Per Desk Audit Procedure Note 2 (Rev. 6/28/07). For standard annual audit use BOY/EOY. For 13-month fieldwork extension, enter current fieldwork count as EOY.
Configure the LLM provider powering Audit Copilot — this page is hidden from regular users
LLM Provider Selection
Choose the provider appropriate for your deployment environment. Government production deployments must use a FedRAMP-authorized provider.
WARNING:
Recommended Models by Use Case
Environment
Provider
Model
Cost (est.)
FedRAMP
Action
Testing
Ollama (local)
llama3.1
Free
N/A — Local
Testing
DeepSeek ⚠️
deepseek-chat
~$0.001/query
NO — Chinese servers
Staging
Groq
llama-3.1-8b-instant
~$0.0001/query
No
Production
Google Gemini
gemini-1.5-flash
~$0.0002/query
Moderate
Production (Govt)
AWS Bedrock
meta.llama3-1-8b-instruct-v1:0
~$0.0003/query
High ✓
Production (Govt)
Azure OpenAI
gpt-4o-mini
~$0.0005/query
High ✓
AI Usage Statistics
—
Total Requests
—
Total Tokens Used
—
Est. Total Cost (USD)
Recent AI Calls
Regulatory Knowledge Base
Loading…
Upload statute PDFs, desk audit procedures, or any regulatory document.
Audit Copilot will search these documents when answering questions and
display clickable citations that open the source directly.
—
Documents
—
Indexed Chunks
—
URL References
Knowledge Coverage
What the Copilot actually knows, per source — not just a chunk count. A collection only shows
Complete when real source text is indexed; Not Loaded means no source document exists anywhere
in this environment, never an estimate.
Fetch and index an online statute page so the AI can cite it with a clickable link.
Fetching and indexing URL…
Loaded Documents
Loading…
Audit Copilot
Powered by Agent F — explains findings, answers audit questions, summarizes evidence, and assists report writing
Not configured
Audit Copilot — Configuration Required
Audit Copilot is not yet available because no approved AI provider is configured and verified. This is a government audit platform — AI fails closed rather than guessing.
To enable:
Configure an approved AI provider (FedRAMP for production)
Validate API credentials
Confirm token budget and an authoritative RAG knowledge base
Contact your system administrator.
Knowledge Base Not Yet Fully Loaded
Knowledge-mode answers are grounded only in documents that have been fully indexed (see Admin → Knowledge Base for the document list). Questions requiring un-indexed regulatory text will correctly return "no authoritative information was found" rather than an answer from general knowledge.
Audit Copilot Chat
Ask anything about your audit data, compliance status, or the audit procedure. Examples: "Which manufacturers have variance >5%?" · "Explain the stamp inventory formula" · "What's next for case TEACS-001?"
Step 3/4/5 — Interview Note Analyzer
Paste your raw field interview notes. The AI will extract structured findings, flag inconsistencies, identify risk indicators, and suggest follow-up procedures.
Step 8 — AI Conclusion Drafter
Enter the audit findings and the AI will draft a formal Step 8 conclusion narrative in OAG legal style. You review and sign — no blank-page drafting.
Explain Risk Flag
Get a plain-English explanation of any risk flag with statutory references and investigation steps.
Extract Non-Standard AG-1/AG-2
If Agent B's column detector fails on a non-standard format, paste the raw text here. The AI will extract brand, manufacturer, period, packs stamped, and sales price.