AQVEN

An engineering environment for AI systems. Define, validate, test, run, evaluate and debug LLM workflows as a system — instead of leaving their logic scattered across prompts, model settings, tools, orchestration code and tests.

📖 Documentation · 🚀 Quickstart · 🧭 Studio guide · 📚 API reference

The problem it solves

A prototype is input → model → output. Production is not. A real AI system combines several models, context sources, tools, retries, judges and verification steps — and once it does, a successful run stops being evidence that anything works.

Changing a prompt, a model, a provider or a schema affects quality, cost and latency differently for different inputs. AQVEN makes those parts and their relationships explicit, so you can answer:

• What system do we actually have?
• Which prompt, model, context, tool or step should change?
• Did that change improve quality for the cases that matter?
• Why did this run produce an almost-right result?
• Can a coding agent understand the system well enough to change it safely?

Install and open it

AQVEN installs as a single package. Studio, the local browser interface, ships inside it — no Node.js toolchain, no separate front-end install, nothing to build.

uv tool install aqven
aqven new my_project
cd my_project
uv run aqven dev

aqven new creates the project from a template, installs its environment and generates typed models. aqven dev starts the local server and opens Studio in your browser.

How a project looks

Workflows are files in your repository, not rows in a database. A node is a YAML file, a prompt is a Markdown file, a code step is a Python function referenced as module:function. Identity is the path on disk, history is git.

my_project/
  aqven.yaml                 providers, data policies, trust defaults
  types/                     types declared once, Python models generated from them
  agents/                    model, settings, output contract, limits
  flows/support_case/
    flow.yaml
    nodes/triage/
      triage.node.yaml
      triage.prompt.md
      triage.py
  evals/                     datasets and scored cases

That layout is the reason a coding agent — Claude Code, Codex, Cursor — can read the system before editing it, and the reason aqven check can validate the whole project before anything runs.

The engineering loop

Define → Validate → Test → Run → Evaluate → Learn → Improve

What it is built on

AQVEN does not reimplement the parts of the stack that already work well. Model-facing agents, providers and tools use Pydantic AI; durable execution, recovery and waits for people use DBOS. AQVEN adds the project format, typed workflow contracts, the compiler, checks, run interfaces and Studio. See Architecture.

Runs locally on SQLite. Python 3.14.

Keywords

AI workflow orchestration · LLM pipelines · prompt engineering · agent evaluation · LLM observability · structured output · human-in-the-loop · MCP server · durable execution · Pydantic AI · DBOS

License

AQVEN License 1.0.0, based on PolyForm Shield 1.0.0 — source-available. Use it for any purpose, including in production and in commercial products you build with it. You may not use it to provide a product that competes with AQVEN, and you may redistribute AQVEN itself only free of charge and for a non-commercial purpose.

Copyright Kirill Burkhanov.