tools i use.

Mac Mini M4

primary host

Primary compute for the self-hosted fleet: Ollama local inference, PM2 services, Redis, SQLite memory stores, Docker workloads, and private routing for sovereignty-sensitive work.

alfred on Tailscale · local model cache · 36-agent control plane

self-host

Claude

primary reasoning

Primary reasoning path for architecture, code review, planning, and high-stakes operator work. Runs through Claude Code and ClaudeSwap where rate windows matter.

No affiliate program · anthropic.com

paid

Gemma

local / air-gapped

Local open-weight route for private prompts, cheap drafts, and offline tests on the Mac Mini. Useful when data should stay on the machine.

Gemma local via Ollama

free

Mistral

codestral / code spec

Codestral and Mistral Small are cost-control routes for code completion, EU-friendly inference, and lightweight tasks that do not need a frontier model.

Codestral · long-context coding route

paid

HuggingFace

121-model fallback

Router for open models and provider experiments. It is where niche models enter the fleet before they earn a permanent route.

DeepSeek, Llama, Qwen, and specialty backends

free

Nous Research

open-weights research

Open-weight research models for experiments, evals, and checking whether a cheaper route can handle work before a paid model gets involved.

free

Perplexity

research / grounded

Grounded research lane when current source lookup matters. It feeds briefs, competitive checks, and sanity passes before decisions get handed to agents.

paid

Ollama

local model runtime

Runtime for private local models and embeddings. It keeps test prompts and sensitive client work inside the Mac Mini boundary.

Gemma · Llama · local embeddings

self-host

Groq

bulk agents / llama

Fast low-cost inference for bulk agent lanes, workers, and routing tests. It handles volume when latency matters more than deep reasoning.

Llama route for high-throughput tasks

free

Cerebras

qwen / gpt-oss

Large-model inference lane for one-shot research and free-tier pressure relief. Best when scale matters and the task is bounded.

No affiliate program · cerebras.ai

free

Cloudflare AI

edge inference

Edge inference and fallback routing when API providers rate-limit. It pairs with Workers and Tunnel in the same Cloudflare account.

Workers AI · edge route · fallback lane

free

OpenRouter

fallback router

Single-key access to many model routes. Useful for fallback tests, cost comparisons, and trying new models before wiring native providers.

Affiliate: openrouter.ai/?ref=kevinz

free

MiniMax

long-context / multimodal

Long-context and multimodal route for experiments where frontier labs are not the right cost fit or where alternate behavior is useful.

paid

Portkey

llm gateway

Gateway layer for model routing, request logs, and cost controls. It gives the fleet one place to reason about provider behavior.

paid

Plausible

privacy analytics

Privacy-first analytics for site and funnel checks. It stays in the toolchain because routing decisions need traffic context.

paid

OpenAI Agent Builder

gpt-native agents

Managed agent surface for clients who want GPT-native tool calls, retrieval, voice, and vision without owning a custom orchestration layer.

No affiliate program · platform.openai.com

paid

Vertex AI Agent Builder

google cloud agents

Enterprise agent path for GCP shops, especially when grounding against BigQuery, Google Cloud data, and Workspace workflows.

No affiliate program · cloud.google.com

paid

Copilot Studio

microsoft agents

Custom copilots over Microsoft 365, Dataverse, and Power Platform. Strong fit when governance and existing enterprise identity matter.

No affiliate program · microsoft.com

paid

HubSpot Breeze

crm agents

Marketing and CRM agents for teams already operating inside HubSpot. Good for fast deployment when the data model already lives there.

No affiliate program · hubspot.com

paid

Salesforce Agentforce

enterprise agents

Sales and service agents inside Salesforce. Best fit when Data Cloud, CRM workflows, and audit trails are already part of the business.

No affiliate program · salesforce.com

paid

OpenClaw

self-hosted gateway

Backbone for the multi-channel agent platform. It routes messages, loads skills per agent, tracks budgets, and keeps the 36-agent fleet coordinated.

Gateway :18789 · 36 agents · 6 channels

github.com/KevinZai

free

Claude Code

primary ide

Primary coding surface. Custom agents, hooks, skills, and repository context all start here before work moves through review gates.

CC Commander · 60 plugin skills

No affiliate program · claude.ai/code

paid

Codex CLI

oauth code agents

OAuth-only coding agents for implementation, review, and local automation. No API key path is used for this lane.

No affiliate program · openai.com/codex

free

Cursor

rapid prototyping

Traditional editor lane for quick visual diffs, prototypes, and work that benefits from a file tree plus agent chat in one window.

Affiliate: cursor.com/?ref=kevinz

paid

Commander

60 plugin skills

Open-source command layer for managing Claude Code, Codex, and agent-driven workflows. It keeps repeatable operations out of memory and inside skills.

github.com/KevinZai/commander

free

Paperclip

issue-first dispatch

Task dispatch and governance layer. It creates issue-first work, routes implementation, and pushes changes through spec and quality gates.

localhost :3100 · branch naming by ticket

self-host

n8n

self-hosted workflows

Webhook and integration layer for non-code automation. It runs self-hosted on Docker for workflows that should stay close to private data.

