Multi-modal design.
Documents, images, and web artifacts, created and ingested. Twelve systems run behind every answer. Select one, or let it play.
Knowledge first
Your indexed corpus is the primary source of truth. Live systems through API and MCP connectors come next, then continuously refreshed licensed sources, then gated web research. Model knowledge is the last resort and is labeled when used.
Engineered specialists
Each bot, such as Tax, Audit, Wealth, Client Accounting, M&A, and IT, runs on its own engineered system prompt that defines its persona, methodology, behavior, process, and guardrails for that discipline. Prompts are managed centrally, versioned, and tested against scenario cases before release.
Routines and Skills
Routines are scheduled, multi-step jobs that run without a user present. Skills are versioned instruction sets that encode a method; they are shared across the firm and attach automatically when a turn needs them.
Multi-model routing
Each turn is classified before any tokens are spent. Lower-cost models gather evidence in parallel, and a frontier model writes the answer. Models from Anthropic and OpenAI run inside the Azure tenant.
Documents in and out
Uploads are read by more than 47 vision and OCR endpoints, so models search extracted text instead of spending tokens on pages. The document engine creates, fills, and edits Word, Excel, PowerPoint, and PDF files in place, on your templates.
Memory that stays small
Long conversations fold into a rolling summary rather than being re-sent. The Second Brain carries what matters between sessions and shapes each answer in the user's own voice.
Image generation
Brand-constrained images stream into the conversation and drop straight into the document being built.
Channels
The same specialists and memory in the web app, Microsoft Teams, iOS, and the MegaSnipper add-ins for Word, Excel, PowerPoint, and Outlook, with SharePoint, OneDrive, mail, and Teams integration.
M&A Diligence is responding to bothCollaboration
Projects hold conversations, documents, instructions, and skills for a matter. Chats can be shared read-only, handed off, or joined live.
Administration
One console for users, specialists, indexes, skills, token budgets, conversation audits, and usage telemetry, gated at the edge and in the application.
Modular by design
Specialists, tools, connectors, and channels are separate services behind one gateway, so capabilities are added or removed through configuration.
Voice
Three voice modes work with every specialist: Conversation for real back-and-forth, Dictate to speak instead of type, and Speak only to hear answers read aloud as they stream. Custom voices through ElevenLabs are fast and responsive, and can use any voice you license or record, including one for your brand.
AutoPilot. Ask in chat, and it does the work on your machine.
From any specialist, a user can hand off a task such as installing and configuring software or mapping a drive. AutoPilot carries it out on their own device through the SAXTech Resolve agent, narrating each step in the chat and asking when it needs an answer.
Sees the screen
Reads what is on screen before every action.
Plans and acts
Clicks, types, and opens apps like a person would.
Checks its work
Verifies each step, retries or escalates on failure.
Asks your specialists
Pulls approved settings and firm knowledge mid-task.
✓ Managed devices only✓ Never enters credentials✓ Asks before big moves✓ Allow and deny lists✓ Full action log✓ Human in the loop
Built for your practice areas during implementation.
Specialists, prompts, skills, integrations, and branding are configured to your firm. The platform underneath stays the same, which is why it can be maintained, audited, and upgraded as one system.