

injectors — context injection

Injectors automatically inject matching collection entries into the LLM system prompt when the user types a matching pattern.

{
  "injectors": [
    {
      "name": "draft-context",
      "regex": "draft\\s+(\\S+)",
      "collection": "main",
      "filterField": "tag"
    }
  ]
}

Field	Type	Description
name	string	Human-readable name
regex	string	Regex pattern matched against the user's prompt
collection	string	Which collection to query on match
filterField	string	Field to use as the filter (typically tag)
artifactPath	string	Optional — file to also inject as raw markdown

Example — lit-review trigger:

{
  "injectors": [
    {
      "name": "lit-review",
      "regex": "review\\s+(\\S+)",
      "collection": "findings",
      "filterField": "tag",
      "artifactPath": "research/synthesis.md"
    }
  ]
}

User types: review transformers → system prompt is pre-loaded with findings tagged transformers + the contents of research/synthesis.md.

extensionCompatibility — opt-in extensions

{
  "extensionCompatibility": {
    "pi-context": {
      "enabled": true,
      "tagPatterns": [],
      "autoEnableAcm": true,
      "enhanceInjectors": false,
      "indexContextEvents": true
    }
  }
}

Field	Type	Default	Description
pi-context.enabled	boolean	false	Master toggle for pi-context integration
pi-context.tagPatterns	string[]	[]	Regex patterns; if a pi-context tag matches, auto-query relevant collections and inject
pi-context.autoEnableAcm	boolean	true	Auto-trigger /acm on session start to link context state
pi-context.enhanceInjectors	boolean	false	Allow pi-context state to refine native injectors
pi-context.indexContextEvents	boolean	true	Append pi-context events to the context_events collection

See also

• Tools — configure_wiki writes this file
• Commands — /wiki setup and /wiki init write the initial config
• Extension wiring — what extensionCompatibility actually wires up at runtime