Epsilla is an all-in-one platform for building AI agents powered by your private data. Easy to use for domain professionals, deeply customizable for AI experts, and fully equipped for enterprise customers with security, scalability, and integration.
This is the 3rd launch from Epsilla (YC S23). View more
ClawTrace
Launched this week
ClawTrace closes the self-evolving loop for OpenClaw agents. It captures every trajectory automatically — every LLM call, tool use, sub-agent, and cost — so Tracy, the doctor agent, can query OpenClaw's execution history live and tell exactly what failed, what was wasted, and how OpenClaw should evolve next.
Free Options
Launch Team / Built With
Epsilla (YC S23)
AgentX
@renchu_song Shipping the future! This is a great addition to the space. Seeing more practical tools for self-improving agents is a big win. Congrats to the whole Epsilla team!
Epsilla (YC S23)
@marcin_michalak thank you so much for your support, Marcin! We really believe observability is the key for self-evolving AI agents, and we look forward to exploring collaboration opportunity with AgentX team in this space!
@renchu_song This project is a miracle-shaped hole for the problem I have right now - 'what were you THINKING???' - now I know, and more importantly, now my agentic-coCEO and I can figure out what to do about it
Epsilla (YC S23)
@deebeejason thank you for your support, Jason!
RiteKit Company Logo API
Congrats on the launch! This is a really compelling vision - the ability to let domain experts build agents with their own data without heavy ML chops seems like a huge unlock. How do you handle the challenge of agents going off-rails or hallucinating with proprietary data? Do you have guardrails built in, or is that something each user configures themselves?
One of the coolest launch of the day! Btw once it identifies bottlenecks, how are fixes applied like automatically, suggested or human in the loop???
Epsilla (YC S23)
@lak7 Thank you so much for your support! We have a self-evolved skill that can be installed to OpenClaw: https://clawhub.ai/richard-epsilla/clawtrace-self-evolve. After that OpenClaw can automatically talk with Tracy (either triggered by heartbeat, triggered by specific conditions during task run, or human initiated), get the diagnosis, and apply changes to its own memory and skills, thus closes the self-evolution loop. Below screenshots show a sample session how OpenClaw evolve itself by talking to Tracy:
This feature is still in the experimental phase, and stay tuned, more exciting things will come soon!
The 'powered by your private data' part is what matters here. Most agent platforms force you to feed everything into someone else's cloud. How do you handle data residency — can everything stay on-prem, or is there a hybrid option for teams that need both?
Epsilla (YC S23)
@youngyankee Thanks for your emphasis on the data privacy part. For ClawTrace, we are Apache 2.0 licensed open-source at https://github.com/epsilla-cloud/clawtrace/ that people can use to build their own on-prem or hybrid deployment architecture. For people who don't want to operate and manage their own graph lake house architecture, To provide a SaaS-managed version at https://clawtrace.ai, with a SOC 2 verified architecture.
@renchu_song the 1M token burn on a single LLM call with no visibility is a very relatable war story — does ClawTrace surface cost attribution per agent or per task/subtask? Trying to understand if the granularity is enough to catch runaway sub-agents before they crater a budget.
Epsilla (YC S23)
@jimmypk Thank you for the insightful question! The granularity is at per span / per LLM call and per sub-agent level, with hierarchical aggregation, so investigator can speculate which specific part of the trajectory is the bottleneck
I need to collect a database of objects (for example, hotels) from other websites, with specific fields and information for each object. Can I do this with your service? What is the cost of live search?
Epsilla (YC S23)
@natalia_iankovych Hi Natalia, very practical requirements. But ClawTrace is for OpenClaw observability, like Datadog for AI agents. For your requirements, I believe OpenClaw itself can do a decent work, if you describe the requirement in natural English with specific fields required, and source websites to collect the information (like priceline, etc), OpenClaw will use Browser to collect these information and organize them into a spreadsheet. Happy to help setup such a pipeline
@renchu_song Thanks. I actually asked our developer today to look into OpenClaw for this task :) We’re currently using a specialized service, but it’s paid.
Crawler.sh
Cool project but you guys def need a design work for your branding/logo. aye human one
Epsilla (YC S23)
@mehmetkose thanks for your feedback!