ReAct Maia: From Answer Bot to AI Agent

Mental-health clinicians already juggle progress notes, scheduling, payment follow-ups and resource hunting before they greet a client. Last year, we launched Maia, a virtual assistant that answered therapists’ questions with retrieval-augmented generation (RAG). About half of the hundreds of professionals who run their practices on Therapyside turned the feature on - they loved it and quickly asked for deeper help with calendars, payments and client prep.

Today, Maia can ReAct, automatically picking the right API or internal tool to lighten that admin load.

From Retrieval to Reasoning

Across the industry, chatbots are graduating to reasoning AI agents that can plan, decide and execute - and Maia now joins those ranks.

“Our early RAG assistant was a hit, but therapists wanted Maia to roll up her sleeves,” says Alessandro De Sario, our co-founder and CEO. “We turned that feedback into an agent that handles real workflow tasks so clinicians can stay focused on care.”

How the New Maia Agent Works

Ask Maia anything - “Reschedule Maria’s session,” “Who hasn’t renewed?,” “Suggest CBT worksheets for Javier.” Behind the scenes, the agent decides which tool to invoke:
‍

● Calendar access - add, move or summarize upcoming sessions.            

● Subscription & payment checks - flag expiring packages and suggest new plan options.        

● Session-note retrieval - surface any detail from previous visits.            

● Science-based resource lookup - deliver articles, worksheets and infographics tailored to therapy goals.

A single chat reply merges those pieces into a day-at-a-glance report or an instant, context-rich answer. If the question is open-ended, Maia falls back on RAG - but now it can also take action.

“Maia can shuffle appointments, cue subscription reminders and drop evidence-based resources right into the chat - all without leaving our platform,” De Sario adds.

NVIDIA Tech Under the Hood

We built Maia’s reason in gloop on the open-source NVIDIA Agent Intelligence toolkit (AIQ), which orchestrates tool calls and error handling. Our language models and NVIDIA NeMo Retriever embedding services run as NVIDIA NIM microservices, giving us up to 3x faster inference on GPUs. Internal APIs - calendar, billing, resource DB - are exposed safely through MCP auto-generated schemas.

Early Impact - and What’s Next

Therapists using Maia’s original RAG assistant save up to 15 minutes per client on note-taking and information lookup, and average Net Promoter Score¹ jumped 25 percent because sessions shifted from paperwork to patient focus.

In early pilots, agentic Maia cuts about 40 percent more admin time - 20 - 22 minutes per client day - by prepping resources, juggling calendars and flagging expiring subscriptions. Quality should climb, too: over 90 percent of clinicians already rely on external science-based materials, and Maia now surfaces them in context for richer, evidence-aligned care.

Coming Soon: Global Doctors SaaS (H2 2025)

The same agentic technology will soon power Global Doctors by Therapyside, a SaaS platform for nutritionists, psychiatrists, physiotherapists and other health professionals. Launching in the second half of 2025, Global Doctors ships with Maia’s full AI toolkit - plus every practice-management feature we’ve refined over the years.
‍

“Admin drag isn’t unique to therapy,” De Sario notes. “By packaging Maia as a service, we give every clinician a tireless digital teammate.”

See Maia in Action

Heading to VivaTech Paris / NVIDIA GTC Paris in June? Visit our booth in the Inception Startup Pavilion at GTC Paris and watch Maia orchestrate a live clinical workflow.

Can't attend the event?

‍[.test-button]Email us[.test-button]

‍Therapyside is proud to be part of the NVIDIA Inception program for startups.

‍
1. Net Promoter Score (NPS) is a common customer-satisfaction metric that ask susers, on a 0-10 scale, how likely they are to recommend a product or service.Respondents scoring 9-10 are “Promoters,” 7-8 are “Passives,” and 0-6 are“Detractors.” The score is calculated by subtracting the percentage of Detractors from the percentage of Promoters, yielding a number from -100 to +100