Concepts

These are the terms Quickstart and the config reference assume. Each one has a deeper page linked at the end of its definition.

Alias. An alias is the name you assign to a configured instance, then reference elsewhere to point at it. You choose the name freely; other parts of the config wire things together by that name. Aliases are lowercase ASCII letters, digits, and single underscores, must start and end with a letter or digit, and cannot contain __ or hyphens. In the config they appear as the <alias> segment of a section header, such as [agents.<alias>] or [providers.models.<type>.<alias>]. See Reference → Environment variables → Alias grammar.

Model provider. A model provider is ZeroClaw’s abstraction over an LLM endpoint. Every chat-completion request goes through a provider, whether the target is a remote API, a self-hosted server, or a local Ollama model. Providers are typed by vendor family, and you can run several named instances of the same family. In the config each one lives at [providers.models.<type>.<alias>]. See Model Providers → Overview.

Risk profile. A risk profile is a named autonomy and sandbox posture. Its level is readonly, supervised (the default), or full, controlling whether tools run automatically, prompt for approval, or are blocked. Each agent references exactly one risk profile. In the config it lives at [risk_profiles.<alias>]. See Security → Autonomy Levels.

Runtime profile. A runtime profile is reusable operational tuning: agentic mode, tool-iteration caps, action and cost budgets, timeouts, context limits, and delegation policy. It is separate from the risk profile, which governs autonomy. Quickstart installs the unbounded preset for new agents; adjust the fields afterward. In the config it lives at [runtime_profiles.<alias>]. See Reference → Config.

Multi-agent. ZeroClaw runs many agents from one install. Each agent has its own set of references (risk profile, model provider, channels), its own workspace directory, and its own memory backend. An agent can spawn an ephemeral SubAgent that inherits its parent’s identity and security policy, and agents can talk to each other when they share a peer group. In the config each agent is an [agents.<alias>] block. See Agents → Runtime internals.

Peer group. A peer group declares an opt-in cross-agent communication set on a single channel. Membership is mutual: two agents are peers only when both appear in the same group’s member list. In the config it lives at [peer_groups.<name>]. See Agents → Runtime internals.

Skill bundle. A skill bundle is a reusable group of skills attached to an agent or channel by alias. It controls which skills load and from where: a source directory plus optional include and exclude lists. Skills themselves are reusable instructions and optional tool definitions. In the config a bundle lives at [skill_bundles.<alias>]. See Tools → Skills.