17-Aug-2026

Progressive Disclosure Skills — A skill can now be a library instead of a wall of text. The agent always carries a short map of what is available and opens a section only when the task calls for it, so an entire methodology can stand behind an agent without being paid for on every single turn. You decide how much of it each agent may ever open: one corpus can serve a designer and a builder while the designer is refused the implementation half outright — and the refusal names what it may read instead, so the agent corrects itself rather than guessing. The library travels inside the skill's signature, so editing any section shows up on the agent's trust badge.

16-Aug-2026

Signed Data Products — A business question now has a signed answer path: declare a data product on your ontology — the entities it exposes, the only joins it may use, its measures — and the platform machine-proves it deterministic before it can be signed, probing the declarations against the live database so a wrong cardinality or a phantom column can never be signed over. A surface-only agent literally cannot query outside its products — raw SQL is refused and the agent is steered to the governed path — and every governed answer seals a receipt into the turn's signed record: product, version, request hash, rows, verifiable offline. Products even compose: a product built from signed products signs with its lineage inside the signature, and goes stale the moment a parent changes.

08-Aug-2026

Your Vote Now Teaches the Agent — Rating an answer used to reach one place. Now every engine learns from it in its own way: an Optimize agent promotes or retires the plan behind the answer, a Focus agent flags the documents that served it, a coordinator passes your rating to the expert that produced it, and facts the agent quietly learned from an answer you rejected stop being recalled. Each vote is also recorded beside the quality scores, so a turn the scorers passed but you rejected is visible where the thresholds are tuned — and hovering a vote button now tells you exactly what it will do.

07-Aug-2026

Trust, Open to Third Parties — Assurance packs can now bring their own checks to the Trust Circle, and their own controls: a pack declares what to look for and which act it wants, and Uderia performs it. Every third-party control arrives dormant, so installing one can never change an answer until you raise it — and a pack must be signed before it can be assigned to anything.

07-Aug-2026

The Trust Panel, Rebuilt — Configuration → Trust is now the Trust Circle itself: pick a guarantee and see what acts on it and what speaks to it. Everything is saved together by one button, and the live dial in the status window now fills in as the turn runs instead of arriving all at once.

05-Aug-2026

Controls That Act Sooner — Three checks stopped being reports and became protections: an agent that keeps making the same call is stopped mid-run instead of graded afterwards, a per-run budget is enforced before the tokens are spent rather than flagged once they are gone, and a fluent answer with nothing to cite is now caught by the grounding gate. Each is off until you set its threshold.

05-Aug-2026

One Home Per Measurement — Every scored check is now counted once, at JUDGED, and grouped by the moment of the turn it measures: Plan, Retrieve, Act, Compose, Deliver. The guarantee stations report what their own machinery did during the turn, so a gate doing its job no longer turns red because a score arrived afterwards.

19-Jul-2026

One Design Language — Every panel now speaks the same visual language: a KPI command center on top, a draggable divider spanning the whole panel, and clean tab strips below — across Anatomy, Trust, Awareness, Intelligence, Components and Ontology. The divider learned gestures: shift-click collapses the upper pane completely, double-click restores it maximized, and your last split is remembered. Panels now answer to their navigator names — Anatomy, Trust, Awareness, Intelligence, Marketplace — and the Ontology gained live estate KPIs, a composition chart, and a click-to-locate insights list.

19-Jul-2026

The Agent Anatomy — Configuration now opens as the agent itself: the five faculties and two foundations of every agent drawn as a living star, inside the trust membrane. Pick an agent and every part shows its real configuration; click any part to land exactly where it is configured. A coordinator reveals its experts as a constellation — a star made of stars — and every profile editor carries its anatomy at a glance.

19-Jul-2026

The Trust Circle — The closed trust loop is now a living instrument inside the platform: eight stations around your agent, every one a live reading, pulsing as real turns pass the gates. Click any station to operate it, press play to hear the life of an agent narrated from its own sealed records — and see how many times the loop has closed. A miniature of the circle now travels with every answer, every roster card, and the chat header.

19-Jul-2026

From Finding to Chunk — A repository security finding is now traceable to the exact text that caused it: the chunk browser marks suspicious chunks with their problem tags, a switch filters the whole corpus to just those chunks, and one click exports the findings as a report. And a benign finding can be acknowledged with a recorded reason — a human judgement frozen into the corpus signature.

19-Jul-2026

Knowledge Answers, Grounded — An answer built from your documents now proves it stayed inside them. Unsupported numbers, quotes and names are caught before delivery — after a second-chance search of the corpus so a real fact is never wrongly blamed — and marked right in the answer text. And when the knowledge base simply doesn't cover a question, the agent says so honestly instead of inventing.

