docketeer 0.0.3

A reasonably sized autonomous AI construction kit

Docketeer

The core agent engine for building autonomous AI assistants with Anthropic and Docket.

Docketeer is a small, opinionated toolkit for running an AI agent that can manage its own memory, schedule its own future work, and extend itself through plugins. The core package provides the agent loop, a persistent workspace for the agent's files and journal, and a plugin system based on standard Python entry points.

Tools

Workspace

• list_files — list files and directories in the workspace
• read_file — read contents of a text file
• write_file — write content to a text file
• delete_file — delete a file
• search_files — search for text across files (case-insensitive)

Journal

• journal_add — add a timestamped entry to today's journal
• journal_read — read journal entries for a day or date range
• journal_search — search across all journal entries

Scheduling

• schedule — schedule a future nudge to prompt the agent at a given time
• cancel_task — cancel a scheduled task
• list_scheduled — list all scheduled and running tasks

Configuration

Variable	Default	Description
DOCKETEER_ANTHROPIC_API_KEY	(required)	Anthropic API key
DOCKETEER_CLAUDE_MODEL	claude-opus-4-6	Claude model to use
DOCKETEER_DATA_DIR	~/.docketeer	Where the agent stores memory and audit logs
DOCKETEER_DOCKET_URL	redis://localhost:6379/0	Redis connection for task scheduling
DOCKETEER_DOCKET_NAME	docketeer	Name of the Docket instance
DOCKETEER_REVERIE_INTERVAL	PT30M (30 min)	Background thinking cycle interval
DOCKETEER_CONSOLIDATION_CRON	0 3 * * *	Cron schedule for daily memory consolidation

Plugins

Docketeer discovers plugins through two entry point groups:

• docketeer.chat — chat backends (how the agent talks to people)
• docketeer.tools — tool plugins (what the agent can do)

Available plugins:

• docketeer-rocketchat — Rocket Chat backend
• docketeer-web — web search, HTTP requests, file downloads
• docketeer-monty — sandboxed Python execution