I build guardrails for agentic AI: the layer that stops an agent from running up a bill, breaking an integration, or losing track of its own work. I'm based in Dhaka and also run AgencyHandy.
| noburn | Hard spending caps for LLM APIs. A pre-flight check runs before every call, so an over-budget agent call is never sent and never billed. Caps per project, per end user and per run. npm i @noburn/sdk · pip install noburn |
| APIDiffGuard | Catches breaking API changes before your clients (or your agents' tools) do. It monitors responses, detects JSON schema drift and diffs versions. Open core: orvi2014/apidiffguard |
| AgencyHandy | Client portal for agencies: services, projects, files, invoices and payments in one place. |
- Baar-Core: budget-aware routing for LLM agents. It routes between cheap and strong models and enforces a hard budget cap locally, before the request leaves your machine.
pip install baar-core - hatswap: run several Claude Code accounts side by side, move a live conversation between them, and hand work to Codex.
- classcatch: a Chrome extension that shows and copies Tailwind classes from any element, and converts plain CSS to Tailwind.
- Multi-agent coordination: the patterns that work and the ones that blow up
- Hallucination in production agents: the real risk profile
- The economics of AI agents: when autonomous beats human-in-the-loop
- How semantic routing cut my LLM costs by 70%
More essays (agents, football, psychology) at robatdasorvi.com.
Shipping agents and worried about runaway spend or upstream API changes? Try noburn or APIDiffGuard, or open an issue on any repo. I read them all.