The 📝OpenClaw Glossary is a collection of terms used across the open-source personal AI assistant platform, for those building, operating, or studying markdown-file-defined agent systems.
Glossary
A
- 📝AGENTS.md — registry of sub-agents an OpenClaw agent can summon or delegate tasks to.
- Agent workspace — the directory (default
~/.openclaw/workspace/) holding an OpenClaw agent's configuration files and memory.
B
- 📝BOOTSTRAP.md — initialization file defining module load order, processing rules, and baseline state for an OpenClaw agent.
C
- Clawdbot — the original name of OpenClaw before rebranding; the project created by Peter Steinberger.
D
- Daily logs — append-only running notes stored at
memory/YYYY-MM-DD.md; today and yesterday auto-load at every session start. - 📝Daytona — secure sandbox infrastructure for AI-generated code, evaluated as an isolated execution layer for OpenClaw agents.
- Dreaming — optional background process that promotes short-term memory signals from daily logs into MEMORY.md after scoring and review via DREAMS.md.
H
- 📝HEARTBEAT.md — manages an agent's autonomous pulse, scheduled tasks, and periodic self-reflection loops.
I
- 📝IDENTITY.md — defines an agent's name, role, creator, organizational affiliation, and operational boundaries.
L
- 📝LLM Export Prompts — extraction prompts for pulling user context out of one LLM to seed another agent's USER.md.
M
- 📝MEMORY.md — curated long-term memory file loaded at every agent session; the distilled, durable layer above daily logs.
- memory_flush — pre-compaction save triggered when a session approaches the context limit; prevents memory loss before summarization.
- memory_get — agent tool for targeted Markdown file reads by path and optional line range.
- memory_search — agent tool for semantic recall over indexed memory files using sqlite-vec embeddings.
- 📝Multi-Agent Routing — distributed orchestration enabling multiple isolated agent instances within a single gateway process.
O
- 📝OpenClaw — open-source personal AI assistant platform with a markdown-file agent definition system.
- 📝OpenClaw Token Caching — prompt caching implementation reducing agent operating costs by up to 90% via cached content billed at 10% of standard token rates.
P
- 📝PDF to RAG Ingestion — locally-hosted knowledge retrieval pipeline that converts PDFs and Markdown into ChromaDB embeddings for agent semantic query.
S
- 📝Security Hardening for OpenClaw — comprehensive security checklist for deploying and maintaining an OpenClaw multi-agent system.
- Session spawning — pattern where parent agents create child agents with specific tasks, enabling delegated execution and multi-agent orchestration.
- 📝SKILLS.md — registry of installable capability modules, each packaged with scripts and a SKILL.md interface.
- 📝Social Media Management — autonomous content workflow using OpenClaw agents for video generation and scheduling.
- 📝SOUL.md — defines an agent's personality, temperament, values, and behavioral guardrails.
T
- 📝Token Optimization for OpenClaw — practices for managing OpenClaw workspace token usage efficiently.
- 📝TOOLS.md — inventory of external capabilities — APIs, CLIs, functions — that an agent is authorized to call.
- 📝TOOLS.md vs SKILLS.md vs MCP — clarifier disambiguating capability inventory, workflow catalog, and connection protocol.
U
- 📝USER.md — long-term dossier on the human user; preferences, habits, professional background.