Start here
Token & Throughput Audit
We price your harness three ways: as it runs today, with prompt caching and a shorter tool list, and on a model trained for it. If the second one hits your target, we recommend no build.
Services and pricing
Three practice areas.
We publish the prices. A price you have to ask for depends on how you look.
Fixed fee, priced on the outcome. 50% up front, 50% on acceptance. We never quote an hourly rate.
Custom Models & Token Compression
A model trained on your own tool calls.
Token & Throughput AuditOne week inside your agent harness, ending in the cost of each completed task and what would lower it.Custom Tool-Calling ModelsA model trained on the calls your harness makes, deployed in your cloud or on your hardware.
S7
Custom Tool-Calling Models
We write the eval set from your real tool calls before we train anything. You see the model's score beside your current model's, and the traffic reaches no third-party inference provider.
Say it out loud in the first call
If your agents make 50 calls a day, a model will not pay back. If caching, a shorter tool list or better retrieval hits your target, you do not need one. We run that check first and tell you when the answer is no.
Agent Harness Engineering
Agents do the volume. Your engineers hold the gates.
Start hereToken & Throughput Audit, $4,500. It opens this area too.
The FactoryAn agent harness in your repositories: agents do the volume, your engineers hold four gates.Fractional AI Engineering LeadSenior engineering leadership two days a week, from someone who still writes code.
S1
The Factory
We report weekly: merged pull requests beside review pass rate, revert rate and token cost per merged pull request.
Say it out loud in the first call
If your tests are broken or absent, the harness multiplies your defects. Fix that first, with us or without us, or we do not start.
S5
Fractional AI Engineering Lead
Compliant AI & Governance
Data flows and controls that survive audit.
Compliant AI Readiness AssessmentTwo to three weeks mapping where your AI data goes, ending in something you can file.The Governed AI WorkspaceA sanctioned AI workspace in your own cloud account, with an audit trail.
Start here
Compliant AI Readiness Assessment
We write it for whoever owns the risk, so you answer your board with a plan and dates instead of a moratorium.
S3
The Governed AI Workspace
Routine questions go to a small model and harder ones to a larger model under a per-seat budget. Every prompt lands in your audit log.
What we decline
The list, before you spend an hour finding out.
- Hourly or time-and-materials work.
- Staff augmentation and body-shopping.
- Strategy decks with no working code or model weights.
- Engagements under $4,500.
- Work where we cannot measure the outcome.
- Agent harnesses on public APIs when your compliance rules forbid sending the data out.
- A custom model when caching, a shorter tool list or retrieval would hit your target.
Buyers test the last two. We turn down the work.
Not sure which?
Send your monthly API spend, your turns per task and the framework you are held to.