Product Roadmap

WebSearch improvements: upgrade to enterprise grade and high quality. Includes (not exhaustive):

• Vertex AI to GA

• PII: Automatic Personal Data Removal

All

• "Automatic Personal Data Removal" feature as part of the Elicitation Check

• Deploy Vertex AI for GA

General Ingestion (including Agentic Ingestion)

All

Improve speed and quality of ingestion. To increase accuracy on responses by better "chunking" documents and sending them to different models

Significant Ingestions Improvement

• MDI + Vision quality and stability improvements

• Error Handling

• Failed Ingestions

• Docling/Alternative MDI

• Speed

Operational Security & Compliance

All

We promise security and compliance to our customers in our contracts, like fixing vulnerabilities based on severity within a given timeframe. Or not shipping images to customers that contain high severity vulnerabilities.
We want to be in compliance with requirements from ISO 9001, 27001, 42001 and our SOC2 controls to keep our certifications, to base our business relationship with our customers on trust.

MCP v1.0 [Spaces 2.3]  

Team Lead / Admin

Connect MCP sources to a chat space and an Agentic table

Unique MCP Hub 

All Users

Standalone middleware client compontent to sell to financial services institution to implement MCP in a secure and compliant way.  This is a critical stratic innovation

KYC SOW (POC)

All

A case-based workspace that helps compliance/AML & RM/front teams complete Source of Wealth assessments faster and with higher audit quality

Code Execution

All

Quants and portfolio managers want to use python and their own code for data analytics

Unique AI Conduct (Agent Harness)

All

The ability to swap out code intelligence so that skills that are created in Unique AI can be used by any code execution and skill provider (Anthropic, Open AI, Google, etc..)

Make LiteLLM Production Ready

All

Easily connect any model

Unique AI Academy

Create an educational program with certification for Internal employees, clients and partners to be a certified on setting up Unique and developing Agents

Zero‑downtime capability when releasing a new version

establishing zero‑downtime release capability to avoid after‑hours deploys and service disruption.

Enable User Model Switching in Chat

All

Allow users to switch the AI model within a chat to match varying task complexity and align with client platform expectations.

Answers UI Refresh

All

Concise Reference in Chat response (not full page)

Enhanced Access Control System

Develop and implement a Policy-Based Access Control (PBAC) system that combines Role-Based Access Control (RBAC) and Attribute-Based Access Control (ABAC) to provide granular, flexible, and scalable access control for Unique AI's platform. T

Widget UI/Answers refresh + MCP support epic description

the Widget is to feature an updated chat UI, a refreshed answers panel, and full MCP support — so that the widget experience is consistent with the broader platform improvements and can interact with MCP tools, handle authentication, and manage MCP connections.

PublicAPIs & SDK

Developers

Open up public APIs for developers to build solutions on Unique AI in their own environments

Commenting

All

ability to leave contextual notes, tag colleagues, and manage comment threads directly within cells — without modifying underlying table data

Unique AI and Spaces 2.0 Rollout and Landing

(70% of clients using Unique AI, 20% of MAU).

Refactoring Hallucination Check

Invest time in improving the overall output quality of Unique AI. This includes  re-optimizing prompts for the latest deployed models, establishing a scalable approach to maintain optimized prompts across different models, as well as general optimizations of the orchestrator (e.g. history management, tool loading, improve speed). This work is critical to ensuring consistent, high-quality responses.

Outbound Access Control

Infrastructure Risk Mitigation

Implement Overall Backup Policy

Infrastructure Risk Mitigation

Platform Operations & Continuous Maintenance

Infrastructure

Connectors / MCP Servers (connector stack rank managed by connector team)

All

Sharepoint V2, Confluence V2, Outlook, Factset, S&P etc…

Workflow and scheduling engine integration

Users require workflow and scheduling capability to increase efficiency and productivitiy

Voice (Speech to text)

Voice - make use of uniqueAI through voice only. Similar to ChatGPT, where users can hold a conversation without typing or reading (e.g. while driving a car). 

Spaces 2.4 - Generic Agentic Table and Setup  (Milestone 4)

Team Lead/Admin

Setup flow for a generalized agentic table

File level access control

All

Needed as a pre-requisite for Sharepoint V2 and other integrations

SWOT Agent

Users

As an equity investor, I want based on research papers, earnings calls & conference call transcripts and web search prepare an comprehensive SWOT analysis result (see example output attached) and share this with my partner / investment committee. The outcome of this SWOT analysis will be used for investment decisions.

Establish and enforce limits:

• LLM Token Limit setting

• rate limits on Public APIs and API Keys

Portfolio Managers

Limit heavy users to ensure resources and prices are properly managed. Currently unique lacks any form of (usage) metrics for both end users, their administrators or infra teams as well as for Unique itself

Enable UniqueAI to work with non-streaming models GPT-5-Pro

UniqueAI does not yet support non-streaming models such as GPT-5-Pro. To enable this, the logic of the orchestrator has to be adjusted

Imersive Agentic Finance Portal

Identify key user FS user challenges (where we are currenlty working) and bring together in a compelling visual formal

PDF paragraph highlighting 

All

Highlight only the source paragraph in the PDF rather than the whole chunk which currently results in a whole page or more being highlited. This makes it hard for the user to validate the accuracy of the statement. 

