Hi HN,<p>We are building pica/os, an infrastructure layer for long-running autonomous AI agents. The core problem we're solving: while there are many agent frameworks, most can't maintain persistent state or run continuously for weeks while handling real-world API integrations.<p>Technical Architecture:
- Rust-based runtime with persistent state management for long-running operations
- Custom scheduler designed for weeks/months of continuous agent operation
- Unified API gateway managing rate limits and authentication across 50+ platforms <- (opensource)
- Memory management system for extending beyond standard context windows
- Built-in approval system for sensitive operations<p>Implementation Approach:
- State persistence through event sourcing
- Distributed task queue for managing long-running operations
- OAuth token management and automatic refresh
- Configurable human-in-the-loop checkpoints
- Automatic retry and fallback mechanisms for API failures<p>Example Use Case We're Building For:
A customer support agent that:
- Monitors tickets across multiple platforms (Zendesk, Email, Slack, etc)
- Maintains conversation context across days/weeks
- Automatically follows up on unresolved issues
- Escalates complex cases to humans<p>Current Status:
- In final development stages
- Planning private beta early 2025
- Focused on developer experience and API design
- Actively working on documentation and SDK<p>Key Questions:
1. For those building autonomous agents, what are your biggest infrastructure pain points?
2. What would make you choose this over existing solutions?
3. Which specific use cases would you want to try first?
4. What security/safety features would you need before considering deployment?
5. What observability features would you need for production deployment?<p>I'm happy to share more technical details about our architecture or discuss specific implementation challenges we're tackling.