appecosystem-ai

Shared, hardware-adaptive, provider-pluggable AI layer for appEcosystem.

Local models are the default. The router probes the machine's capability tier, picks a model that will actually run on it, and falls back gracefully when the preferred one is unavailable. Cloud providers are opt-in, configured per task — so you can send security analysis to a frontier model while keeping high-volume chat local and free.

Install

pip install appecosystem-ai

Requires Python 3.9+. Local inference talks to Ollama over HTTP and needs no extra Python packages.

Cloud providers are optional extras — install only what you use:

pip install "appecosystem-ai[anthropic]"
pip install "appecosystem-ai[openai]"
pip install "appecosystem-ai[gemini]"

Ask a local model

import asyncio
from ecosystem_ai import AIProfile, build_router
from ecosystem_ai.providers import ChatMessage

async def main():
    router = build_router(AIProfile())      # detects this machine's tier
    result = await router.chat(
        [ChatMessage(role="user", content="Summarize today's alerts.")]
    )
    print(result.content)

asyncio.run(main())

build_router() detects the hardware tier when you do not pass one. Skipping that detection leaves the stock selected_model="auto" profile with no model to resolve, so let it detect unless you have a specific tier in mind.

Route a task to the cloud

task selects which provider handles the call, per the profile's routing table:

result = await router.chat(messages, task="security")

With task_providers={"security": "anthropic", "chat": "ollama"}, security analysis goes to Claude while chat stays local. Tasks that need judgement warn when the selected model is too small to be trusted with them.

Check what this machine can run

from ecosystem_ai import detect

info, tier = detect()
print(f"{info.ram_gb} GB RAM, {info.cpu_cores} cores, GPU={info.has_gpu} — tier {tier}")

License

MIT — see LICENSE.