OpenClaw Glossary

An autonomous AI entity powered by an LLM that can understand, reason, plan, and execute tasks. In OpenClaw, agents use a language model as their "brain" combined with tools to interact with the world. Agents maintain conversation context and can remember past interactions through memory systems. You can run multiple agents for different purposes.

A unique authentication token required to access LLM services. OpenClaw uses API keys to communicate with providers like OpenAI, Anthropic, or Google. These keys are typically stored in environment variables or the gateway config and should never be shared publicly. Each provider has its own API key format.

The ability to automatically trigger actions based on events or schedules. In OpenClaw, automation is achieved through cron jobs, event listeners, and skills. Common automations include sending daily summaries, monitoring systems, or responding to specific triggers.

An automated user on a messaging platform that can send and receive messages. OpenClaw acts as a bot on channels like Discord, Telegram, or Slack. Bots typically have a username, avatar, and can respond to commands or messages automatically.

Using OpenClaw to control a web browser programmatically. The browser skill allows your agent to navigate websites, fill forms, extract data, and take screenshots. Useful for web scraping, testing, and automating web tasks.

A messaging platform or protocol where OpenClaw can communicate. Channels include Discord, WhatsApp, Telegram, Slack, Signal, SMS, iMessage, and more. Each channel has its own API and authentication method. Multiple channels can be connected simultaneously.

Anthropic's family of large language models. Known for strong reasoning, helpful responses, and the Constitutional AI approach. Available in OpenClaw via API. Popular models: Claude 3.5 Sonnet (balanced), Claude 3 Opus (powerful), Claude 3 Haiku (fast).

Settings that control how OpenClaw behaves. Main config file is gateway.yaml (or gateway.json). Contains channel settings, model selection, skills, memory configuration, system prompts, and more.

The maximum amount of text an LLM can process in a single request. Measured in tokens (roughly 1 token = 1 word). Larger context windows allow for longer conversations and more information per request. GPT-4o supports 128K tokens, Claude 3.5 supports 200K.

A scheduled task that runs automatically at specified times. Uses standard cron syntax: minute hour day month weekday. Used for reminders, periodic reports, automated backups, and proactive actions. Example: 0 9 * * * runs daily at 9 AM.

Installing and running OpenClaw on a server or hosting platform. Can be cloud VPS (DigitalOcean, Linode), serverless (Cloudflare), or self-hosted (home server). Each has different setup requirements and costs.

A containerization platform that packages OpenClaw with all its dependencies. Docker images make installation consistent across systems. Run with docker run -it openclaw/openclaw. Recommended for production deployments.

A numerical representation of text that captures semantic meaning. Used by vector databases for semantic search. OpenClaw converts text to embeddings to enable memory search - finding relevant past information even without exact keyword matches.

A variable stored outside your config file, typically in your shell. Used for sensitive data like API keys. Prefix with OPENCLAW_ or use ${VAR_NAME} in config. More secure than hardcoding secrets.

Training a base model on custom data to improve performance on specific tasks. OpenClaw doesn't do fine-tuning directly but can work with fine-tuned models from providers. More advanced than prompt engineering but requires more resources.

The main entry point and configuration hub for OpenClaw. The gateway.yaml file controls everything: which channels to use, which LLM, what skills are loaded, memory settings, and agent behavior. Think of it as the brain's configuration.

Generative Pre-trained Transformer - OpenAI's LLM family. Available in OpenClaw: GPT-4o (latest, multimodal), GPT-4 Turbo (powerful), GPT-3.5 Turbo (fast, cheap). Known for strong general-purpose abilities and tool use.

Hypertext Transfer Protocol - the foundation of web communication. OpenClaw uses HTTP to communicate with LLM APIs, webhooks, and web-based channels. Understanding HTTP helps with debugging and building integrations.

The process of setting up OpenClaw on your system. Can be done via npm (npm install -g openclaw), Docker, or from source. After installation, run openclaw init to configure.

