These are planning bands, not a checkout. There is no order form or payment on this site. No customer has been quoted or has bought at these prices, and no sealed acceptance test has been run yet, so the acceptance steps below describe our plan. The final quote depends on corpus size, connectors, permissions, benchmark design, model size, security, and deployment target — and we publish the quote before you commit.
Managed endpoint, tenant-isolated. Fastest route to a working system.
$12k–$25k
implementation · enterprise $50k–$100k
then
$750–$2.5k / month
enterprise $5k–$10k + usage
Runs inside your boundary. Notebook, department, VPC, or air-gapped.
$25k–$60k
notebook/dept · VPC or on-prem $75k–$175k
then
15%–20% annually
support, updates, re-testing
Tuning is added only where it beats search alone on the measured questions.
$40k–$85k
build & validation · enterprise from $125k
then
$2k–$7.5k / month
enterprise from $10k + compute
The full transfer: weights, index, harness, runbook — inside your boundary.
$55k–$95k
notebook · workstation $90k–$175k · rack $175k–$400k+
then
15%–20% annually
or $2.5k–$10k / month
How payment tracks proof
20%
Corpus, use cases, benchmark charter, and rubric approved
25%
Governed corpus and working search system delivered
30%
Candidate passes visible validation on the target hardware
25%
Only on passing the sealed question set — or a reserve set after one fix cycle
What happens if we miss
You receive category-level failure diagnostics — never the hidden questions. We fix the system against the open development questions and run the reserve set once.
If the reserve set also misses, you keep the corpus, index, harness, and results, and ordinary liability is capped at the unpaid acceptance milestone plus a predefined service credit.
A 10%–20% premium applies to a passed acceptance test, because we carry the fix and payment risk. Hardware, exceptional compute, and third-party licences are quoted separately.