Docker service · webhook triggers

self-host

Langfuse

llm tracing / cost

Trace inspection, latency analysis, and per-request cost visibility for the model lanes that need deeper observability.

paid

Promptfoo

eval regression

Evaluation framework for prompt changes, routing regressions, and provider comparison before a route is trusted in production.

free

Datadog

observability

Infrastructure and app observability for hosted services where Langfuse is the wrong shape or logs need a broader operational view.

paid

OpenWebUI

local model ui

Local model interface for testing Ollama routes, comparing prompts, and giving non-terminal access to private models.

self-host

Inngest

durable workflows

Durable background workflow option for jobs that need retries, visibility, and clearer state than a raw queue.

paid

Logsnag

events / alerts

Lightweight event tracking for agent actions, deployment markers, and operator alerts that should show up fast.

paid

AWS

iam / primary cloud

Primary cloud account for infra pieces that need IAM, mature primitives, and long-running reliability.

paid

Vercel

frontend hosting

Hosts kevinz.ai and static frontend properties. Fast deploys, edge delivery, and clean rollback paths for public web surfaces.

Affiliate: vercel.com/?ref=kevinz

paid

Railway

backend services

Quick backend deployment lane for services where a small always-on app beats serverless complexity.

paid

Cloudflare

tunnel / workers / cdn

DNS, CDN, Workers, and Tunnel for the home lab. Keeps local services reachable without opening inbound ports.

paid

Tailscale

sovereign mesh

Private network between machines, services, and admin surfaces. The Mac Mini is reachable as alfred without public exposure.

self-host

Coolify

self-hosted deploys

Self-hosted deployment control plane for apps that belong on owned infrastructure instead of a managed platform.

self-host

Docker

containers

Container baseline for n8n, local services, sandboxes, and repeatable dev environments on macOS and servers.

free

BigQuery

analytics / grounding

Warehouse layer for analytics and Google Cloud grounding when client data already sits inside the GCP world.

paid

Supabase

postgres / auth

Postgres, auth, and storage when a product needs real app primitives without standing up a custom backend stack.

paid

Fly.io

region failover

Always-on backend services and regional failover when serverless is the wrong operational model.

Affiliate: fly.io/?ref=kevinz

paid

OrbStack

docker on macOS

Fast Docker runtime on Apple Silicon. Lower friction than Docker Desktop for local services and agent-adjacent containers.

paid

PM2

persistent services

Process manager for long-running local services. Restarts failed agents and keeps the home-lab process list sane after reboots.

20+ persistent services

free

Caddy

reverse proxy

Reverse proxy for local services, paired with Cloudflare Tunnel for clean routing and auto-TLS where needed.

free

Redis

cache / queue

Cache and queue layer for agent coordination, session state, and fast handoffs between services.

free

GitHub

source / ci

Source of truth for code, issues, reviews, releases, and public open-source work.

github.com/KevinZai

free

Sentry

error tracking

Error tracking for production web surfaces and APIs where failures need context, grouping, and alerts.

paid

Stripe

billing

Billing and SaaS revenue infrastructure. It remains the cleanest default for payments, subscriptions, and invoices.

paid

Resend

transactional email

Transactional email and newsletter delivery for site forms, notifications, and publication workflows.

Affiliate: resend.com/?ref=kevinz

paid

PostHog

product analytics

Product analytics, events, and funnels when a project needs more behavioral detail than page analytics.

paid

Twilio

sms / voice

SMS and voice rails for customer-facing automations, verification, and agent-assisted communications.

paid

1Password

op:// secrets

Secret source of truth. Agents and services read credentials through op:// references, never hardcoded keys.

paid

Obsidian

second brain

Personal knowledge base, notes, vault sync, and long-lived thinking. It feeds writing and agent memory workflows.

free

Notion

specs / client docs

Client-facing docs, specs, research summaries, and structured pages where collaborators need readable context.

paid

Slack

team comms / alerts

Team communication and agent alert surface. It is where operational updates land when they need human attention.

paid

Discord

community / dev forums

Community, developer forums, and live discussion channels where projects and tools surface early signal.

free

Todoist

tasks / inbox

Personal task inbox for work that should not become a repo issue, plus quick capture before agent routing.

paid

Home Assistant

local automations

Local automation and device state. It gives home workflows a private control plane instead of another cloud dependency.

self-host

Linear

personal dev tasks

Issue tracker for implementation discipline. Agent work starts with a ticket, not a vague request floating in chat.

Affiliate: linear.app/?ref=kevinz

paid

Attio

crm / lead scoring

CRM and relationship database for cleaner lead tracking, customer context, and agent-accessible contact records.

paid

Clay

data enrichment

Data enrichment and go-to-market operations. Useful when a workflow needs structured company or contact context.

paid

Granola

meeting ingestion

Meeting notes and transcripts routed into memory and follow-up workflows. Calls become structured inputs for agents.

paid

BlueBubbles

imessage bridge

iMessage bridge for personal automation experiments and channel testing where mobile messages need an agent lane.

self-host

Raycast

launcher / snippets

Launcher, snippets, quick commands, and local workflow shortcuts. It is the operator console for small actions.

paid