Back to Projects
Autonomous AI agent daemon with persistent memory and proactive behavior.
Problem
Traditional AI workflows lose context between sessions, require manual prompting, and can't act proactively on schedules or events.
Approach
Built a TypeScript daemon using Claude CLI as runtime with a three-tier memory system, lane-based task queue, Discord transport, and Google Workspace integration. Runs as a systemd service with self-healing diagnostics.
Impact
186+ source files, 250+ tests, autonomous daily pipelines for social media, email monitoring, and proactive task execution.
What I Built
- Architected a three-tier memory system (working, episodic, semantic) backed by SQLite for persistent context across conversation sessions.
- Built a lane-based task queue with priority scheduling, dependency resolution, and automatic retry with exponential backoff.
- Integrated Discord.js as the primary transport layer for bidirectional human-agent communication with rich embeds and reactions.
- Implemented Google Workspace integration for autonomous email monitoring, calendar management, and document creation.
- Deployed as a systemd service with self-healing diagnostics, automatic restart on failure, and structured logging for observability.
Technologies
TypeScriptClaude CLIDiscord.jsGoogle APIsSQLiteStagehand