A collection of documents or information that OpenClaw can search and reference. Can be imported from PDFs, websites, or custom documents. Enables your agent to answer questions about your specific data.

An AI model trained on vast amounts of text to understand and generate human-like text. The "brain" that powers OpenClaw's reasoning. Examples: GPT-4, Claude 3, Gemini. Different models have different strengths, speeds, and costs.

An LLM that runs entirely on your own hardware instead of calling external APIs. Using Ollama, you can run models like Llama 3, Mistral, or CodeLlama locally. Benefits: no API costs, complete privacy, works offline. Tradeoff: requires powerful hardware.

OpenClaw's ability to remember information across conversations. Uses vector databases for semantic storage. Can be configured to forget information after a set time. Types: short-term (conversation), long-term (vector store), working (current context).

A service that offers access to LLMs via API. OpenClaw supports many: OpenAI, Anthropic, Google, Ollama, OpenRouter, Azure OpenAI, and more. Each has different models, pricing, and capabilities.

Node Package Manager - the standard way to install OpenClaw. npm install -g openclaw installs OpenClaw globally. npm also hosts skill packages that extend OpenClaw's capabilities.

A tool for running LLMs locally on Mac/Linux. OpenClaw can use Ollama as a model provider, enabling fully local operation. Popular models: Llama 3, Mistral, CodeLlama, Phi3. Great for privacy and cost savings.

A unified API that provides access to 100+ LLMs from multiple providers. One API key accesses all models. Useful for trying different models and accessing specialized models. OpenClaw supports OpenRouter natively.

An extension that adds functionality to OpenClaw. Similar to skills but may refer to integrations or middleware. Can add new channels, tools, or capabilities. See ClawHub for available plugins.

The text input given to an LLM. In OpenClaw, you craft system prompts that define your agent's personality, capabilities, and rules. The better your prompt, the better your agent performs. Prompt engineering is the art of crafting effective prompts.

Restrictions on how many API calls you can make in a given time period. All LLM providers have rate limits based on your plan. If exceeded, you'll get 429 errors. Solutions: upgrade plan, add delays, use caching, or switch providers.

Running OpenClaw on your own infrastructure rather than using a cloud service. Gives you full control over data, configuration, and costs. Can be on a home server, VPS, or even a Raspberry Pi (for basic use cases).

A conversation thread between a user and the agent. Sessions maintain context - what was discussed previously. Can be configured to last for specific durations. Long sessions consume more tokens.

A reusable extension that adds capabilities to OpenClaw. Skills can add tools, define behaviors, or provide knowledge. Installed via openclaw install @skill-name. Over 700+ skills available at ClawHub.

Converting spoken words into written text. Enables voice input for OpenClaw. Providers: OpenAI Whisper, Deepgram, Google Speech. Required for voice conversation capabilities.

A setting that controls randomness in LLM outputs. Range 0-2 (usually 0-1). Lower = more focused/deterministic. Higher = more creative/unpredictable. For coding, use 0.1-0.3. For creative writing, use 0.7-0.9.

The basic unit of text processing for LLMs. Roughly 1 token = 1 word in English. LLM pricing is per token (input + output). Context windows are measured in tokens. API calls include token counts in response.

A capability the agent can use to interact with the world. Tools include: file operations, command execution, web browsing, message sending, API calls. The agent decides when to use tools based on the user's request.

Converting text into spoken audio. Enables voice output for OpenClaw. Providers: ElevenLabs (natural), OpenAI TTS, Google Cloud TTS, Azure Speech. Voice selection affects quality and character.

A database optimized for storing and searching embeddings. Used by OpenClaw's memory system for semantic search. Popular options: Pinecone, Weaviate, Qdrant, Chroma, Milvus. Enables "remembering" relevant information from past conversations.

An HTTP POST request that notifies your application of events. OpenClaw can receive webhooks from external services (GitHub, Stripe, etc.) and send webhooks to trigger other services. Useful for integrations and automation.