Metadata-Version: 2.4
Name: aqven
Version: 0.0.3
Summary: Engineering layer for reliable AI workflows: files as the source of truth, a compiler, a checkpointed runtime and a local Studio.
Keywords: ai,llm,workflow,agents,evaluation,pydantic-ai
Author: Kirill Burkhanov
License-Expression: LicenseRef-AQVEN-1.0.0
License-File: LICENSE
Classifier: Development Status :: 3 - Alpha
Classifier: Intended Audience :: Developers
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.14
Classifier: Topic :: Software Development :: Libraries :: Python Modules
Classifier: Typing :: Typed
Requires-Dist: aqven-llm==0.0.3
Requires-Dist: pydantic==2.13.5
Requires-Dist: ruamel-yaml==0.19.1
Requires-Dist: python-liquid==2.3.1
Requires-Dist: httpx2==2.13.0
Requires-Dist: pydantic-ai-slim[mcp]==2.43.0
Requires-Dist: pydantic-evals==2.43.0
Requires-Dist: dbos==2.31.1
Requires-Dist: fastapi==0.141.1
Requires-Dist: uvicorn==0.53.0
Requires-Dist: python-multipart==0.0.32
Requires-Dist: mcp==2.2.0
Requires-Dist: rfc8785==0.1.4
Requires-Dist: watchfiles==1.2.0
Requires-Dist: genai-prices==0.1.7
Requires-Dist: scipy==1.18.1
Requires-Dist: numpy==2.5.3
Requires-Dist: opentelemetry-sdk==1.44.0
Requires-Dist: opentelemetry-exporter-otlp-proto-http==1.44.0
Requires-Dist: claude-agent-sdk==0.2.154
Requires-Dist: openai-codex==0.147.0
Requires-Dist: httpx==0.28.1
Requires-Dist: tenacity==9.1.4
Requires-Dist: python-dotenv==1.2.3
Requires-Dist: aqven-llm[anthropic]==0.0.3 ; extra == 'anthropic'
Requires-Dist: aqven-llm[bedrock]==0.0.3 ; extra == 'bedrock'
Requires-Dist: aqven-llm[cohere]==0.0.3 ; extra == 'cohere'
Requires-Dist: aqven-llm[google]==0.0.3 ; extra == 'google'
Requires-Dist: aqven-llm[groq]==0.0.3 ; extra == 'groq'
Requires-Dist: aqven-llm[huggingface]==0.0.3 ; extra == 'huggingface'
Requires-Dist: aqven-llm[mistral]==0.0.3 ; extra == 'mistral'
Requires-Dist: aqven-llm[xai]==0.0.3 ; extra == 'xai'
Requires-Python: >=3.14, <3.15
Project-URL: Homepage, https://aqvenstudio.com
Project-URL: Documentation, https://aqvenstudio.com
Project-URL: Repository, https://github.com/lastonoga/AQVEN
Project-URL: Issues, https://github.com/lastonoga/AQVEN/issues
Provides-Extra: anthropic
Provides-Extra: bedrock
Provides-Extra: cohere
Provides-Extra: google
Provides-Extra: groq
Provides-Extra: huggingface
Provides-Extra: mistral
Provides-Extra: xai
Description-Content-Type: text/markdown

# 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](https://aqvenstudio.com)** · 🚀 **[Quickstart](https://aqvenstudio.com/engineering/quickstart/)** · 🧭 **[Studio guide](https://aqvenstudio.com/studio/)** · 📚 **[API reference](https://aqvenstudio.com/engineering/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.

```bash
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
  datasets/                  cases a flow runs on
  experiments/               variants, checks and a question over the 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

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

| Command | What it does |
| --- | --- |
| `aqven check` | validates the project and simulates every flow **without network access or tokens** |
| `aqven run` | runs a flow locally and streams run events as they happen |
| `aqven dev` | project server plus Studio: canvas, runs, datasets, human review |
| `aqven serve` | HTTP API and MCP server, without a browser |
| `aqven mcp` | MCP over stdio, so a coding agent can inspect and edit the project |

## 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](https://ai.pydantic.dev/); durable execution, recovery and waits for people use
[DBOS](https://www.dbos.dev/). AQVEN adds the project format, typed workflow contracts, the compiler, checks,
run interfaces and Studio. See [Architecture](https://aqvenstudio.com/engineering/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](https://polyformproject.org/licenses/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.
