Agent Learning and Data Services DigitalOcean's Agent Learning Services flag silent production failures and user frustration in AI agents that never appear in logs, converting those sessions into one-click evaluations and simulations that replay fixes on the real harness before shipping. Agent Data Services ingests call recordings, documents, scanned files, and agent logs into a single semantically searchable catalog so agents can retrieve context by meaning without custom pipelines. Both run on isolated per-session microVMs with Action Gateway providing governed access to 20,000+ tools.
Visit productFeatured media
Loading media…
View on XCosigns Received
“The bug I keep shipping looks like a success. The recall matcher on kept returns a model number and sounds sure. It is the wrong SKU. Nothing in the log says it failed. I find out when I check the furnace label myself. A replay of that one session would be worth more to me than another benchmark.”
“wired this into my stack today... my evals write themselves NOW: - users rage quit silently -> Signals catches that exact session - evals written from imagination -> one click turns the real failure into one - fixes that break something else -> Simulations replays it on your actual harness first - data buried in calls, pdfs and logs -> Agent Data Services makes it one searchable catalog - agents holding your api keys -> Action Gateway gives them 20,000+ tools and zero keys the move: stop writing evals from imagination your angriest user already wrote them for you Signals and Simulations are in public preview right now P.S. thanks DigitalOcean for providing an access 🙏”
“this looks like a game changer especially for solo founders. I'm the whole team, so a bug nobody sees just sits there, your customer will go to your competitor, and you won't even have an idea what's going on.”
“the worst bug in AI right now is the silent 200 OK tool call fails. agent says sorry and makes something up. user leaves. logs say 0 errors you’d never even know it happened signals flags those sessions and turns them into evals. simulations replays your fix on them before you ship. both in public preview this is the stuff i actually want”
“the Simulations part is what I'd want to try first. you take a chat that went wrong in prod and run your fix against it before shipping. way better than me writing fake test chats by hand. it's in public preview, and works with Claude Code, Codex and OpenCode agents for now.”
“LESS time hunting through logs, MORE time fixing what frustrated the user 🙌 I love the direction :)”