VibeSys: Generating Bespoke Systems with AI Agents

An agentic framework that generates bespoke systems from application requirements, workload characteristics, and the underlying hardware.

One of VibeSys's first initiatives is VibeServe, which asks whether AI agents can generate a bespoke LLM serving system for each model, workload, and hardware target. The figures, blog post, and paper below document that initiative.

Updates

• 2026-05 — VibeServe blog post: Let AI Agents Write Your Serving Stack with VibeServe.
• 2026-05 — Paper released on arXiv: 2605.06068.

Introduction

VibeSys explores a broader approach to systems development: use application requirements, workload characteristics, and hardware capabilities as the inputs to an agentic search process that creates a purpose-built system. Each target defines its own implementation contract, correctness checks, and performance benchmark, allowing VibeSys to work across domains rather than assuming a single runtime, programming language, or deployment shape.

The framework is organized as a multi-agent optimization loop. An outer loop plans the search over system designs using persistent state such as issues, memory, and git history, while an inner loop implements candidates, validates correctness against target-specific requirements, and measures performance on the target workload and hardware. VibeServe is the first substantial initiative built on this approach; its serving-focused results include predicted-output decoding, hybrid prompt caching, streaming ASR, constrained JSON decoding, multimodal inference, and Apple Silicon deployment.

Architecture

The framework factors the work along two axes:

• Outer loop — a fresh designer selects one falsifiable causal hypothesis from git history, profiling evidence, and durable roadmap/progress memory, then hands it off until it is proven, disproven, or otherwise terminated.
• Inner loop — a hypothesis-scoped implementer session edits the candidate, chooses targeted experiments and parameter ranges, and reports whether to continue or nominate the result.
• Independent judge — a fresh, read-only reviewer checks the implementation, activation evidence, invariants, and reward-hacking risks at a sparse cadence. After a PASS, the framework—not an agent—runs and records the canonical accuracy and benchmark commands.
• Performance evaluator — profiles the implementation (Nsight Systems, PyTorch profiler) and feeds bottleneck hints into future design decisions.
• Skills library — Agent Skills entries distilled from existing serving engines and research literature (continuous batching, paged-KV, FlashInfer/FlashAttention, MLX, hybrid-cache management, …). New model families, hardware platforms, and optimization techniques are added by writing a skill, not by modifying the framework.
• Execution environment: an isolated runtime view where candidate source is writable while evaluator-owned inputs and framework metadata are read-only and integrity-checked. It exposes the target hardware (local CUDA, Modal, Docker, or Apple Silicon) plus profilers.

Each round is recorded in git and a framework-owned audit. Provisional rounds remain explicitly unreviewed; only judge-approved candidates receive official accuracy and performance results.

Quickstart

Install Python 3.12+, Git, and uv. Linux also requires bubblewrap, and a kernel that lets it create an unprivileged user namespace; macOS includes the required sandbox-exec command. Where user namespaces are blocked and installing bubblewrap needs root you do not have, VIBESYS_AGENT_SANDBOX=landlock selects a weaker but root-free backend (see the CLI reference for what it stops enforcing). Tasks whose candidate is a container topology, such as the microservices examples, also need Docker Engine reachable without sudo (add your user to the docker group) and a Go toolchain on PATH. Then install VibeSys:

uv tool install vibesys

Install and authenticate a supported coding-agent CLI. For Codex CLI, run codex login; see the CLI reference for other supported agents.

From the root of the project you want to optimize, add a named task under .vibesys/tasks/:

• .vibesys/tasks/<task>/OBJECTIVE.md describes what to optimize and the constraints the result must preserve.
• .vibesys/tasks/<task>/vibesys.input.toml identifies the problem domain and the programs that check correctness and benchmark performance.
• agent.toml optionally selects the coding agent, model, and hardware backend. Keep it untracked.

For example:

[model]
name = "gpt-5.4"

[agent]
backend = "cli"
# Optional. AgentShim is the default; Omnigent is an alternate runtime.
driver = "agentshim"
cli_provider = "codex"

[backend]
name = "cpu"

See examples/ for complete objectives and manifests across data structures, model serving, and microservices.

Run from the project root:

cd /path/to/my-project
vibesys validate --task my-task
vibesys --task my-task --max-rounds 4

The directory must be its Git repository root, or outside Git so VibeSys can initialize a repository. An existing repository needs a baseline commit and a clean worktree. See Running VibeSys for copied projects, legacy input bundles, Docker, Modal, remote repositories, resume, and alternate search loops. The CLI reference documents every flag. Contributor setup belongs in docs/contributing/development.md.

Citation

If you use the VibeServe initiative in your research, please cite:

@misc{kamahori2026vibeserveaiagentsbuild,
      title={VibeServe: Can AI Agents Build Bespoke LLM Serving Systems?},
      author={Keisuke Kamahori and Shihang Li and Simon Peter and Baris Kasikci},
      year={2026},
      eprint={2605.06068},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2605.06068},
}