Skip to main content
Operator is a local agent runtime that connects chat messages to LLMs via LiteLLM. It is intentionally small and file-driven:
  • Markdown files for agent prompts, jobs, and skills
  • SQLite for durable runtime state (memory, messages, runs, jobs)
  • A single execution path for both inbound chat and scheduled jobs

Quickstart

Install and run Operator in minutes.

Configuration

Set up models, agents, memory, and transports.

Agents

Define agent personalities with markdown files.

Jobs

Schedule recurring tasks with cron expressions.

​
Core Features

  • Multiple agents with separate prompts and workspaces
  • Model fallback chains (if one model errors, the next is tried)
  • Slack transport with thread continuity
  • Skills discovery from markdown files
  • Scheduled jobs with prerun gating and postrun hooks
  • Vector memory with automatic harvesting and semantic search
  • Key-value store for operational state
  • Turn-safe context truncation against model token budgets

​
Architecture

~/.operator/
├── operator.yaml           # runtime config
├── SYSTEM.md               # system preamble
├── logs/
├── state/
│   └── operator.db         # SQLite database
├── agents/
│   └── <agent>/
│       ├── AGENT.md        # agent prompt
│       └── workspace/      # agent working directory
├── jobs/
│   └── <job>/
│       ├── JOB.md          # job definition
│       └── scripts/        # prerun/postrun hooks
└── skills/
    └── <skill>/
        ├── SKILL.md        # skill definition
        └── scripts|references|assets/