$ tail -f studio/live · 5 systems running ▍
Not a portfolio page. These are the tools we assemble from enterprise-grade open source, and they're running right here on this page. Watch them work, then imagine them on your stack. Your cloud, your data, no per-seat pricing.
01 · Observability
See the outage before your customers do.
OpenTelemetry wired through your services, landing in Grafana. Golden signals per service, alerts that page before the support inbox fills. The demo is the real thing: watch the p95 spike, the alert fire, the on-call get paged.
p95 latency
340ms
throughput
214 rps
error rate
0.10%
14:12:12 info checkout-api 200 GET /cart 41ms
14:12:15 info payments 200 POST /charge 118ms
14:12:18 info catalog 200 GET /sku/8812 12ms
14:12:21 warn payments retry gateway=razorpay attempt=2
14:12:24 info auth 200 POST /token 9ms
02 · Security & access
Fortune-500 security posture, zero license invoice.
One login across every internal tool, secrets out of Slack, vulnerability scans in CI. The console below runs the same loop we hand over: Keycloak granting and denying access, Trivy tearing through a repo, OpenBao rotating a key.
Access events · keycloak
14:12:12 priya@acme.in → metabase granted
14:12:19 dev-ci@acme.in → registry granted
14:12:26 rahul@acme.in → espocrm mfa challenge
14:12:33 unknown (49.36.x.x) → grafana DENIED
14:12:40 meera@acme.in → n8n granted
14:12:47 arjun@acme.in → warehouse granted
CI scan · trivy · acme/checkout-api
scanning layers… 0%
HIGHopenssl 3.0.2CVE-2026-0778→ 3.0.16
MED lodash 4.17.20CVE-2025-8203→ 4.17.23
MED pillow 9.4.0CVE-2025-4863→ 10.3.1
openbao · secret payments/razorpay-key rotated ✓ 14:12:12
03 · AI platform
Every model behind one gateway. Every rupee accounted for.
One API in front of every provider, one key per team, full tracing of prompts, cost and quality. Watch requests route across Claude, GPT and Llama, a guardrail catch PII, and spend attribute itself to the team that caused it.
requests this month 18,400
spend $1246.20
Request ledger · litellm + langfuse
supportclaude-sonnet-52.1k$0.011640ms
growthclaude-haiku-4.50.8k$0.002210ms
opsllama-3.3-70b3.4k$0.004890ms
supportclaude-sonnet-51.7k$0.009570ms
financegpt-5.21.2k$0.014730ms
growthclaude-sonnet-5⛔ guardrail: PII in prompt
opsclaude-haiku-4.50.6k$0.001190ms
Spend by team
eval score (7d) 96.2% ✓
04 · Data platform
Scattered SaaS exports in. One queryable warehouse out.
The substrate under most of our client builds: every source synced into one warehouse, modelled once, tested on every run, served to operators through BI. The pipeline below runs end to end on a loop: sources to dashboard in seconds.
sources
shopify
tally erp
espocrm
sync
airbyte
warehouse
clickhouse
models
dbt · 84
bi
metabase
syncing sources… ▍
Revenue by channel · refreshed 47s ago
05 · Automation
Workflows that run themselves. Humans where it matters.
A hardened n8n deployment plus the automation library we rebuild for every ops team: routing, enrichment, scoring, approvals, follow-ups. This run holds at the human-approval step on purpose. That's the point.
workflow lead-fastlane · runs 1,284
run in progress
waiting for trigger…
Shipped for clients
10 builds
We run what we ship
Our website, publishing, CRM and growth engines are all self-hosted open source. If we wouldn't operate it ourselves, we don't hand it to you.
$ 10dem infra ls
site react · lambda · cloudfront a few $/mo
blog ghost · self-hosted ec2
crm + ops espocrm + n8n ec2
growth instagram + x engines in-house agents
4 systems · 0 license fees · 0 lock-in ▍
Bring us the mess. We'll tear it down, pick the right stack, and ship something your team owns in weeks, not quarters.