AI Backends Guide

This guide covers the Python AI backend package used by Jumpshell skills and extension workflows.

Canonical Paths

  • Python package source: src/python/ai-backends
  • Package module: src/python/ai-backends/ai_backends
  • Package metadata: src/python/ai-backends/pyproject.toml

What It Provides

The ai-backends package provides a unified interface for multiple providers:

  • gemini
  • openai
  • anthropic
  • github-api
  • copilot-cli
  • cursor

It also supports quality-tier selection and model discovery/registry caching.

Install Options

From repository root:

pip install -e ./src/python/ai-backends

Optional extras:

pip install "ai-backends[openai]"
pip install "ai-backends[anthropic]"
pip install "ai-backends[gemini]"
pip install "ai-backends[all]"

Extension Integration

The Jumpshell extension can install/update bundled ai-backends during skill updates when:

  • jumpshell.installAiBackendsOnSkillsInstall = true

Related settings:

  • jumpshell.aiBackendsPath
  • jumpshell.skillsPath

Quick Usage 

import ai_backends

cache = ai_backends.ensure_registry()
backend, model = ai_backends.resolve_quality("normal", cache, vision=False)
text = ai_backends.call_backend(backend, "Summarize this topic", model=model)