The Platform is the foundation layer of the Archipath AI-of-record stack. Every product above it inherits the same governance, routing logic, evidence trail, and AI Index inputs. Automatic provisioning is operator-hosted. Customer-cloud containment and operator access require a separate deployment review.
Every request is classified and routed to the lowest-cost model that can handle it, and premium models run only when the work needs them. How much you save depends on your workload, provider pricing and quality requirements; we measure it with you during the first measurement period instead of promising a fixed percentage.
Formatting, translation, extraction. Runs on Bedrock Llama or local OSS.
~$0.00 per requestSummaries, simple Q&A, routine drafting. Pennies, not dollars.
~$0.002 per requestResearch, coding, analysis, document processing. Quality-cost balanced.
~$0.01–$0.03 per requestLegal analysis, architectural decisions, regulator-facing output. Premium only when warranted.
~$0.05–$0.15 per requestIn most cases, names, phone numbers, email addresses, Social Security numbers, and card and account numbers are masked before your company's prompts reach the AI provider. These controls are separate from how a deployment handles prompt text. Provider retention and training terms still require their own review.
Agents read from and write to your firm's systems. In most cases, names, phone numbers, email addresses, Social Security numbers, and card and account numbers are masked before your company's prompts reach the AI provider. Every write requires explicit user approval. Every action is logged. Four built-in connectors plus a universal MCP adapter cover the long tail.
OAuth2 auth. Read across SharePoint and email. Write calendar events, draft emails, upload files — with approval.
Query client lists and ledgers. Generate invoices on approval. Reconcile payments. Cost attribution maps cleanly to client codes.
OAuth2 auth. Read mail and Drive. Draft emails and create events on approval.
Any MCP-compatible server. Tools discovered automatically. Hundreds of community servers. Extensible without engineering work.
Agents do not chat — they propose actions and wait. The agent identifies the right action, builds the payload, shows the user a full preview card, and waits for explicit approval. Nothing auto-executes. Every approval lands in the audit trail and contributes to the AI Index.
"I will send the appointment letter to 15 clients. To: 15 recipients · Subject: Q2 tax appointment reminder · Body preview below." Action stays unsigned and unsent until the user clicks approve.
Full preview of the action with all details visible. Approve, reject, or edit before sending. Reviewer can drill into recipient list, attached files, or the underlying data the agent pulled.
"15 emails sent · view details in compliance log." The audit trail records what was approved, by whom, against what data — with the execution result and any provider response codes.
Archipath does more than log prompts. It records whether an action was shaped by company goals, which employee-trained skills were used, what memory and documents were consulted, whether a human accepted or edited the result, and where evidence is complete or missing. AI Impact separates adoption, reviewed effectiveness and comparison evidence; company context supplied to a model does not establish strategic alignment.
Agent actions can point back to the goals and standards that guided the output.
Approved skills show up in the proof trail when they influence future work.
Approvals and edits become quality signal instead of disappearing into chat history.
Skill reuse, acceptance, alignment, audit coverage, and skill health roll up over time.
Upload firm documents and they are chunked, embedded, and indexed in seconds. The top-k relevant chunks are auto-injected into every conversation. The model cites which documents it used. In most cases, names, phone numbers, email addresses, Social Security numbers, and card and account numbers are masked before your company's prompts reach the AI provider.
Drag and drop. Documents are processed and indexed in seconds. No manual tagging required.
The top-k relevant chunks are injected into every agent conversation. The agent cites which documents it used in the response.
In most cases, names, phone numbers, email addresses, Social Security numbers, and card and account numbers are masked before your company's prompts reach the AI provider.
Per-client, per-matter, per-project knowledge scopes that conversation memory persists into. Cross-engagement contamination is impossible by design — critical for conflict-of-interest protection in legal and financial firms. When an engagement ends, the container is purgeable while the audit trail (what happened, not what was said) remains.
Kyro22 provisions the infrastructure. You add your provider API keys and start configuring. The platform is live the day you sign. The first measurement period establishes actual usage, evidence coverage and suitable comparison baselines. Savings require reviewed outcomes and financial evidence; they are not guaranteed within a fixed period.
Add provider API keys (Anthropic, OpenAI, Google, AWS Bedrock). Configure the first connector. Platform is live.
Configure professional roles and tier floors. Add firm-specific terms to the blocklist. Review the first compliance logs. Upload firm documents to the knowledge base.
Context containers accumulate client knowledge. Usage and evidence coverage become visible. AI Impact compares reviewed work with documented baselines where the installed release supports it. Recorded time differences do not establish cash savings or causal attribution.
Security and trust details, including what is mapped and what is not audited, are on a separate page. Security & trust →
Security controls documented and mapped. Audit-ready documentation for compliance teams.
Trust Service Criteria documented across availability, security, processing integrity, confidentiality, and privacy.
All sensitive data encrypted at rest. Keys managed via AWS Secrets Manager or local configuration.
Strict transport security, content security policy, X-Frame-Options DENY, X-Content-Type-Options, referrer policy enforced on every response.
Multi-tenant architecture with strict isolation. No cross-tenant data access. Context containers enforce isolation at the application layer.
Routing, batch processing, safety queue, memory, connectors, and security controls all covered by automated test suites.
Built-in safety rules that cannot be bypassed: malware, impersonation, jailbreak, AI-disclosure, advice disclaimers. Plus unlimited custom rules.
Prompts are sent only through provider APIs whose business terms prohibit training on your data.
Routing, batch processing, safety queue, memory, connectors, and security controls are covered by automated tests that run on every change.
Firms can add their own safety rules. Archipath does not claim that built-in rules cannot be bypassed.
30 minutes. We walk through routing, redaction, action approval, proof trails, and the AI Index on a live deployment. You bring the questions.