Autonomous agent infrastructure with scheduled wakeups

[jakemannix]

Summary

Adds infrastructure for autonomous agent operation, enabling agents to wake up periodically and perform tasks without human intervention.

New Features

Scheduled Agent Wakeups (src/modaletta/scheduled/wakeup.py)

• Modal cron job that wakes agents every 15 minutes
• Agent roster management (stored in Modal volume)
• Custom wakeup prompts per agent
• Wakeup logging to /data/logs/<agent-id>.jsonl

Digest Cron Job (src/modaletta/digest_cron.py)

• Fetches Bluesky feeds and delivers summaries to agents
• Configurable via digest_config.yaml

Documentation

• CLAUDE.md - Development guidance for AI coding assistants
• docs/design-autonomous-infrastructure.md - Architecture for scheduled wakeups, persistent volumes, bidirectional Modal↔Letta integration
• docs/design-mcp-tools.md - Design for MCP tool servers on Modal

Architecture

Modal Platform
├── Scheduled Wakeups (Cron)
├── Digest Delivery (Cron)
├── Persistent Volumes (/data, /artifacts, /cache)
└── MCP Tool Servers (future)
        ↓
    Letta Cloud
    └── Agent Memory ↔ LLM Reasoning

Test Plan

Unit Tests

uv sync --all-extras
uv run pytest tests/ -v

Scheduled Wakeup Testing

# Create Modal secret with Letta credentials
modal secret create letta-credentials \
  LETTA_SERVER_URL="https://api.letta.com/" \
  LETTA_API_KEY="<your-key>"

# Initialize agent roster
modal run src/modaletta/scheduled/wakeup.py --init --agent-id <agent-id>

# Test one-time wakeup
modal run src/modaletta/scheduled/wakeup.py --agent-id <agent-id>

# Test with custom prompt
modal run src/modaletta/scheduled/wakeup.py --agent-id <agent-id> --prompt "Check for updates"

# View logs
modal run src/modaletta/scheduled/wakeup.py --logs --agent-id <agent-id>

# Deploy (runs every 15 min)
modal deploy src/modaletta/scheduled/wakeup.py

Digest Cron Testing

See src/modaletta/TESTING_DIGEST_CRON.md for detailed instructions.

File Changes

• 10 files changed
• +1,708 lines

👾 Generated with Letta Code