coworker

Overview

Coworker is an open-source AI agent that acts as a team member rather than a tool. It remembers past conversations, tracks tasks across sessions, manages your Google Workspace, and can build internal dashboards — all through natural language chat. This is a fork of the original coworker by Array Ventures.

Built on Mastra and designed for team deployment on the Astro platform.

Usage

Chat naturally. Coworker picks the right tools based on context:

• Ask questions — it searches memory for relevant past context before answering
• Request tasks — it tracks multi-step work with a visible task list and checks in when done
• Google Workspace — "summarise my unread emails", "schedule a meeting Thursday 3pm", "share the Q2 doc with the team"
• Build tools — "create a dashboard showing our open GitHub issues" — it scaffolds, commits, and maintains the app
• Schedule work — "every Monday morning, summarise last week's activity and post to Slack"
• Install skills — "install the Jira skill from ClawHub" to extend what it can do

Inputs

Input	Required	Description
MODEL	No	Model to use, e.g. anthropic/claude-sonnet-4-5 or openai/gpt-4o
ANTHROPIC_API_KEY	If using Anthropic	Anthropic API key
OPENAI_API_KEY	If using OpenAI	OpenAI API key
GOG_GOOGLE_CLIENT_ID	For Google Workspace	OAuth client ID from Google Cloud Console
GOG_GOOGLE_CLIENT_SECRET	For Google Workspace	OAuth client secret
GOG_KEYRING_PASSWORD	Recommended	Encrypts stored Google OAuth tokens at rest

Limitations

• Google Workspace requires a Google Cloud OAuth app — see README for setup steps
• WhatsApp sessions are not persisted across redeployments (re-scan QR on restart)
• Scheduled tasks run in-process — they pause if the agent container restarts
• Local LLM providers (LM Studio, Ollama) require the agent to share a network with the provider