PAI as a full autonomous platform — how far should the repo go?

[MarvinDontPanic]

Picking up from #542 (PAI vs OpenClaw comparison) — I've built a few pieces on top of PAI that close most of the gap with OpenClaw, and I'm curious whether this direction is something the project wants to pursue.

The core three (what makes it autonomous)

1. Job execution gateway — A small service that accepts work via API, manages claude -p runs, handles concurrency/timeouts/callbacks, and streams progress to a dashboard. ~500 lines.

2. Scheduling system — Fetches task definitions from an API, executes them on schedule. LLM tasks go through the gateway, bash tasks run directly. ~300 lines.

3. Messaging integration — Bidirectional bot so the AI can reach you outside the terminal (job results, alerts) and you can reach it from your phone. I built mine on Slack, but the interface is simple enough that Telegram, Discord, or any other client could implement it.

Optional but useful

• Control center dashboard — Web UI for task management, job history, system health
• Email handling — Automated fetch, rules, AI triage as a scheduled task
• Document library — Delivery system for AI-generated artifacts
• CLI tools — Thin wrappers for common operations

The point

The Algorithm, memory, skills, hooks — that's 80% of the work and it's already done. The three core pieces above aren't architecturally complex, and with them in place you don't need a separate system like OpenClaw.

I have all of this built and running for myself, and it's been working well — it's replaced OpenClaw completely for me. So I know the approach is viable. The question is whether it's useful to the broader community.

Questions

1. Does PAI see itself expanding into infrastructure like this, or is the scope intentionally the cognitive layer only?
2. If this direction is interesting, what are useful next steps? I could start contributing architectural patterns, reference implementations, or submit PRs — what would be most valuable?

[jschulman]

I did the same thing as you but... OpenClaw has a lot of bloat. it's trying to be everything to everybody. That's great for them, but I'm not sure that's what we're building here. So my suggestion would be an integration framework instead of building connectors. You want to talk to Slack/Discord/etc? It doesn't matter, these are the end points you use. Same for email, web, etc. It keeps PAI smaller and each user can build their version.

[danielmiessler]

Yes, this is the way we're thinking about it as well. We're trying to do things a bit slower in the PAI way.