Metadata Sorting: Solving the recency (model) problem

All

Across all functions being able to sort and filter based on date

Metrics and Analytics

All

Clients see and report internally on usage. Unqiue can improve products

RAG Connectors / Knowledge Base / Ingestion

Enable a root admin role who can see hidden scopes/folders (not files or content)

Process MSG files in RAG (KB and Upload&Ask)

All Users

In order to be also able to work with content from Mails out of Outlook (Project Data, Deal Information etc) it is valuable for our customers to be able to process .msg files (as this is the format mails are saved when they get safed directly from Outlook)

Knowledge Base UI refresh

Admins

"Drag and Drop"-Functionality in the Knowledge Base. They would like to have a behavior as we know it from our every-day work with folders, to move them via dragging and dropping. 

Enable “Canvas Mode” Editing for Long Responses in Chat

All users

Enable a “Canvas mode” (similar to ChatGPT’s editing interface) that allows users to open very long AI-generated responses in an editable canvas, make updates, and then send the revised content back into the chat.

Improve error messages and in app alerting

Admins, users

"The current issue reported by users is that banners, such as those displayed during outages, are often dismissed without being read. This leads to unnecessary questions and escalations, as users either miss critical information or remain unaware of whether the issue is still ongoing.

Requested Solution: During banner set up there should be a flag to set if banner should be dismissible by users (as today) or non-dismissible  meaning the banner would stay visible as long as the admin removes the banner."

Unique AI Usability improvement:  Modularize Unique AI.

• Upload to chat UI fixes

• Make context window independent and configurable from the tool. (i.e. Break hardcode dependency of "Select files to chat" on "Internal Knowledge search" tool.  )

• Enable custom tools

• Enable select files for custom tools

Additional needs:

• Create new chat and bring over the current  Select Files to Chat settings. 

• Prompt to users to create new chat  to avoid hitting token and memory issues.  

All

Feedback suggests that many users remain in the same chat.  Help users understand that it is better to create new chats and help them easily create new chats with the same configuration of Select Files to Chat

Basic query output into Word or Excel. Output template creator (Powerpoint and Word).  Pre configured templates

RM, Portfolio Managers

Easy exporting of AI output into a document

Spaces 3.0 - Templates, Users Selected Models, User Spaces (tbc)

A user can upload documents into their own space and prompt against those documents

Zitadel role mapping

Admin

The goal is that the client can define via their Groups who gets more permissions on Zitadel (e.g.: members of the INFRA_ADMIN group get the IAM_OWNER permission) and do not have to manually manage these permissions in Zitadel itself.

Get user to first high quality prompt fast with high quality answers

User research shows that stickiness increases when a user gets to a first high quality prompt in the first minutes of interaction…and multiple high quality prompts in the first days

Unique AI & Agent Ready Benchmarking and Evals (Golden Set of Evals)

Managers

To show value of investment  to Unique Clients 

Space configuration versioning and history

All User

Introduce versioned UniqueAI configs and the ability to revert to any prior version. 

User Management UI improvements

Managers

Encrypted self-service configuration and secret store in Unique product

Users, Admin Users, Administrators, Unique Staff

 encrypted configuration store both for values and secrets where surely administrators (of the product) or even end users can add their own secrets without involving infrastructure teams.

KYC Recertification

Market Insights Email Agent

Select Files to Chat interface meta data filter

All

Earnings Call Agent

RM

Identification of the correct stock

Improve token expiration issue

End Users

In the new teams integration we have the issue that they should on purpose not use our Recording UI. Therefore, a background service registers the users with their credentials in our service. There are several reasons why Microsoft can invalidate this login token (device change, password reset, email address change, etc.) and then the token is invalid and the user has no transcript, report, etc. without knowing it. There is no UI that can tell him to relogin. We can improve this by introducing a scheduled check if the tokens are valid and if not send an email with a warning asking the user to relogin.

Retry mechanism for transcript retrieval

End Users

Currently, there is no retry mechanism in place, which creates challenges when Microsoft unexpectedly prevents us from retrieving the transcript from Teams. This issue has already occurred for the client and was resolved only after manually retriggering the process. This approach is far from ideal, as it results in the user not receiving the report in their CRM. Consequently, the user reports the issue to the client’s project team, who then escalate it to us. We investigate, manually retrigger the process, and inform the client to check again. This workflow leads to a poor user experience and unnecessary manual effort for all parties involved. To address this, we should implement a retry mechanism to automatically attempt to retrieve the transcript, thereby improving the success rate and streamlining the process.

Global Glossary for Centralized Term Management and Tone of voice

All users across spaces within the system

Ensures consistent understanding of company-specific terms and products by providing centralized term management, automatic recognition, and easy access to definitions, potentially supporting multiple languages.

Prompt Library/Sharing

RM

Users can share prompts, vote on prompt and select from popular prompts to improve quality and efficiency

Direct Image Embedding

Chat against transripts

Benchmarking & Analytics: Scheduled Run

Unique

In-document Translator (PDF)

All

Upload a document and get it translated, priotiy is PDF

MP3/Wav Ingestion

RM

editing the user prompts

End Users

Bing Grounded Websearch

Vertex AI websearch