← Home Stack The systems we orchestrate · production · interconnected
Twenty-plus systems. One operator.
The work in our cases isn’t one tool, one model, one harness. It’s twenty-plus production systems — model providers, agentic harnesses, orchestration, infrastructure, data, automation — interconnected and running flawlessly. This page is the actual stack, with what each piece does and where the seams live. No logo wall. No vendor pitch. Just the architecture.
Layer 01 · Models Foundation models · routed per task
Pick the right brain. For each job.
Different models are good at different things. Reasoning, coding, voice, search, vision, low-latency. The trick is routing each task to the right model and not paying premium prices for tasks a smaller model handles fine.
Anthropic · Claude Primary reasoning, long-context architecture work, agentic code, sensitive client work.
Google · Gemini Multimodal tasks, large-document ingestion, complementary reasoning.
OpenAI · Codex Code-focused completion lanes inside the build pipeline.
Perplexity Live web research with citations — agents call it for fresh-data lookups.
Qwen Alternative reasoning + cost-efficient inference for high-volume agent tasks.
DeepSeek Cost-efficient reasoning for high-volume tasks where Anthropic is overkill.
ElevenLabs Voice synthesis for any agent that needs to speak — client-facing, content, calls.
Layer 02 · Orchestration The harnesses · routing · coordination
Where the agents actually live.
Models alone don’t do work. Agents do. And agents need an environment to live in: tools, memory, eval, observability, hand-off, deployment. This layer is the part most "AI consultants" never touch — and it’s where the leverage actually compounds.
MyCLAW Agentic environment harness — bespoke, cutting-edge, the foundation most production agents run on.
OpenCLAW Open variant of the same pattern — for cases where the harness ships with the deliverable.
Custom harnesses When a specific case needs something neither MyCLAW nor OpenCLAW gives you, we build it.
OpenRouter Model-routing layer — one call surface, many models, cost & latency optimised per task.
Direct API calls For the tasks where OpenRouter overhead doesn’t pay — we go direct to provider APIs.
Custom model orchestration Multi-model pipelines, eval rigs, retry & fallback logic, agent hand-off, lineage.
Layer 03 · Infrastructure Cloud · compute · uptime
Where the systems actually run.
AWS Primary cloud — compute, storage, networking, deployments, observability.
DigitalOcean Lighter-weight services, isolated workloads, simpler provisioning where AWS is overkill.
Layer 04 · Data + Knowledge Stores · second brain · embeddings
What the agents remember.
Supabase Application database + auth + edge functions for production apps.
MySQL Transactional data where the SQL is mature and the schema is established.
Obsidian Second brain — knowledge, briefs, drafts, project memory. Agents read and write into it.
Notion Client-facing deliverables, structured docs, runbooks. Agents draft; human reviews.
Layer 05 · Ops + Automation Project mgmt · productivity · browser automation
The connective tissue. Where humans and agents meet.
Plane Project management — tickets, roadmaps, sprints. Agents move tickets; humans set priority.
GSuite Calendar, mail, drive — where the operator’s day actually lives. Agents read, write, schedule, send.
Playwright Browser automation — agents drive web UIs where there’s no API: scraping, form-filling, testing.
Interconnect How it all stays glued together
The hard part isn’t the tools. It’s the seams.
Anyone can subscribe to all the systems on this page. The leverage comes from how they talk to each other — model routing decisions, eval rigs, lineage tracking, deployment gates, secrets management, error handling, retry logic, cost monitoring, observability, hand-off contracts, runbook automation, sub-agent dispatch.
That’s the work. Every case on /cases runs on this stack — the differences between them are which layers carry more load, which models route where, and which harnesses orchestrate which agents. The Audit maps your version of this. The Sprint builds it. The Retainer runs it.