Pricing
Three shapes.
One team.
The audit is free. Embed is quoted per scope. Run is a monthly retainer. You move between tiers as the engagement matures; we don't move you down.
Start here
Audit
We map your current agent surface area and write you a 4-page audit. No charge, no obligation. Most engagements start here.
Free1 week
- Written 4-page audit
- Latency & cost breakdown
- Eval coverage map
- Follow-up embed or run
- On-call coverage
Most chosen
Embed
Two senior AI engineers embed with your team and ship the missing piece to production. Knowledge transfer is part of the contract.
Quote on request4–8 weeks
- Written 4-page audit
- Latency & cost breakdown
- Eval coverage map
- Follow-up embed or run2 senior engineers
- On-call coverage
Long term
Run
We take the on-call. Model swaps, drift, capacity, paging, monthly architecture review. 99.9% SLA with a real runbook behind it.
Monthly retainermonthly
- Written 4-page audit
- Latency & cost breakdown
- Eval coverage map
- Follow-up embed or runFull team
- On-call coverage99.9% SLA
Side by side
What's included.
| Capability | Audit | Embed | Run |
|---|---|---|---|
| Latency & cost breakdown | |||
| Eval coverage map | |||
| Risk register & remediation plan | |||
| Production deployment | |||
| 99.9% uptime SLA | |||
| On-call rotation (HKT) | |||
| Quarterly model-swap review | |||
| Cost-cap on LLM spend | |||
| Source code transfer |
Common questions
On pricing.
Is the audit really free?
Yes. One week of engineering time, no charge, no obligation. We've refunded the audit before - we'd rather earn the work.
Do you do equity / sweat-equity?
Sometimes. We have a small dedicated pool for early-stage AI products. Pitch us.
Can we pay in HKD / RMB / USD?
Yes. We invoice in HKD by default; USD and RMB on request. Wire or local bank transfer.
What's the cap on Embed?
Embed is capped at 8 weeks. After that you either move to Run (managed ops) or wrap. This forces a real handover, not an indefinite dependency.
What does the Run monthly retainer cover?
Everything post-MVP: on-call during HKT hours, model swaps, schema drift, capacity, prompt regression, eval suite maintenance, monthly architecture review, quarterly business review.