19-Jul-2026

A Screened, Governed Knowledge Base — Every document is screened as it enters a repository (hidden instructions, personal data, exfiltration links), a corpus can only be signed once it is fully screened, and a poisoned document is one click from reversible quarantine. Every answer shows its sources' integrity, freshness, and sensitivity — and a statistical monitor spots documents crafted to hijack retrieval.

19-Jul-2026

Proven, Not Promised — The knowledge-lane protections come with a published benchmark anyone can reproduce offline at zero cost: 0% false alarms, honest misses listed by name. Pair it with live red-team probes that plant real poisoned documents in a test collection and verify the defenses actually held — payloads cleaned up and the cleanup reported.

18-Jul-2026

Trust, Made Visible — See trust, don't just take it on faith. An agent's declared data scope is drawn right on its knowledge graph, with any out-of-scope drift shown crossing the boundary in red. Every answer's sealed provenance appears as a verifiable chain you can check offline, and every step of a coordinator's work carries its own trust badge.

16-Jul-2026

Test Your Own Defenses — Probe any agent with a live attack corpus and see whether its guards actually held — prompt injection, personal-data leakage, scope drift — then export a signed, offline-verifiable audit pack that maps your posture to EU AI Act, NIST AI RMF, and GDPR controls.

15-Jul-2026

Injection Firewalls — A hidden instruction can't hijack the agent through the back doors. Poisoned memories are quarantined, poisoned documents are defused, and poisoned tool results are flagged — closing every indirect path that a filter on the user's prompt would miss.

13-Jul-2026

Human Review — A new safeguard that withholds a risky answer exactly like enforcement, but keeps the original for a person to approve or uphold. Nothing is lost, nothing is silently delivered — ideal for the scheduled and autonomous runs nobody was watching.

11-Jul-2026

Why Trust This Answer? — Every answer now carries a plain-language trust capsule: the sources it drew on, whether its figures came from a governed metric, what the guardrails caught, and a signed record of how it was produced — exportable as a receipt anyone can verify offline.

09-Jul-2026

Assurance Inbox & Policy Backtesting — Trust incidents become worked items in one triage queue instead of buried log lines. And before you tighten any policy, replay it over your real history to see exactly what it would have blocked or masked — so you decide on evidence, not guesswork, at zero token cost.

07-Jul-2026

Behavioral Trust Score — An agent's trust now reflects its measured conduct, not just a signature: sustained good behavior earns trust, and decline visibly withdraws it. A built-in circuit breaker can alert you — or freeze an agent's unattended use — the moment its conduct crosses the line.

05-Jul-2026

The Trust Journey — Every turn shows the full assurance pipeline it passed through — identity, scope, tool checks, guardrails, grounding, and the final seal — as one compact, glanceable strip, so you can see exactly which protections acted on any answer.

03-Jul-2026

The Trust Membrane — A tamper-evident, cryptographically-signed trust badge on every agent and every shareable asset. An agent is trusted only when everything it's built from is — and that trust survives the marketplace, so whoever installs or forks it sees the same honest, verifiable badge.

02-Jul-2026

Agent Authority — When an agent acts on its own — a scheduled task, a coordinator's expert, a step in a flow — it runs as a limited, fully-attributable version of its owner, never able to exceed the authority you granted, with every autonomous action recorded as done by the agent, not by you.

01-Jul-2026

Governed Metrics & Open Standards — Define a business metric once in your ontology and have both the AI and your BI tools consume it identically — and export your knowledge graph and metrics as recognized open W3C semantic standards for true, lock-in-free interoperability.

30-Jun-2026

Platform Ontology — A living, agent-centric map of your whole estate. Every agent, what grounds it, what it remembers, the components it can use, and all your configuration (LLMs, MCP servers, vector & graph store backends, skills, extensions, packs) become navigable nodes — built on read so it's always in sync. Explore the graph or the navigator tree, surface trust insights, and jump straight into the screen that manages any piece. Admins can span any user.

28-Jun-2026

Agent Memory — The agent remembers durable facts and preferences about you across every session — kept per-agent by default or shared globally, recalled automatically into every conversation, and fully inspectable and erasable in a dedicated read/write canvas.

28-Jun-2026

Uderia as an MCP Server — Expose any governed Uderia profile as a tool to external MCP clients (Claude Desktop, Cursor, other agents). Every answer runs the full pipeline — grounding-gated and guardrailed before it leaves — and the tool surface is hard-gated by the access token's tier.

27-Jun-2026

AI Guardrails — Built-in protection on both edges of the conversation: outbound PII redaction and toxicity blocking, plus inbound prompt-injection blocking that stops the attempt before it ever reaches the model. Deterministic, all-engine, fail-safe, governed off by default.

26-Jun-2026

Grounding Gate — Anti-hallucination, enforced. A runtime gate checks every answer against the Knowledge Graph's declared scope and can pass, annotate, or block confident-but-wrong drift — with a published benchmark (0 false negatives, governance drift 6% → 0%).

