Future Enhancements

Centralpoint Future Development Roadmap

Roadmap · Next 12 months

Centralpoint has been developed and maintained for more than 25 years, with updates released on a bi-weekly cadence. While our roadmap extends three years out, we consider it reliable only for the next twelve months. Technology changes so rapidly that even planned items are frequently re-prioritized as new capabilities emerge — which is why we never look beyond a single year in our active development effort. Accordingly, we publish only the coming twelve months of planned enhancements. Over the coming year, that roadmap centers on deeper AI integration: ingesting Microsoft Teams and Microsoft 365 content into the governed index, connecting the AI assistant live to back-office systems under secure role-based access, agentic multi-step workflows, multimodal vector indexing, and expanded on-premise AI governance and cost controls. For more information, please see the planned enhancements below.

This year's focus

Microsoft Teams & 365 ingestion Live back-office access Agentic multi-step workflows Multimodal vector indexing On-premise AI governance & cost controls

25 years of evolution

Our heritage of data inferencing

More than two decades of shipping enhancements every two weeks — the foundation of governed data that today's AI inferencing is built on. See the full release history behind the roadmap.

View Our Heritage

Microsoft Teams Ingestion & Meeting Intelligence

Microsoft Teams has quietly become an enterprise silo of its own. Meetings, decisions, and attachments live there and nowhere else. This release brings Teams into Centralpoint as a first-class governed source. A scheduled connector ingests each meeting's transcript, start and end time, duration, attendee list, and any shared attachments, writes them to a records module, and enrolls that content in the hybrid vector index. Once indexed, meeting knowledge becomes answerable through the Knowledge Assistant and Page Assistant under the same audience, role, and sensitivity rules that govern every other record. Ask 'what did we commit to in last week's steering call?' and get a cited answer drawn from the transcript, without the content ever leaving your environment. Pre-index sensitivity filtering runs before embedding, so restricted discussions stay restricted.

Fri, Aug 14, 2026

Live Back-Office Connectors (Governed AI Data Bridge)

Now that DataSources feed directly through the LLM assistant, the next step is reaching past Centralpoint's own records into the systems that run the business. This release introduces governed back-office connectors. A DataSource can query an ERP, HRIS, financial, or ticketing system in real time and render the result, a live chart, table, or record, directly inside the Page Assistant reply. Access is enforced by the asking user's roles and audiences, so a finance user sees live payables while a field user sees only what their clearance allows. Nothing is copied or retained beyond policy. The assistant simply becomes a governed, plain-language window onto live operational data, on demand.

Mon, Sep 14, 2026

Microsoft 365 Unified Ingestion (Outlook, SharePoint, OneDrive)

Building on Teams ingestion, this release widens the net across the Microsoft 365 estate. Scheduled delta transfers bring Outlook mail threads, SharePoint libraries, and OneDrive documents into the governed index, capturing only what changed since the last run to keep sync light. Records are de-duplicated across sources, so a document living in three places is answered as one. As with every source, pre-index sensitivity filtering and per-record enrollment control decide what is embedded and what is held back. The result is a single governed knowledge surface that spans the tools your teams already use, searchable, answerable, and fully inside your environment.

Wed, Oct 14, 2026

Agentic Workflows - Multi-Step Governed AI Actions

The AI Prompt Manager already supports Following chains that run one prompt after another. This release surfaces that capability as end-user agentic workflows. A request such as 'draft the vendor summary, check it against policy, and route it for approval' becomes a governed sequence of steps, each with its own skills, token budget, and audit entry. Human-in-the-loop gates pause the chain for review before any consequential action, and every step is logged for accountability. Because each node runs under the same governance envelope, PII redaction, brand and compliance guardrails, and token metering, automation never outruns oversight.

Sun, Nov 15, 2026

AI Governance & Cost Dashboard

As governed AI moves into daily use, administrators need to see it clearly. This release delivers a single dashboard that reports token consumption by prompt, model, and audience; cache hit rates and the spend they save; which skills are loading and how often; and every PII-redaction and guardrail event. Budgets and alert thresholds are managed in the same place, so a team approaching its monthly limit is flagged before it is exceeded. The dashboard turns the platform's existing metering and audit logs into a live operational picture, proof that AI is both useful and under control.

Mon, Dec 14, 2026

Voice-Enabled Conversational Assistant

This release adds speech to the Page Assistant. Users can ask questions out loud and hear spoken, cited responses, with on-premise speech-to-text and text-to-speech options for environments that cannot send audio to the cloud. Voice interaction is grounded in the same governed vector index and honors the same role and audience rules as text chat, so nothing is spoken that the user could not already read. Hands-free access opens the assistant to field, floor, and accessibility use cases without loosening a single governance control.

Sun, Jan 17, 2027

Multimodal Vector Indexing (Images, Diagrams, Scans, CAD)

Centralpoint's hybrid index has always understood text. This release teaches it to see. Images, diagrams, scanned documents, and CAD drawings are processed through OCR and image embeddings and folded into the same vector space as written content, so a search or an assistant question can surface a relevant figure, floor plan, or scanned form, not just prose. Multimodal content flows through the same enrollment, sensitivity, and security controls as everything else. Knowledge that used to be locked inside a picture becomes findable and answerable.

