Sudhi Seshachala
AI enthusiast. Failed founder. Open-source builder. Houston, TX.
The backstory
I built Xervmon — a cloud management and monitoring platform — and spent years trying to turn it into a business. It didn't work. Not for lack of effort or conviction, but because the timing was off, the market moved differently than I expected, and building a company is genuinely hard in ways that aren't obvious from the outside.
I learned more from that failure than from anything else I've done. About what engineers actually need. About where tools break down at scale. About the gap between what sounds good in a pitch and what actually gets used at 2am when something is on fire.
The return
AI changed everything — not as a buzzword, but as a genuine shift in what a single developer can build and ship. I got obsessed. I started using Claude Code, building agentic workflows, wiring up MCP servers, and watching my token bills climb.
And I noticed the same pattern that killed Xervmon: powerful tools with rough edges. Context replay burning thousands of dollars a day. Models routing to Opus when Haiku would do. Files read in full when three functions would suffice.
This time I'm not building a business around the fix. I'm just shipping it.
The philosophy
Every idea I have, every tool I build, every pattern I discover — it goes on GitHub. MIT licensed. No freemium gates. No “contact us for enterprise pricing.”
If it's useful, fork it. Improve it. Use it at 2am when something is on fire. That's what I wanted from Xervmon. That's what I'm building now.
The tools I'm shipping — Agent Booster, RTK, Conduct — are the infrastructure I wish had existed when I was running Xervmon. Context efficiency, smart model routing, governed agent workflows that a team can actually own and modify as YAML. Small, composable, open.
What I'm building
Agent Booster
AST-level context routing for AI coding tools. smart_read, semantic search, smart model routing. Cut Claude Code costs 3–15x.
RTK — Rust Token Killer
A transparent CLI proxy that strips token noise from git, build, test, and package manager output before it reaches the model. 93% savings in production.
Conduct
22 ready-made agent playbooks for tickets, PRs, alerts, and incidents. DORA-lite metrics, per-agent A–F scorecards, vector memory across runs, and ConductGuard policy enforcement — all in one platform. Install a playbook, connect an environment, run.
Find me
If you're building with AI agents and want to compare notes, I'm always up for it.