Definetely the scenario people(vibecoder) are in...therefore made vibegaurd @ @polaris_code x @truefoundry 's agent that act hacakthon
Demo on the way
#agentsthatact
Admin was caught in the act at @polaris_code × @truefoundry 's hackathon leave
Sleep is totally visible 👀 shoot your shot's on how terrible he is trying to act out there🤣
#agentsthatact
Build Agents That Act is live at Polaris Campus, Bangalore.
900+ Round 1 registrations, one day, every agent on TrueForge. With @polariscodes.
The bar: reach a real system, run it live in front of the judges, stop before anything irreversible.
Updates below 👇 #agentsthatact
side quest started.
4 of us, one hackathon, and absolutely no idea how this day is going to end.
At @polariscodes for @truefoundry’s Agents That Act.
Got some good perspective from Anant Agarwal this morning. Now we build.
#agentsthatact#buildinpublic
@truefoundry@polaris_code Team rocket phir se haar gyi🥀🥀
Learnt a lot....met a lot of good folks.. stings that we were not on the stage
But a day well spent
Admin gonna go on hibernation now🥹
4 hours. Countless iterations.
Built OpenSwarm Autopilot—a planning layer that decides what should happen before the multi-agent system starts executing.
Demo day. @polaris_code@openswarm#OpenSwarm#Hackathon#AI
5 days of silence.
here's what was actually happening:
backend: locked. 14 tables. 40+ routes. $0/month infra. now we're in the part nobody talks about —
making the UI not suck.
turns out "paste conversation → get vibe" sounds simple until you're debugging hydration mismatches at 2am because className="dark" exists on the server but not the client and React is having an existential crisis about it.
suppressHydrationWarning is now our most used prop.
UI is close. GhostMode is getting a face soon.
FourDevs — building in public
@shipdailyx the sha256 is just the start next step is semantic deduplication. two slightly different pastes of the same conversation shouldn't both hit the API either. but that's a v2 problem.
ISSUE 9 [SOLVED]
"same text pasted twice = two Claude API calls. two charges."
no deduplication. every submit hits the AI.
the fix:
sha256(text.trim().toLowerCase()) → stored as text_hash. before calling AI, check: same user + same hash + last 24 hours?
if match → return cached result instantly. { ...existing, cached: true }
zero API cost. zero latency. user doesn't notice.
GhostMode — building in public
10 issues found. 10 issues killed.
2 day sprint. backend is clean.
what we shipped: → $0 AI infrastructure (was going to be $3,750/month)
→ auth that actually works
→ WebSocket security fix
→ dark mode without the retina burn
→ deduplication saving API credits
→ abuse protection for free tier
→ email reliability
→ feed performance
backend is locked. no more touching it.
next: the UI.
GhostMode is about to have a face.
ISSUE 10 [SOLVED]
"what happens if 100k free users hit the API on the same day?"
100k × 5 analyses = 500,000 AI calls. even on free APIs, rate limits would explode.
the fix:
Redis counter. keyed by date. incremented per free analysis.
platform:free:daily:2026-05-28 → increments to 10,000 → stops.
over cap? 503: service_busy. upgrade to premium.
worst case day: $0. always.
GhostMode — building in public