Nexus AI Harness

Build LLM agents by composing the pieces you want.

Every part of an agent is a plugin: the loop, model, memory, tools, permissions, subagents, and more. You pick the plugins you need and register them on a harness, and you can swap any piece without touching the rest.

Why Nexus AI Harness?

• Start in one line. A batteries-included harness gives you a working agent right away.
• Compose, don't configure. Build an agent by adding the plugins you want.
• Swap any part. Change the model, memory, or tools without rewriting the rest.
• Many models, one interface. Anthropic, OpenAI, Gemini, Bedrock, and more, all called the same way.
• Extend without forking. Add your own plugin by implementing an interface.
• Safe by default. Built-in permissions, sandboxing, and spend and time limits.
• Ready for real work. Memory, subagents, skills, MCP tools, and session resume come included.

Install

pip install nexus-ai-harness

Set the API key for the model you'll use (copy .env.example to .env and fill it in). For Anthropic that's ANTHROPIC_API_KEY. See Models for other providers.

Usage

Start with the batteries-included harness:

from nexus_ai_harness.plugins import default_harness
from nexus_ai_harness.plugins.models import AnthropicModel

harness = default_harness(AnthropicModel(model="..."))
answer = harness.run_sync("What is 128 * 47?")   # or: await harness.run(...)

Or compose your own:

from nexus_ai_harness.harness import NexusAIHarness
from nexus_ai_harness.plugins.loops import AgenticLoop
from nexus_ai_harness.plugins.models import AnthropicModel

harness = (
    NexusAIHarness()
    .use(AgenticLoop())
    .use(AnthropicModel(model="..."))
    .use(...)                       # add the plugins you want
)

Documentation

See docs/ for the architecture, the plugin guides, and how to write your own plugin.

Contributing

See CONTRIBUTING.md.

License

MIT