Audit readiness & scoping
$1,500Architecture discovery, risk assessment, connection plan, preliminary suite, and a fixed-scope proposal.
- One discovery workshop
- Architecture profile
- Risk and evidence map
- Credited toward an approved full audit
Pricing & preliminary facts
Fixed-scope work with visible boundaries.
The package covers the audit method. Architecture, capabilities, case volume, security requirements, and remediation cycles determine the final statement of work.
Engagement packages
Start with the smallest defensible scope.
Audit readiness is credited toward an approved full engagement. Third-party model, voice, telephony, and observability usage is separate and passed through without hiding it in the service fee.
Essential audit
$4,500For one response-only assistant that does not retrieve private knowledge or take actions.
- Up to 40 cases
- Baseline audit
- Findings register
- Evidence report
Standard audit & fix
$7,500For one chat or RAG architecture with one controlled remediation and regression cycle.
- 60–100 cases
- Baseline and fixed comparison
- One remediation cycle
- Audit trail and evidence export
Advanced agent audit
From $15,000For voice, tool-calling, workflow, agentic RAG, document, data, or multimodal systems.
- Up to 150 cases or scripted calls
- Action and failure-path testing
- Two remediation cycles
- Latency and observability review
Enterprise / multi-agent
From $25,000For multiple agents, channels, environments, regulated data, or extensive security boundaries.
- Custom case volume
- Cross-agent and authorization testing
- Security and retention requirements
- Stakeholder acceptance package
After the first audit
Continuous assurance
Scheduled regression, drift monitoring, release gates, and maintenance of newly discovered failure cases.
Complexity modifiers
Add-ons expand the evidence contract.
A capability may already be included in the selected base package. The scope builder removes included charges automatically.
RAG knowledge
Retrieval, freshness rules, reranking, and citations.
Action-taking tools
Typed tool calls that read or change business state.
Long-term memory
Customer or organizational memory retained across sessions.
Human handoff
Escalation, approval, or transfer to a person or queue.
Model/provider fallback
Retries, alternate providers, circuit breakers, and degraded modes.
Safety guardrails
Input/output controls, policy checks, and protected-topic handling.
Trace observability
Model, retrieval, tool, cost, and latency traces such as Langfuse.
PII and compliance
Consent, disclosure, redaction, retention, and regulated-data boundaries.
Before a formal quote
Ten facts we need from the customer.
Unknown answers are acceptable. They become explicit discovery tasks instead of hidden assumptions in the quote.
- 01
Business use case
What the AI does, who uses it, and the outcome it supports.
- 02
Architecture
Known system diagram or permission to begin with the discovery profile.
- 03
Channels
Chat, voice, SMS, email, API, documents, or multimodal input.
- 04
Models and providers
LLM, speech, retrieval, telephony, and observability vendors.
- 05
Knowledge and tools
Source-of-truth systems and any actions the agent can execute.
- 06
Safe access
A staging endpoint, replay export, or trace-import path.
- 07
Data sensitivity
PII, financial, health, payment, geographic, or retention constraints.
- 08
Evaluation volume
Approximate number of cases, calls, languages, and configurations.
- 09
Service boundary
Audit-only, customer-implemented fixes, or full audit-and-fix.
- 10
Timeline and approval
Deadline, decision owner, and who accepts manual-review outcomes.
These are launch prices and preliminary estimator rules, not a binding quote. The signed statement of work controls scope, deliverables, assumptions, exclusions, payment schedule, and acceptance criteria.