# Strukto

> Strukto builds foundational infrastructure for AI agents: Mirage, a virtual terminal that mounts storage, tools and runtimes as one agent-native workspace behind a single shell.

## About
Strukto is a B2B AI infrastructure company building foundational infrastructure for AI agents. Its work centers on Mirage, a virtual terminal: a virtual filesystem for context, virtual CLIs for tool use, and routed runtimes for where a command lands, with per-agent profiles governing all three. The same agent reads a Google Doc, queries Postgres, and ships a Linear issue with the same handful of primitives. Founded by Zecheng Zhang.

## Projects
- [Mirage](https://docs.mirage.strukto.ai): a virtual terminal for AI agents, and a virtualization layer between the agent and the infrastructure underneath it. A virtual filesystem mounts storage and services as one tree; virtual CLIs answer git, gh, slack, discord, linear and ntn itself, so an agent drives a service with nothing installed; runtimes route where a command actually runs, from Monty and Pyodide in process to Docker, E2B or an SSH host; a policy engine allows, asks or denies each command and hides paths outright; cloud authentication resolves a credential into the session; notifications turn an external change into an event on the mount; and per-agent profiles govern all of it. Portable enough to run inside FastAPI or Express, in a browser tab, behind the mirage CLI, or as a real mountpoint over FUSE and FSKit. Python and TypeScript SDKs and a CLI. Drops into OpenAI Agents SDK, Claude Agent SDK, Vercel AI SDK, LangChain, Pydantic AI, CAMEL, Agno, Mastra, Haystack and OpenHands. Available now. Live demo at https://www.strukto.ai/terminal. Docs at https://docs.mirage.strukto.ai. Source at https://github.com/strukto-ai/mirage.

## Research
- [SEAR: Schema-Based Evaluation and Routing for LLM Gateways](https://www.strukto.ai/research/sear). Accepted at [CAIS 2026](https://www.caisconf.org/program/2026/papers/sear-schema-based-evaluation-and-routing-for-llm-gateways/). Turns LLM-as-judge evaluation and gateway routing into SQL over a typed relational schema: ~100 signals per session, 95.5% boolean accuracy, 90-92% cost reduction versus naive routing.
- [Full research corpus](https://www.strukto.ai/llms-full.txt). Plain-text dump of all published papers for direct AI citation.

## News
- [Introducing Mirage: A Unified Virtual Filesystem for AI Agents](https://www.strukto.ai/news/introducing-mirage) (2026-05-06). Mirage launches as an open-source virtualization layer for AI agents, mounting data sources, services, and tools as one virtual filesystem with a bash interface instead of writing a tool function per backend or stitching MCP servers. Python (`mirage-ai`) and TypeScript (`@struktoai/mirage-node`, `@struktoai/mirage-browser`) SDKs plus a CLI; adapters for OpenAI Agents SDK, Vercel AI SDK, LangChain, Pydantic AI, CAMEL, Mastra, OpenHands.

## Team
- Zecheng Zhang, Founder (Stanford, UIUC, ex-AWS, YC alumni). Background: ML systems, relational databases, structured data, LLM evaluations, AI agents. Co-author of SEAR (CAIS 2026). Open-source contributor to PyTorch Geometric, PyTorch Frame, RelBench, DeepSNAP, CRAB, CAMEL.

## Links
- Homepage: https://www.strukto.ai/
- Mirage: https://www.strukto.ai/mirage
- Virtual Terminal, live in the browser: https://www.strukto.ai/terminal
- About: https://www.strukto.ai/about
- Research: https://www.strukto.ai/research
- News: https://www.strukto.ai/news

## Contact
- Booking: https://cal.com/strukto/30min