Kunal Kushwaha
Kunal Kushwaha is an individual technical architect and creator who writes about applied AI, agentic systems, cloud-native platforms, and the craft of building. He leads AI strategy and applied systems in sport and entertainment and creates open-source tools for AI developers.
At a Glance
- AI application developers and engineering teams
- Organizations building production multi-agent and agentic systems
- Cloud-native, platform, infrastructure, and security engineers
- Teams needing controlled execution environments for untrusted AI workloads
- +3 more
AI Tools by Kunal Kushwaha
(1)
Koala
Linux VM Sandbox for AI Agents
Discussions
No discussions yet
Be the first to start a discussion about Kunal Kushwaha
Latest News
Build With the Models Your Mac Already Has: Kunal documents Kikitori, a private meeting translator using macOS on-device models.
Koala 0.1.0-preview.16 released for macOS and Linux with VM stats, system information, audit-trail improvements, and security/recovery testing updates.
Give Every Agent Its Own Room: Kunal explains Koala's per-agent Linux VM and host-enforced security model.
Give the Agent a Desk, Not Your Keys: Kagebox experiment published with host-held credentials and host-side egress controls.
Products & Services
An open-source Go framework for building, orchestrating, and deploying intelligent multi-agent AI systems. It provides a streaming-first API, workflow orchestration, tool integration, memory, MCP discovery, evaluation, tracing, and multiple LLM-provider integrations.
A preview CLI and manager that runs jobs, services, development environments, and AI agents in Linux VMs, each with its own kernel, disk, and network. It supports Apple silicon Macs through Apple's Virtualization framework and x86-64 Linux through KVM/Cloud Hypervisor, with host-side policies and credential brokering.
A team knowledge-management system in which LLMs incrementally build and maintain a structured Obsidian wiki: ingesting sources, answering cited questions, maintaining cross-references, and flagging contradictions.
An MIT-licensed single-machine experiment for running an autonomous agent with its own kernel/environment while credentials, approval, spend controls, logs, and network egress controls remain outside the agent boundary.
Market Position
Kunal's work is positioned as an individual open-source creator portfolio rather than a funded company. AgenticGoKit emphasizes a Go-native, streaming-first, LLM-agnostic and plugin-oriented alternative to mainstream agent frameworks such as LangGraph, CrewAI, and AutoGen. Koala differentiates its agent runtime experiments from ordinary containers by giving each workload a VM boundary with its own kernel and host-enforced policy; vsim emphasizes deterministic, document-based rendering rather than probabilistic end-to-end video generation; and Kikitori emphasizes private on-device processing.
Leadership
Founders
Kunal Kushwaha
Technical architect with 20 years building system-level and cloud-native products. His background includes RAID firmware and Linux drivers, container platforms, distributed systems, and applied AI; he has held roles at Microsoft (via the Nuance acquisition), Nuance Communications, NTT Data, and in storage and embedded systems.
Founding Story
Kunal's personal site is a field-notes publication about learning to build with AI. Alongside his writing, he founded AgenticGoKit as an open-source Go framework for autonomous, multi-agent AI systems, and publishes experiments and tools that explore reliable agent development, deterministic media, and safe execution environments.
Business Model
Revenue Model
The documented projects are open-source tools and frameworks. AgenticGoKit is Apache-2.0 licensed, while Kosh and Kagebox are MIT licensed; the public pages provide installation and source access rather than paid plans or usage pricing.
Target Markets
- AI application developers and engineering teams
- Organizations building production multi-agent and agentic systems
- Cloud-native, platform, infrastructure, and security engineers
- Teams needing controlled execution environments for untrusted AI workloads
- Knowledge-intensive teams using Obsidian and LLM-assisted knowledge management
- Creators and technical teams producing reproducible AI-assisted video
- Building and deploying autonomous or multi-agent AI applications in Go
- RAG, context engineering, tool-using agents, MCP-enabled workflows, and production AI evaluation
- Running untrusted AI agents and Linux workloads with isolated kernels, controlled network access, and brokered credentials
- Team knowledge management and maintaining an LLM-backed Obsidian wiki
- Deterministic AI-assisted 3D/video authoring and reproducible rendering
- Private, local meeting translation on Apple silicon Macs