What is JellyClaw?

JellyClaw is the Local AI Office: a multi-agent orchestration platform that runs a hierarchy of AI agents — a CEO agent, department heads, and workers — entirely on your own hardware via Ollama. You describe the team in YAML, hand it a goal, and it works overnight, reporting results to your Telegram.

Architecture overview

Three layers, one direction of delegation:

• CEO Agent — takes your goal and breaks it into department-level work.
• Department Heads — plan their area, delegate tasks to workers, review results.
• Worker Agents — execute individual tasks on local models and report back up.

Around the hierarchy sit the MVP-scope pieces: Ollama runs the models locally, Claude API is an optional escalation path for hard tasks, SQLite stores state on disk, Telegram carries reports to you, and GitHub is where agents read and write code.

Configuration

A team is one YAML file. The dev-team template concept:

# teams/dev-team.yaml — planned format for the CLI MVP
team: dev-team
ceo:
  model: qwen2.5:14b          # runs locally via Ollama
departments:
  engineering:
    head: llama3.1:8b
    workers: 2
report:
  channel: telegram

Next: Quick start.