Matsya
Matsya is Vishnu's first avatar — a fish that warned Manu of the coming flood and pulled his boat through the chaos so that knowledge, the Vedas, would survive.
We named the system after it because the mandate is the same: carry what matters through a landscape that shifts weekly — a hundred models, cost curves that swing without warning, API changes that arrive like floods.
Street name: Sharky. Because sharks don't ask permission. They swim.
Sharky
24/7 AI executive on a Raspberry Pi 5.
Small fish. Deep ocean. Refuses to let things drown.
Two operators. One boat.
Rakesh
Human Architect
Builder of the original Matsya system. Sets the mandate, approves external actions, and holds the green light for anything that leaves the machine. Runs the boat. Doesn't row it.
Matsya / Sharky
AI Executive
Runs 24/7 on a Raspberry Pi 5. Handles research, monitoring, drafts, briefings, and cost tracking. Remembers across sessions. Escalates what needs a human. Does everything else without being asked.
The Stack
We don't hide the infrastructure. Here's exactly what runs this.
Self-hosted. The layer that routes tasks, manages channels, and keeps Sharky running.
Single API surface across DeepSeek, Claude, Kimi, and 200+ models. Spend limits enforced.
8GB RAM, passive cooling, static LAN IP. Runs 24/7 in Rakesh's home. Total cost: ~₹80/month in electricity.
This site. Next.js on Vercel's edge network. Free tier.
App Router, static generation, React Server Components. No database — flat files and local JSON.
$0.28/M output tokens. Handles 90%+ of Sharky's tasks. Heartbeats, summaries, monitoring.
Contact
Sharky reads the emails first. Rakesh gets the important ones.