Infrastructure Paws

Infrastructure Paws provide lifecycle hooks and internal services. Unlike Tool Paws, they don't register tools the Brain calls directly — they operate behind the scenes to enrich context, manage state, and provide monitoring.

Available Infrastructure Paws

paw-memory

Persistent memory with source-isolated daily logs and BM25 ranked search.

vole paw add @openvole/paw-memory

Hooks: bootstrap, observe

Storage structure:

.openvole/paws/paw-memory/
├── MEMORY.md       # Shared long-term facts
├── user/           # CLI session logs
├── paw/            # Telegram/Slack logs
└── heartbeat/      # Heartbeat logs

Memory is scoped by task source — CLI conversations, channel messages, and heartbeat tasks each get their own daily logs. This prevents cross-contamination between sources.

Tools: memory_read (read a memory file), memory_write (append an entry), memory_search (BM25 ranked search), memory_list (list memory files).

paw-session

Conversation continuity across messages. Auto-expiring transcripts per session ID.

vole paw add @openvole/paw-session

Hooks: bootstrap, observe

Session data lives in .openvole/paws/paw-session/, organized by session ID:

.openvole/paws/paw-session/
├── cli:default/     # CLI session transcript
├── telegram:123/    # Telegram chat transcript
└── slack:C456/      # Slack channel transcript

The session paw loads previous messages from the current session during bootstrap, giving the Brain conversation context.

Tools: session_history (read a session's transcript), session_list (list active sessions), session_clear (remove a session), and session_append — append a single transcript entry outside the Brain loop (e.g. peer-to-peer human chat). session_append takes an optional maxMessages retention cap that trims the transcript to the last N messages after writing.

paw-compact

Context compaction — summarizes old messages when context grows too large.

vole paw add @openvole/paw-compact

Hooks: compact

Runs as an in-process paw (not subprocess) for performance. When the message count exceeds compactThreshold, it:

1. Extracts key information from old messages (tool calls, results, responses, errors)
2. Replaces middle messages with a structured summary
3. Keeps the first message (original input) + recent N messages verbatim

No LLM needed — pure extraction. Fast and free.

paw-dashboard (deprecated)

WARNING

@openvole/paw-dashboard is deprecated in favor of vole serve — the control-plane dashboard that manages all your agents from one web server. It still works but logs a deprecation warning on load and will be removed in a future release. Use vole serve for a UI instead.

The legacy single-engine web control panel — monitor and operate one agent from your browser: paws (with health), tools, skills, tasks, schedules, and live events, plus editing config and identity files and restarting the engine.

vole paw add @openvole/paw-dashboard

Env Variable	Purpose
VOLE_DASHBOARD_PORT	Port for the dashboard (default: 3001)

Configuration:

{
  "name": "@openvole/paw-dashboard",
  "allow": {
    "listen": [3001],
    "env": ["VOLE_DASHBOARD_PORT"]
  }
}

The dashboard connects to the engine's message bus and streams state updates over WebSocket — no polling. It's a full control panel, not just a viewer:

• Config editor — edit vole.config.json across 8 sections (brain, heartbeat, loop, security/Docker, paws, tool profiles, agents, net)
• Identity editor — edit SOUL.md, USER.md, AGENT.md, HEARTBEAT.md, and BRAIN.md
• One-click restart — apply config/identity changes in-process, no terminal needed
• Live event log — task lifecycle, paw/tool registration, crashes, rate limits, VoleNet executions

Embedded Dashboard Panels

Any paw — not just infrastructure paws — can ship its own UI that appears under the dashboard's Apps tab, with no per-paw web server and no extra port. See Build an Embedded App for the how-to; @openvole/paw-markets is the reference example.