Skip to content

Ajolopy

The Python framework for building AI-native applications in production.

What Rails was for database-backed web apps and what NestJS is for enterprise Node services: the default choice when what you're building has LLMs, agents, tools, prompts, evals, streaming, and MCP as core ingredients — not as an addon.

Ajolopy does not compete with FastAPI or LangChain. It competes for the complete AI-native application framework category, which is currently empty in Python.

Status: v0.1 — first public release

APIs are stable for the ten primitives; expect additive changes in v0.1.x. Pin exact versions if you depend on the framework today. Active roadmap and current work live in board.json.

Why Ajolopy

Building an AI-native app today means stitching together:

  • a web framework (FastAPI),
  • an LLM SDK (Anthropic / OpenAI / Gemini),
  • a streaming layer,
  • a tool-call dispatcher,
  • an evaluation harness,
  • prompt versioning,
  • MCP client and server transports,
  • observability (OpenTelemetry, Langfuse, etc.),
  • a project layout convention.

Ajolopy ships all of that as one opinionated framework with ten primitives.

The 10 primitives

AI (7) Framework (3)
@Agent, @Tool, @Stream, @Eval, @Metric, @Workflow, @MCP @Module, @Injectable, @Controller

Every primitive ships with a magical default that covers the 90% case with zero ceremony, and an escape hatch (subclass or override) for the 10% that needs full control.

Where to go next

  • :material-rocket-launch: Quickstart Five minutes from uv pip install ajolopy to a working agent answering curl requests.

  • :material-school: Tutorial The three-step killer demo arc — agent, evals, multi-agent + MCP — in roughly 55 lines.

  • :material-chart-line: Recipes Plug Ajolopy into Langfuse, Sentry, Grafana, Honeycomb, or Datadog in under 10 minutes — same pipeline, different exporter.

  • :material-package-variant: Install Core install plus the full optional-extras matrix (observability, MCP, memory backends).

  • :material-map-marker-path: Next steps Per-primitive reference, example projects, observability and deploy recipes, and how to contribute.

Design tenets

  • Default magical, escape hatch always available. Every primitive works as a one-line decorator and lets you subclass when you need full control.
  • No 11th primitive in v0.1. Anything that would require a new decorator goes to v0.2+. The surface stays small on purpose.
  • Production from day one. Observability, env validation, deploy templates, and evals are first-class — not addons.

License

MIT. Source on GitHub.