Sun, Feb 14, 2027

Guided Skill Authoring & SKILLSHARE Library

The AI Skill Manager holds the reusable capabilities the assistant loads at runtime. This release makes those skills easy to build and share. A guided authoring experience walks an administrator through defining a skill's purpose, retrieval bucket, token budget, and guardrails, validating it before it goes live. A curated SKILLSHARE library lets proven skills be published and reused across sites and tenants under governance, so hard-won expertise is captured once and spread, never rebuilt from scratch.

Sun, Mar 14, 2027

Retrieval Re-Ranking & Relevance Feedback

Vector recall is fast but blunt; this release adds precision. A re-ranking pass reorders the top candidate chunks before they reach the model, lifting the most relevant material to the top of the context window. A lightweight feedback loop, a thumbs up or down on an answer, feeds signal back into tuning, so retrieval improves with use. The result is fewer near-misses, tighter citations, and answers that stay grounded in the best available source rather than merely the first one matched.

Sun, Apr 18, 2027

Governed Assistant Memory & 360 Continuity

Today each conversation starts fresh. This release introduces governed memory. With the user's context and permission, the assistant can carry a 360-degree profile and prior-session continuity forward, so a returning user is greeted with relevant history rather than a blank slate. Memory is scoped to the individual, held inside the environment, and subject to retention and redaction policy like any other governed data. Continuity makes the assistant feel like a colleague who remembers, without ever becoming a system of record it was not authorized to be.

Sun, May 16, 2027

Pre-Index Sensitivity & DLP Classifier

Governance is strongest when it happens before content is ever embedded. This release adds an automated sensitivity classifier to the ingestion pipeline. As records arrive from any source, they are scored for PII, confidentiality, and policy-restricted content, and the platform enforces the matching action, redact, hold, or exclude, before a single vector is written. Administrators tune the policy; the pipeline applies it consistently across Teams, Microsoft 365, and every other source. Sensitive material is handled correctly at the door, not discovered after the fact.

Sun, Jun 20, 2027

Model Evaluation & A/B Test Center

Choosing a model should be evidence-based. This release adds an evaluation center where administrators run the same question set against multiple providers and models, on-premise and cloud, and score the results on accuracy, grounding, latency, and cost against the organization's own data. A/B comparisons are repeatable and logged, so a model choice can be justified and revisited as the field moves. Because the platform is already model-agnostic, switching the winner into production is a configuration change, not a rebuild.

Sun, Aug 15, 2027

External AI Egress Governance (Interstitials 2.0)

Interstitials already let a Centralpoint assistant live on an external site with one line of JavaScript. This release hardens that egress path for scale. Per-origin policy controls which sites may host an assistant, what it may answer, and how much it may spend, with rate and budget limits enforced at the edge. Every off-platform conversation is logged back to the governed environment for audit. The assistant can go anywhere the business needs it, while governance stays firmly at home.

Sun, Sep 19, 2027

Sovereign / Air-Gapped AI Deployment Profile

For the most sensitive environments, this release packages a sovereign deployment profile: the complete governed AI stack running with no outbound connection at all. On-premise language models and local embeddings handle generation and retrieval entirely inside the client's boundary, with cloud providers disabled by policy rather than convention. Every capability, Knowledge Assistant, Page Assistant, skills, prompts, metering, and audit, operates unchanged. Organizations that cannot send a single token outside their walls get the full platform without exception.

Sun, Nov 14, 2027

Autonomous Governed Agents

The horizon of governed AI is delegation. This release introduces long-running agents that monitor Data Triggers and act on their own within a strict policy envelope, escalating an exception, drafting a filing, opening a ticket, or notifying an owner when conditions are met. Agents inherit the same guardrails, budgets, approval gates, and audit trail as every other governed prompt, so autonomy never means opacity. The organization defines the boundaries; the agent works inside them, tirelessly, and shows its work.

Sun, Jan 16, 2028

Zero-Config RAG Autopilot

This is the automation of RAG for any organization. Point Centralpoint at a repository, a share, a database, a mailbox, and the platform does the rest: it discovers the content, classifies its sensitivity, selects chunk sizes and overlap, generates embeddings, and stands up a working assistant, with no configuration to write. What used to be a project becomes a switch. Every automatic decision is recorded and adjustable, so governance is never traded for convenience. The result is a RAG deployment that any organization can turn on in an afternoon, not a quarter, while every guardrail, audience rule, and retention policy applies from the first indexed record.

Sun, Mar 12, 2028

Self-Tuning Retrieval Engine

Retrieval stops being a set-and-forget configuration. This release lets the index watch its own performance, grounding accuracy, citation quality, and user feedback, and adjust chunk size, overlap, similarity thresholds, and re-ranking weights per module on its own. When a content set shifts, retrieval re-tunes to match, without an administrator touching a slider. Every adjustment is logged and reversible, and administrators can pin any setting they want to control by hand. The governance question expands from 'who can see the data' to 'is the retrieval itself behaving,' and the platform answers it continuously.

