1) iMessage Support. iMessage is now a supported integration for interacting with Praxos. You can simply add it from the integrations menu.
2) Integration magic links. AI can now send you magic links to integrate an app without having to manually provide credentials for an integration. Click the link and an oauth window will open.
3) Audio, Video, Image generation support. We have always supported multimedia and file input (despite some issues xD). You can now directly ask for generation of any of the above.
4) New tools for Notion and Gmail, allowing much more extensive 'find' operations as well as creating databases and pages much more freely.
Under the hood,
Below is a (non)comprehensive summary of Praxos's updated capabilities:
–Email management. Find, summarize, respond to, or draft emails. Try asking it to "find emails about X" or "draft a reply to Y." It'll figure out the remaining context on its own and draft it in your voice.
–Answer questions from your data. Ask me things like "what's the tracking number for my package?" It'll comb through the collective sum of all the data sources it's linked to.
–General knowledge: Ask anything from "how tall is the Eiffel Tower?" to "what's the weather in Toronto?"
–Set up automations. It can automatically create schedules and keep them in memory. This means that it'll forward emails, notify you about important messages, and handle repetitive tasks for you.
–Calendar management: Create, update, or delete events and get reminders.
–Research. Look things up for you online, from restaurant recommendations to market research.
All of these are available over both text and voice across iMessage, Whatsapp, Telegram, and Discord.
We are working towards creating an assistant that knows you shockingly well in your pocket, with full awareness of:
-What you’re working, and with what deadlines
-How you slept every morning via whoop & oura and keeps you accountable with your workout / health goals.
-What you like to eat based on your restaurant reservations, takeout receipts, and grocery purchases.
–What you post, like on socials, and newsletters you read, so that it can curate content for you and help you get to the nuggets of information that you really care about faster.
-Who your family, friends, and coworkers are, so that it proactively proposes and schedules activities for you and them..
Only hypothetically structured well. In truth, basically every carrier structures their data just a bit differently, and then on top of it, few insurance packages in commercial insurance policies from one carrier. and then you have to deal with appraisals, tech specs, and all the other external data that is basically randomly structured.
As well as pesky websites with either zero or crappy API support. You end up biting the bullet and entering the data by hand or taking a screenshot and hoping OCR does the job.
We're Lucas and Soheil, and we are the co-founders of Praxos.
Praxos is a kernel for AI agents, combining rock-solid stability with LLM flexibility.
The Problem: Building AI agents that reliably handle complex, stateful tasks is incredibly hard. Current approaches are often brittle, leading to unpredictable behavior and constant fires that have to be put down one by one.
Our Origin: We faced this when building AI for Insurance. LLMs struggled with the intricate, interconnected data, which as a bonus also wasn’t limited to a single domain. This made building a robust system very difficult, and connecting the different agents and workflows basically impossible. We knew there had to be a better way to combine AI's intelligence with system-level dependability.
How Praxos Works: We are building the full Praxos Kernel brick by brick, and we started with the most critical challenge: solving the data problem for AI, once and for all. Today, this foundational layer of Praxos is live and available through our SDK.
As of now, we can parse any data source, from unstructured PDFs and API streams to conversational messages to structured databases—and automatically transform them into a single Knowledge Graph with deep semantic types (e.g., 'PolicyID', 'CountryName') that allow you to ensure that you’re retrieving and updating exactly the type of info you want, while also supporting vector search. It also makes extracting structured information super easy, because all you have to do is look for the right types, no re-extraction necessary. Need new types on the fly? Not a problem. Simply create your own types, or extend ours through our pydantic compatible SDK.
Next Steps:
We have laid out our vision and how we want to go about building this in our white paper, linked in the post. That said, the TLDR is that we are aware that a kernel is not just about managing memory. Our next goals will be to incorporate lifecycle management, syscall boundaries, and finally, what we are the most excited about: breaking down LLM tasks into simpler, repeatable instructions. Let us know if you think there’s a different step we should tackle first!
Try it Out & Share Feedback:
Website: www.praxos.ai
1) iMessage Support. iMessage is now a supported integration for interacting with Praxos. You can simply add it from the integrations menu.
2) Integration magic links. AI can now send you magic links to integrate an app without having to manually provide credentials for an integration. Click the link and an oauth window will open.
3) Audio, Video, Image generation support. We have always supported multimedia and file input (despite some issues xD). You can now directly ask for generation of any of the above.
4) New tools for Notion and Gmail, allowing much more extensive 'find' operations as well as creating databases and pages much more freely.
Under the hood,
Below is a (non)comprehensive summary of Praxos's updated capabilities:
–Email management. Find, summarize, respond to, or draft emails. Try asking it to "find emails about X" or "draft a reply to Y." It'll figure out the remaining context on its own and draft it in your voice.
–Answer questions from your data. Ask me things like "what's the tracking number for my package?" It'll comb through the collective sum of all the data sources it's linked to.
–General knowledge: Ask anything from "how tall is the Eiffel Tower?" to "what's the weather in Toronto?"
–Set up automations. It can automatically create schedules and keep them in memory. This means that it'll forward emails, notify you about important messages, and handle repetitive tasks for you.
–Calendar management: Create, update, or delete events and get reminders.
–Research. Look things up for you online, from restaurant recommendations to market research.
All of these are available over both text and voice across iMessage, Whatsapp, Telegram, and Discord.
We are working towards creating an assistant that knows you shockingly well in your pocket, with full awareness of:
-What you’re working, and with what deadlines
-How you slept every morning via whoop & oura and keeps you accountable with your workout / health goals.
-What you like to eat based on your restaurant reservations, takeout receipts, and grocery purchases.
–What you post, like on socials, and newsletters you read, so that it can curate content for you and help you get to the nuggets of information that you really care about faster.
-Who your family, friends, and coworkers are, so that it proactively proposes and schedules activities for you and them..