Skip to main content

Agent Framework

CrewAI

Multiple AI agents working as a team

CrewAI lets you define a team of AI agents — each with a role, goal, and tool access — that collaborate to complete complex tasks. One agent researches, another writes, a third reviews. Complex workflows that would be a single long prompt elsewhere become structured, reliable processes.

Book a consultation to set up CrewAI →

What it is

AI as a team, not a single assistant

A single AI model trying to do everything produces mediocre results. CrewAI breaks complex tasks into roles: a Researcher with web access, a Writer focused on style, a Reviewer enforcing standards. Each agent is focused on its job and passes work to the next.

The result is more reliable than a single prompt for multi-step workflows. Each agent's output becomes context for the next. Roles enforce quality constraints that a single model conversation doesn't.

We configure your crew for your workflow — the agents, their roles, their tools, and the task sequence. You run the crew; it executes the full pipeline.

How it works

Define once, run repeatedly

A Crew is defined as a Python configuration: agents with roles and goals, tasks with descriptions and expected outputs, a process flow (sequential or hierarchical). Once defined, you trigger the crew with an input, and it runs to completion.

All agents share the same local Ollama instance. Each agent is a separate model call with its own role context. The orchestrator manages task delegation and output passing between agents. Everything runs locally on your Mac.

Who it's for

Anyone with repeatable, multi-step workflows

  • Content teams running research → write → review pipelines
  • Businesses with repeatable multi-step analysis workflows
  • Researchers automating literature review + synthesis
  • Developers building AI-powered pipelines that need reliability
  • Anyone whose complex AI tasks benefit from structured role separation

Unique advantage

Complex workflows that would be a single exhausting prompt elsewhere become structured, repeatable processes. Define the crew once; run it for every client, every report, every cycle. Quality comes from structure, not from hoping the model stays on track.

Full stack

What gets installed

LayerComponentPurpose
AI EngineOllama (MLX backend)Runs models on Apple Silicon
Agent FrameworkCrewAIMulti-agent orchestration with role-based collaboration
Chat UIOpen WebUIBrowser chat, always available
NetworkingTailscaleSecure access from anywhere
RuntimePython environmentCrew definitions and tool integrations
SecurityHardened configLoopback binding, egress logging

Security

Local models, local execution

All agents run against your local Ollama instance — no API calls to cloud AI services. Tool use (web search, file access) is configured with defined scope. All data processed by the crew stays on your Mac unless you explicitly configure an external integration.

Recommended models

Models that pair well

Qwen 3 32B

Reliable instruction following — agents stay on task and execute roles consistently

Llama 3.3 70B

Complex multi-agent reasoning — strong at planning and delegation across agents

DeepSeek R1 32B

Planning and reasoning — excellent for the orchestrator role in complex workflows

Ready to set up CrewAI?

Book a consultation. We'll design your crew for your specific workflow and configure it on your Mac.

Book a consultation to set up CrewAI →