Uniphore

Uniphore says customers can connect agents directly to their existing systems, data and infrastructure, with enterprise integration by design in the Agentic Layer. The Data Layer has data connectors and a zero-copy fabric that connects data across any platform, app or cloud. There is no public connector catalog, named system list or action scope, so it is not stated whether integrations only read or can also take actions, or which are native. Outside agents can work with the platform through Uniphore's Multi-Agent Collaboration Protocol and through MCP.

Sourceuniphore.com/platform/business-ai-cloud/agentic-layer/, /business-ai-cloud/; readread 2026-09-16

The Agentic Layer has an AI Agent Orchestration Engine that coordinates interactions across multiple agents, systems and third-party tools through a centralized control layer, plus Visual Agentic Workflows built on a BPMN-compatible framework that maps to business logic. BPMN-based orchestration, announced with the Business AI Cloud launch, deploys agents into workflows across sales, marketing, service and HR.

Sourceuniphore.com agentic-layerread 2026-08-31

Retrieval over enterprise data runs through RAG and knowledge graphs, which are Knowledge Layer components alongside process discovery and the SLM factory. The platform connects data across any platform, app or cloud and transforms it into contextual knowledge, and customers can choose components including vector DBs and knowledge graphs.

Knowledge graphs and vector databases over enterprise data persist, scale past the context window and stay queryable, and new knowledge is added through RAG over connected data without retraining. How sources are refreshed and permissioned, whether retrieval is visible and whether answers carry citations are not stated publicly, and the figure of over 90% retrieval accuracy is Uniphore's own claim. Small language model fine tuning, by contrast, compiles knowledge into weights.

Sourceuniphore.com/platform/business-ai-cloud/; uniphore.com/the-business-ai-company/; readread 2026-09-16

The Model Layer provides AI security through built-in guardrails and policy control, with Guardrails among its components, and the Agentic Layer builder lets customers connect to enterprise data and apply their own policies. There is no approval step, checkpoint or pause on agent actions, so where a person could approve (node, workflow or policy layer) is not stated, and no developer documentation is public.

Sourceuniphore.com/platform/business-ai-cloud/, /agentic-layer/; readread 2026-09-16

Customers manage their data with role-based access controls and embedded guardrails, so users only access what they are authorized to see. Uniphore is ISO/IEC 27001:2022 certified, SOC2 Type II certified and a Level 1 PCI DSS service provider with an annual Qualified Security Assessor assessment and a current Attestation of Compliance.

Its other controls, staff MFA and least privilege on the production network, SIEM, intrusion detection, pen tests and EDR, are Uniphore's internal security and do not tell a customer who inside their own organization can make an agent do what. SSO, SAML or SCIM, and how far role based access reaches over agents, tools and models, are not stated, and no developer or admin documentation is public.

Sourceuniphore.com/platform/business-ai-cloud/; uniphore.com/security/ (web fetcher; sandbox blocked by Cloudflare WAF); readread 2026-09-16

Uniphore names observability among the Business AI Cloud's platform protections, alongside embedded guardrails and AI security, but does not describe it. Its SIEM gathers logs from Uniphore's network devices and host systems, which covers Uniphore's own infrastructure, not agent behavior. No public page shows whether prompts, tool calls, retrieved knowledge and outputs can be inspected step by step, and there is no run trace or agent audit log.

Sourceuniphore.com/the-business-ai-company/; uniphore.com/security/; read 2026-08-31 andread 2026-09-16

There is no stated session state, resumable conversation state or memory across sessions, and with no developer documentation public, the memory types on offer (session, conversation, workflow, long term) are not stated. Perpetual self-learning, one of the platform foundations, adapts models and does not keep agent state. Fine tuning small language models changes weights, which are not a memory layer, and knowledge graphs and RAG over enterprise data are retrieval, not memory.

Sourceuniphore.com/platform/business-ai-cloud/, /agentic-layer/, /knowledge-layer/; readread 2026-09-16

Uniphore hosts service data in AWS data centers in the United States and Europe, and customers can choose to keep their service data US-only or in the EEA. The platform runs on any public cloud, private cloud or on-premises with full control over data and models, and the platform foundations include On-Premises and Multi-cloud. Uniphore describes the Business AI Cloud as a sovereign, composable and secure platform.

Sourceuniphore.com/security and uniphore.com the-business-ai-companyread 2026-08-31

The Agentic Layer ships pre-built enterprise-grade AI agents trained on industries and use cases alongside the custom builder, and the builder offers Prebuilt Templates and Actions with common blueprints and reusable building blocks to accelerate setup. The Business AI Suite includes named ready-to-deploy agents such as a Real-time Guidance Agent and a Conversation Insights Agent.

Sourceuniphore.com agentic-layer and business-ai-suiteread 2026-08-31

Multi-Modal Support in the Agentic Layer lets agents understand and interact via voice, text or structured inputs, adapting to any user interface or channel, and the Business AI Suite covers contact center and customer-facing channels. How work reaches the agent is not stated, and there is no event trigger, scheduled run, webhook or inbound queue mechanism.

Sourceuniphore.com agentic-layerread 2026-08-31

The Model Layer offers model orchestration across open, closed or fine-tuned models, with Model Garden, Industry Models and BYO Models among its components, and customers can integrate with both open and closed-source LLMs.

The Knowledge Layer's SLM Factory lets customers fine-tune or create domain-specific models, and the Agentic Layer builder offers Bring your own data & models, with fine-tuning on task-specific SLMs. Bringing their own models, selecting from a model garden and fine-tuning their own SLMs put the model choice with the customer. The models in the garden and per agent model assignment are not listed publicly.

Sourceuniphore.com/platform/business-ai-cloud/, /agentic-layer/; readread 2026-09-16

The Agentic Layer offers external integration through Model Context Protocol (MCP) and MCP based orchestration. Through Uniphore's Multi-Agent Collaboration Protocol, third-party agent integration works with external agents and tools, with a control layer for custom and third-party agents. Communications with Uniphore's UI and APIs run over TLS, which is transport encryption, not a callable interface. There is no public API reference, SDK or developer documentation, and Uniphore does not say whether MCP support means consuming MCP servers, serving one, or both.

Sourceuniphore.com/platform/business-ai-cloud/, /agentic-layer/; readread 2026-09-16

Customers can run their own agent through a simulator before launch. The agent builder's Test Before You Launch capability simulates agent behavior, audits logic flows and refines responses before going live, so a change can be evaluated before it reaches customers. There is no regression detection or version comparison surface, and no score or pass rate is shown.

Sourceuniphore.com agentic-layerread 2026-08-31

There is no agent that operates a browser, desktop or application screen. Large Action Model and process discovery are Knowledge Layer components, the capability Uniphore acquired with Orby AI, but no public page says what that component does. The Agentic Layer's agents act through orchestration, enterprise integration and MCP, and its multi-modal support covers voice, text or structured inputs, which is how users reach agents, not agents driving interfaces.

Sourceuniphore.com/platform/business-ai-cloud/, /agentic-layer/; uniphore.com/orby-ai/ (read 31 August); readread 2026-09-16