Skip to content
View convee's full-sized avatar
:octocat:
...
:octocat:
...

Block or report convee

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please donโ€™t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this userโ€™s behavior. Learn more about reporting abuse.

Report abuse
convee/README.md
Convee โ€” Agent systems architect and AI researcher

Hi, I'm Convee ๐Ÿ‘‹

I design and build AI agent systems, with a focus on architecture, secure execution, evaluation, and the engineering required to turn model capabilities into dependable products. I also research reasoning, tool use, context and memory, and the boundaries of current AI systems.

An agent is not a model call. It is a system of models, tools, state, runtimes, permissions, evaluations, and feedback loops.

I am particularly experienced in:

  • Agent Architecture โ€” multi-agent collaboration, task orchestration, tool routing, context and memory, human-in-the-loop workflows
  • Secure Runtime โ€” isolated execution, workspace lifecycle, multi-tenant authorization, quotas, and failure containment
  • Evaluation & Observability โ€” reproducible evals, tracing, failure guards, and quality/cost feedback loops
  • AI Research โ€” evidence-first discovery, source tracing, knowledge engineering, and automated research workflows

Flagship projects

Built and maintained under hullwork โ€” open, self-hosted infrastructure for dependable AI agents.

hullwork/sandbox โ€” Secure runtime for AI agents

A self-hosted execution sandbox for AI agents. Each runtime runs in its own Kubernetes gVisor Pod, while the control plane manages workspaces, tenant authorization, quotas, credentials, and lifecycle. It is accessible through an HTTP API, Python SDK, CLI, and MCP bridge.

The project is concerned not only with running code, but with where an agent runs, what it can access, how it recovers from failure, and how the system preserves its security boundaries.

Kubernetes gVisor Python MCP Multi-tenant Fail-closed

hullwork/site โ€” Verified website delivery for AI agents

A deployment control plane that lets an AI agent ship a website to Kubernetes through HTTP, CLI, or MCP. It handles tenant admission, quotas, desired state, builds, workloads, ingress, observability, and scale-to-zero activation.

The system treats deployment as a claim that must be proven: once a workload is ready, the control plane makes a real HTTP request and records the response status and body digest. โ€œDeployedโ€ is a measurement, not an exit code.

Kubernetes Python MCP Multi-tenant Scale-to-zero HTTP verification

Architecture & research

Area Questions I work on
Agent systems How should planners, executors, tools, memory, and feedback loops compose into systems that can recover and evolve?
Runtime & security How can untrusted code run in isolation with least-privilege identity, resource, and network boundaries?
Evaluation How can traces, offline evals, production signals, and failure analysis form a continuous improvement loop?
AI research How can evidence-first collection, source tracing, synthesis, and knowledge capture produce trustworthy research?
Engineering How can Go, Python, Kubernetes, Docker, and GitOps turn architecture into operable systems?

More builds

An automated research and content archive that continuously collects high-value signals, preserves traceable sources, and turns scattered information into reusable research material.

A lightweight Markdown publishing system built with Go, MySQL, and server-rendered templates, including content management, search, and a secure admin surface.

Six single-file HTML5 games written for my 7-year-old โ€” zero dependencies, zero build step, portrait-first. A reminder that not every system needs a runtime.

Find me

  • ๐ŸŒ convee.cn โ€” AI engineering intelligence and knowledge systems
  • ๐Ÿ—๏ธ hullwork โ€” open infrastructure for dependable AI agents
  • ๐Ÿ’ป GitHub Projects โ€” more open-source work
Architect the system. Ground it in evidence. Ship what works.

Pinned Loading

  1. goblog goblog Public

    Go Markdown Blog ๅšๅฎข็ณป็ปŸ

    JavaScript 93 22

  2. artgo artgo Public

    go web framework

    Go 3

  3. kids-games kids-games Public

    ็ป™ 7 ๅฒๅฐๆœ‹ๅ‹ๅš็š„ H5 ๅฐๆธธๆˆๅˆ้›†๏ผš6 ไธชๅ•ๆ–‡ไปถ้›ถไพ่ต–ๆธธๆˆ๏ผŒๆต่งˆๅ™จ็›ดๆŽฅ็Žฉ

    HTML 2

  4. daily-content-archive daily-content-archive Public

    Cross-platform daily content archive with continuous, source-traceable collection

    Python 1

  5. hullwork/sandbox hullwork/sandbox Public

    gVisor-isolated Kubernetes sandbox with an HTTP API, SDK and MCP bridge

    Python 1 1

  6. hullwork/site hullwork/site Public

    Website build and deployment control plane

    Python 1