Sun, May 14, 2028

Governance Policy-as-Code

Governance becomes an artifact you can version, review, and prove. This release expresses access, redaction, retention, and guardrail rules as declarative policy that is checked into history and applied uniformly across the Knowledge Assistant, Page Assistant, ingestion pipeline, and every connector. A change to policy is a reviewable event with an author and a date, not a quiet setting buried in a console. Roll a policy forward, roll it back, or compare two versions to see exactly what changed and when. When an auditor asks how the organization governs its AI, the answer is a document, not a description.

Sun, Jul 16, 2028

Continuous Compliance Attestation

Compliance shifts from an annual scramble to a continuous state. This release has the platform continuously check itself against the frameworks that matter to the organization, generating living attestation reports that show, at any moment, that access, redaction, retention, and accessibility policies are being met. When configuration drifts away from policy, the drift is flagged and routed for correction before it becomes a finding. The evidence auditors ask for is produced as a by-product of normal operation, timestamped and traceable, rather than reconstructed after the fact.

Sun, Sep 17, 2028

Federated Cross-Enterprise Knowledge Mesh

The envelope of governance widens past a single organization. This vision lets separate Centralpoint environments, subsidiaries, partners, agencies, form a governed knowledge mesh in which a question can span nodes while each node's data never leaves home. Retrieval is federated: the mesh routes a query to the nodes entitled to answer it, applies each node's own access and sensitivity rules, and returns cited results without centralizing a single record. Sovereignty and collaboration stop being a trade-off. Each member keeps control of its own data and still benefits from the collective knowledge of the whole.

Sun, Nov 19, 2028

Explainable AI Provenance Ledger

Governance you can prove, answer by answer. This vision attaches a tamper-evident provenance record to every AI response: exactly which chunks were retrieved, which model and version produced the text, which policy version was in force, and which user asked, written to an immutable ledger. Any answer the assistant ever gave can be reconstructed and defended months later. When a decision was informed by AI, the organization can show precisely what the AI saw, what it said, and under what rules, turning explainability from a promise into a receipt.

Sun, Feb 18, 2029

Autonomous Data Steward

Governance turns inward, onto the data itself. This vision introduces an autonomous steward that continuously watches the governed corpus for staleness, duplication, orphaned records, and taxonomy drift, then proposes or, within policy, performs the housekeeping that keeps an index trustworthy. It flags content past its retention window, reconciles duplicates across sources, and keeps classifications consistent, so the knowledge the assistant draws on stays clean without a person auditing it by hand. A well-governed answer depends on well-governed data, and the steward makes that upkeep continuous rather than occasional.

Mon, Apr 16, 2029

Real-Time Bias & Drift Monitoring

Governing an AI means watching how it behaves, not just what it can access. This vision continuously samples the assistant's outputs for bias, hallucination rate, and drift from expected behavior, and raises an alert, or automatically quarantines a model or skill, when quality falls outside policy. A provider that degrades after an upstream change is caught in hours, not by a user's bad experience weeks later. The organization sets the tolerances; the platform enforces them around the clock, so a governed assistant stays a trustworthy one.

Mon, Jun 18, 2029

Adaptive Intent-Aware Guardrails

Guardrails grow up. Today's rules match keywords and patterns; this vision lets them understand context and intent, so policy responds to what a user is actually trying to do, weighed against their role and the sensitivity of the moment. The same question can be answered fully for one user, redacted for another, and gently declined for a third, based on situation rather than a static list. Governance becomes proportionate: firm where risk is real, frictionless where it is not, and always explainable after the fact.

Mon, Sep 17, 2029

Organizational Digital Twin (Graph + RAG Fusion)

The governed index becomes a living model of the organization. This vision fuses a knowledge graph, people, systems, contracts, assets, decisions, and the relationships among them, with the existing vector index, so the assistant can reason over how things connect, not just which document mentions them. Ask how a policy change ripples across departments, or which contracts touch a given vendor, and get an answer grounded in structure as well as text. The organization gains a queryable digital twin of itself, governed by the same rules as every other record.

Sun, Dec 16, 2029

Portable Governance Packages

Governance becomes a transferable asset. This vision lets an organization export a complete governed AI configuration, its skills, prompts, policies, taxonomy, and index schema, as a signed, portable package that another Centralpoint instance can adopt intact. A proven governance posture built for one site, agency, or division can be handed to the next and stood up in hours, with its provenance and integrity verifiable on import. Hard-won governance stops being trapped in one environment and becomes something an organization can standardize, share, and inherit.

Sun, Mar 17, 2030

Universal Connector Fabric

The endgame of 'any organization, any system.' This vision turns any system with an API into a governed Centralpoint source or action target through a self-describing connector fabric, with no bespoke integration to write. The fabric reads a system's own description, maps it into the governed model, and exposes it to the assistant as live data to read or an action to take, all under the same role, audience, and audit controls as native content. Governed AI stops being limited to the data you remembered to import and reaches, safely, across the whole estate.

Sun, Jun 16, 2030