We're working to improve reusability of the tools and software libraries you've already created across projects This should allow you to flesh out your own software "ecosystem" that grows more as you develop with PRSM

Chilling on da computer tweaking system prompts for the last 12 hours ama

We are going to be meeting up and working all weekend, look forward to a few streams

On one of our 7 computers and not including any codex

What do you guys think of a q&a stream tomorrow night post beta launch? We're working very hard right now to make sure everything is polished for the beta, and we appreciate the continued support

Prompt compiler lookin good

Used a couple more tokens 😅

Closed Beta Launch and Q&A stream tn at 10PM EST

We are live to announce our closed beta launch and a Q&A session x.com/i/broadcasts/1…

A piece of feedback we've received from our beta testers is that they would like an agent-friendly frontend so they can plug their existing agentic setups directly into prsm and provide them full visibility into the system. We are working on a non-gui workflow that allows you to

It’s really wonderful when you watch the system literally iterate and self-improve. It reviews initiative segmentation, and then optimizes its own system prompts to make sure that the components are designed in such a way that they can be easily reused by future projects.

Little UI update (Feedback appreciated)

Iterations and solidifying the orchestration flows is getting expensive 🙃 85%+ cache hit rate in testing though

It's truly impressive how much Codex refuses to abide by agentic workflows. We have implemented a provider-specific prompt injection layer to tune behavior for each model for each state. This means, that different models with different behaviors can be tuned to be Hypervisors,

We are implementing a dynamic tool hinting system that tells agents about relevant identifiers that they can pick from, rather than just reminding them of the correct tool call schema like most orchestration tools. We are finding that this methodology makes tool calls more

We have received feedback from our beta testers that they want to be able to customize/tune some of the per state prompts, so we are creating an interface for them to do that. Down the line, we will implement a FSM compiler and plug-in interface for you to build full custom