A comprehensive reference for OpenViber terminology and concepts.

A

Artifact

A living document within a Space that evolves over time. Artifacts can be:

• Documents (Markdown, text, HTML)
• Code files
• Data files (JSON, CSV)
• Binary files (images, PDFs)

Each artifact maintains version history, allowing you to track changes and rollback if needed.

Adapter

A storage backend implementation. OpenViber supports:

• Local Adapter: SQLite + Filesystem (default)
• Supabase Adapter: PostgreSQL + Supabase Storage (cloud)

C

Context

The accumulated information that tasks use to understand and respond to requests. Context includes:

• Conversation history
• Artifact contents
• Space goal and metadata
• Previous decisions

Conversation History

The complete record of messages between users and tasks within a Space. History persists across sessions, enabling tasks to maintain continuity.

D

Daemon (Viber Runtime)

The local process running on a Viber that executes tasks and connects outbound to the Gateway. In docs, “daemon” and “Viber runtime” are used interchangeably.

G

Gateway

The central coordinator that routes messages between Viber runtimes (daemons) and the web app. Started via viber gateway. Vibers connect outbound to the gateway via WebSocket; the web app (Viber Board) talks to the gateway via REST and SSE. This is distinct from the Channels server (enterprise channel webhooks) and from the Skill Hub (external skill registry).

H

History

See Conversation History.

Hub (Skill Hub)

The external skill registry for discovering and importing skills from sources like OpenClaw, GitHub, npm, and others. See src/skills/hub/ for the implementation. Not to be confused with the Gateway (central coordinator).

Channels (enterprise channel server)

The HTTP server that receives webhooks from enterprise channels (DingTalk, WeCom, Discord, Feishu). Started via viber channels. Implemented by ChannelGateway in src/channels/gateway.ts. Distinct from the Gateway (central coordinator for Vibers).

J

Job

A scheduled task defined as a YAML file that runs automatically on a cron timer. Jobs specify a schedule, a prompt, and optional configuration (model, skills, tools). They are stored in ~/.openviber/jobs/.

When a job fires, the JobScheduler creates a Task (Agent) with the job’s configuration and executes the prompt. Jobs can leverage skills for domain knowledge — for example, a health-check job uses the antigravity skill.

See Jobs for full documentation.

Job Scheduler

The JobScheduler class that manages cron-based job execution. It reads YAML job files from disk, registers cron triggers via Croner, and creates Task instances to execute job prompts on schedule.

M

Message

A single unit of communication within a conversation. Messages have:

• role: user, assistant, or system
• content: The message text
• metadata: Additional information (timestamps, task info)

Metadata

Additional information attached to messages, Spaces, or artifacts. Used for tracking, filtering, and coordination.

Mode

The operational mode for task interactions:

• Always Ask: Task asks before each action
• Task Decides (or Viber Decides): Task acts within policy boundaries
• Always Execute: Maximum autonomy, minimal interruption

P

Persistence

OpenViber’s core capability of saving and restoring workspace state. Persistence enables:

• Session continuity across restarts
• Multi-day workflows
• Collaboration across time

Provider

An LLM service provider. Supported providers:

• OpenAI: GPT-4, GPT-4o, GPT-4o-mini
• Anthropic: Claude 3.5 Sonnet, Claude 3 Opus
• DeepSeek: DeepSeek Chat, DeepSeek Coder
• Google: Gemini 1.5 Pro, Gemini 1.5 Flash
• OpenRouter: Access to multiple providers through a single API

S

Skill

A reusable bundle of domain knowledge that teaches a task how to approach specific tasks. A skill is a directory containing a SKILL.md file (frontmatter + instructions) and optionally an index.ts that exports specialized tools.

Skills are discovered from:

• ~/.openviber/skills/ (user-defined)
• src/skills/ (built-in: antigravity, cursor-agent, codex-cli, github, tmux)

Unlike tools (which provide actions), skills provide knowledge and context that gets injected into the task’s system prompt. Skills can also bundle their own tools — for example, the github skill provides gh_list_issues, gh_create_pr, etc.

See Skills for full documentation.

Skill Registry

The SkillRegistry class that manages skill discovery, loading, and tool registration. It scans skill directories for SKILL.md files, parses metadata, and lazily loads tools from index.ts.

Space

A working directory that tasks operate in. Spaces live at ~/openviber_spaces/ by default, but tasks can be pointed at any directory (e.g., an existing Git repo). A Space can be:

• A cloned Git repository (code projects)
• A research folder (non-code work)
• An output directory (reports, generated content)

Multiple tasks can work on the same Space.

T

Tool

A capability that extends what tasks can do. Tools allow tasks to:

• Fetch external data
• Read/write files
• Search the web
• Execute code
• Interact with APIs

Task

A role-scoped unit of work or assignment that runs on a Viber. Each task has its own:

• Persona — Name, personality, communication style (adopted by the Viber for this task)
• Goals — What it’s designed to accomplish
• Tools — What actions it can take
• Skills — Domain knowledge it applies
• Model — Which LLM provider it uses

Tasks are configured through YAML files in ~/.openviber/vibers/.

V

Viber

A machine running the OpenViber runtime that executes tasks. The Viber is the AI Worker. A Viber provides:

• Runtime — The Viber runtime (daemon) process that executes tasks and connects to the Gateway
• Scheduler — Cron-based job scheduling for automated tasks
• Credentials — Shared account access for hosted tasks
• Config — Identity and settings at ~/.openviber/ (lightweight, portable)
• Spaces — Working data at ~/openviber_spaces/ (repos, research, outputs)

Vibers connect to the OpenViber Board via a one-time token command (npx openviber connect --token ...). Multiple tasks on one machine coordinate through external systems (GitHub, email) rather than direct inter-task messaging.

ViberAgent

The core class that orchestrates a task’s execution. ViberAgent:

• Processes user requests through an LLM
• Coordinates tool calls and skill loading
• Maintains context across sessions
• Reports progress and results

Quick Reference

Core Concepts

Common Patterns