24-Jun-2026

Ontology — Ingest Databricks Industry Data Models into a two-layer (logical + physical) Knowledge Graph, then deploy the physical model to a real Teradata database (DDL, soft-RI keys, metric views, demo data) — all deterministic.

21-Jun-2026

Agent Observation & Evaluation — See and trust how every agent performs. Each turn and flow run is automatically scored on correctness, cost, and safety, and a health lens tracks quality, regressions, and cost per agent — with optional AI-judge panels for deeper review.

21-Jun-2026

Next Step Recommender — After every answer, get smart, one-tap next steps — relevant follow-up questions and actions (chart it, pin to a dashboard, build a flow) grounded in what your agent can actually do. Token-light by default, and dismissible whenever you want.

10-Jun-2026

Flow Parameters & Visual Debugging — Parameters — global variables, per-flow values, and named node outputs — turn any flow into a reusable template that serves many inputs. And you debug it visually: run a single node, or everything downstream, without re-executing the whole graph, inspecting each node's real input and output as you go.

05-Jun-2026

Sharper Self-Learning — The agent now learns only from turns that genuinely retrieved data, ignoring presentation-only follow-ups (charts, dashboards, reports) so its memory of proven approaches stays accurate and surfaces better suggestions on future queries.

17-May-2026

Dashboard Component — Pin any chart or table from the conversation to a persistent, multi-tile dashboard. Tiles refresh live via direct MCP calls or by re-running the original query, support drag-to-resize and grid reordering, and the panel reopens automatically when switching back to the session.

16-May-2026

Visual Flow Orchestration — Design and execute multi-step agent workflows through natural conversation. The agent builds a live execution graph, runs each node through the full Fusion Optimizer pipeline, and streams per-node status in real time — with support for decision branches, parallel paths, and version snapshots.

14-May-2026

New LLM Providers — MLX for Apple Silicon local inference and xAI for Grok models, expanding provider flexibility from hyperscalers to sovereign on-device execution.

12-May-2026

KG Store Abstraction Layer — Knowledge Graphs are now storage-agnostic. Start with the embedded SQLite backend and scale to Neo4j for enterprise graph workloads — without changing profiles, agent packs, or workflows.

10-May-2026

Infinity Context — RAG-powered conversation memory that never forgets. Every prior exchange is stored in a per-session vector store and retrieved by semantic relevance — giving the agent effectively unlimited context depth without destructive truncation.

09-May-2026

Model Guide — One-click access to the curated Uderia Model Guide (llm.uderia.com) directly from the LLM Configurations panel. Compare providers, recommended models, context windows, and pricing before you configure.

03-May-2026

Native Connector Architecture — The agent now acts across your entire digital workplace. Connect Google, Microsoft 365, and Slack out of the box; extend with any public connector from the registry. Each user authenticates their own accounts — no shared credentials.

02-May-2026

Always-Fresh Knowledge — Knowledge repositories stay current automatically. The agent detects what changed, re-indexes only what's new, and removes what's gone — without manual intervention across any environment.

02-May-2026

SAML 2.0 & JIT Provisioning — Connect to corporate identity providers via SAML 2.0; user roles update automatically at login based on IdP group membership.

01-May-2026

Task Scheduler — Schedule any agent query to run automatically on a cron or interval; manage schedules through conversation or the split-panel UI, with results delivered by email or webhook.

28-Apr-2026

Parallel Query Execution — Multi-step queries run faster: independent plan phases and date-range iterations now execute in parallel instead of sequentially.

16-Apr-2026

Knowledge Graphs in Agent Packs — Export and import Knowledge Graphs as part of agent packs; import restores graphs without displacing existing active graphs.

10-Apr-2026

Genie Coordinator Pass-Through — Single-expert queries skip the synthesis step, cutting coordinator token cost and response time in half.

04-Apr-2026

OpenRouter Support — Access 100+ models (Llama, Mistral, Gemma, and more) via a single API key.

15-Apr-2026

Teradata Hybrid Search — Enable BM25 hybrid search on Teradata VantageCloud knowledge repositories directly from the repository inspector.

10-Mar-2026

Hybrid Search — Choose between semantic, hybrid (RRF), and keyword (BM25) search on Qdrant Cloud knowledge repositories.

08-Mar-2026

Teradata VantageCloud Backend — Store and search knowledge in Teradata VantageCloud with server-side embedding and enterprise-grade connection resilience.

07-Mar-2026

ChromaDB & Qdrant Cloud — Zero-config local knowledge storage (ChromaDB) or managed cloud vector search (Qdrant), selectable per repository.

01-Mar-2026

Context Window Management — Control how much of your conversation history and knowledge the agent sees; configure per profile with live token budget visualization.

29-Feb-2026

Canvas Component — Write and run SQL directly in the chat canvas with syntax highlighting and live result rendering.

27-Feb-2026

Data Visualization — Turn query results into interactive charts (16 types) automatically; the agent picks the right visual for your data.

23-Feb-2026

Extensions — Transform agent responses on the fly: extract JSON, classify, summarize, generate PDFs, or chain custom transformations with a single # trigger.

22-Feb-2026

Skills - Pre-processing prompt injections with 5 built-in skills and visual editor

15-Feb-2026

Dual-Model Cost Breakdown - Live Status displays strategic vs tactical costs for Fusion Optimizer dual-model executions with color-coded visualization

14-Feb-2026

Theme-Aware Token/Cost KPIs - Live Status token and cost displays now adapt to both dark and light themes with consistent visibility

13-Feb-2026

n8n Integration - Visual workflow automation with three production-ready templates (Simple Query, Scheduled Reports, Slack Integration) and comprehensive deployment guides

08-Feb-2026

Agent Packs - Bundle complete agent teams (coordinator, experts, knowledge collections) into portable .agentpack files for one-click install, export, and marketplace sharing

01-Feb-2026

Text to Speech - Voice output now available for all profile classes with Escape-to-cancel and smart text truncation

31-Jan-2026

Document Upload & Multimodal Analysis - Attach documents and images in chat with native multimodal delivery and automatic text extraction fallback across all providers

30-Jan-2026

All Models Available - Access all provider models with ★ recommended highlights

30-Jan-2026

Light Theme - Full light mode support with proper contrast and visibility across all components

25-Jan-2026

Legacy Theme - Fixed color bleeding issues with opaque backgrounds and consistent gray palette

23-Jan-2026

Modern Theme - Enhanced glass-panel effects and refined slate/blue color palette

17-Jan-2026

Genie Profiles - Hierarchical AI Organizations with multi-level autonomous coordination (Parent → Child agents)

16-Jan-2026

Session Primer - Auto-initialize sessions with domain knowledge, transforming generic LLMs into pre-educated specialists

16-Jan-2026

Genie Profile Type (Beta) - Multi-profile coordination with LangChain orchestration

16-Jan-2026

Genie UI Features - Inline progress cards, collapsible child sessions, split view access

09-Jan-2026

Knowledge Focused (RAG) Profile Type - Mandatory knowledge retrieval with anti-hallucination safeguards

11-Jan-2026

Profile Classes - Four Execution Modes (Conversation, Efficiency, Knowledge, Genie)

10-Jan-2026

Export/Import Knowledge Repositories

09-Jan-2026

Export/Import Planner Repositories

02-Jan-2026

OAuth Implementation - Google, Github

02-Jan-2026

Email Validation for Registration

20-Dec-2025

Extended Prompt Management System - Dynamic Workflow Prompts

19-Dec-2025

Extended Prompt Management System - Dynamic Variables

19-Dec-2025

Extended Bootstrapping - Enhanced Bootstraping Parameter Configuration

12-Dec-2025

Enhanced Prompt Encryption/Decryption Process using Database Encryption

12-Dec-2025

Migration from File based Application Configuration to Database Schema

05-Dec-2025

Consumption Profile Enforcement - Rate Limiting and Usage Quotas

07-Dec-2025

Financial Governance - Dashboards and LiteLLM Integration

06-Dec-2025

Planner Constructor: SQL Query - Document Context

05-Dec-2025

Planner Constructor: SQL Query - Database Context

29-Nov-2025

Knowledge Repository Constructor - Document Storage

28-Nov-2025

Knowledge Repository Integration

28-Nov-2025

Multi-User Authentication - JWT tokens, access tokens, user tiers

22-Nov-2025

Profile System - Modular Configuration & Temporary Overrides

21-Nov-2025

Planner Repository Constructors - Modular Plugin System

21-Nov-2025

Modern UI Design

15-Nov-2025

Flowise Integration

15-Nov-2025

Airflow Integration

14-Nov-2025

Self-Improving AI (RAG)

07-Nov-2025

UI Real-Time Monitoring of Rest Requests

31-Oct-2025

Fully configurable Context Management (Turn & Session)

26-Oct-2025

Turn Replay & Reload Plan

25-Oct-2025

Stop Button Added - Ability to immediately Stop Workflows

24-Oct-2025

Robust Multi-Tool Phase Handling

11-Oct-2025

Friendly.AI Integration

10-Oct-2025

Context Aware Rendering of the Collateral Report

20-Sep-2025

Microsoft Azure Integration

19-Sep-2025

REST Interface for Engine Configuration, Execution & Monitoring

12-Sep-2025

Significant Formatting Upgrade (Canonical Baseline Model for LLM Provider Rendering)

05-Sep-2025

Conversation Mode (Google Cloud Credentials required)