ORION Enterprise AI Platform
The Orion platform

One platform for knowledge, agents and processes.

The AI platform you own.

Cited answers from your own knowledge, AI agents that do the work, and processes with your people in control, built on the systems you already run, in your cloud or data centre, with the models you choose.

  • In your cloud or data centre
  • Choose your models, no lock-in
  • Every action approved & audited
Why Orion

Why organisations choose Orion.

Own it

Orion runs in your cloud account or data centre. Your documents, data and model traffic stay under your control.

Choose your models

Commercial, open-source hosted or self-hosted — one model for the whole platform, or a mix. As models overtake each other, switch without rebuilding.

Trust every answer

Answers cite their sources. People approve sensitive actions. Every step is recorded for audit.

Keep your systems

Orion works on top of the email, documents, ERP and core systems you already run. No rip-and-replace.

Start small, grow

Begin with one module — most start with Knowledge Hive — and add the rest on the same foundation.

Configure, don’t code

Set up sources, models, access and branding through an admin console — configuration, not custom code.

Architecture

How Orion fits together.

Orion sits on top of the systems you already run. People engage at the top, your data flows in from the left, Orion reasons and acts in the middle, and results flow back to your systems — all governed.

Orion architecture: people and channels at the top; your systems and data flow in on the left; the Digital Assembly Line, Agent Hive and Knowledge Hive run on a Shared Foundation in the middle; results are written back to your systems on the right; trust and governance, platform operations and infrastructure layers apply across everything.
People & channels
Web appChatREST / SSE APISlack (approvals)
Your systems & data
  • Email & PST archives
  • SharePoint & files
  • Databases
  • ERP & core systems
  • Scans & images
Orion The AI platform you own
Where work happens Digital Assembly Line
  • Agent + human workflows
  • Approvals & retries
  • Visual designer
  • Triggers: events, schedules, APIs
The workforce Agent Hive
  • Agent templates
  • Smart agents
  • Tools (MCP)
  • Versioned & replayable
What you know Knowledge Hive
  • Cited answers
  • Hybrid search
  • Permission-aware
Shared Foundation · the power behind every module
Your choice of models, one gatewayTools & integrations (MCP)Identity & accessCost controls
Back into your systems
  • Write-back via APIs & MCP
  • Human-approved
  • Safe to repeat (idempotent)
  • Cited answers to people
Trust & governance
Single sign-onRole & document-level accessPersonal-data maskingGuardrailsAudit trail
Platform operations
Quality evaluationTracing & monitoringCost trackingVersioned releasesAdmin console
Runs on
AWSMicrosoft AzureGoogle CloudOracle Cloud (OCI)Your data centre

Want the component-level view? See the technical architecture

The platform

Three modules, one shared foundation.

Find answers, delegate tasks and run whole processes. Each module runs alone or together on the shared foundation — most start with Enterprise Knowledge and add the rest without replacing anything.

01 · Find answers · Orion Knowledge Hive

Enterprise Knowledge

Finds and explains what your organisation knows — with the source for every answer. Ask in plain language and get a cited answer, drawn only from what each person is allowed to see.

Explore Enterprise Knowledge
Step 1ConnectEmail and PST archives, SharePoint, file storage, uploads.
Step 2UnderstandPDFs, Word, scans, tables, images; personal data masked.
Step 3AnswerSearch, re-rank, draft, check — and cite every claim.
Step 4ImproveAnswers scored on your data; every question traced.
  • Hybrid keyword and semantic search with re-ranking
  • Self-correcting answers, checked against their sources
  • Document-level access that mirrors source permissions
  • Collections refresh without downtime
02 · Delegate tasks · Orion Agent Hive

AI Agents

Build, version and run AI agents that use your tools and knowledge, within guardrails you set. An agent is a versioned definition that Orion runs reliably — and records in full.

Explore AI Agents
Research and answerSearches knowledge and tools, reasons over the results, and returns a cited answer.
Extract and validateReads documents, pulls out structured fields, and cross-checks them against other sources.
Triage and routeClassifies a request, drafts a response, and routes exceptions to the right person.

Anatomy of an agent

InstructionsRole, goals and rules of engagement
ModelPrimary model with fallback, chosen per agent
ToolsMCP servers and APIs from a governed catalogue
KnowledgeRetrieval from Knowledge Hive, under the caller’s access
Sub-agentsDelegation to specialist agents
MemoryPer-run state and long-term store, scoped by namespace
GuardrailsPermissions, tool limits and monitoring
  • Registry of immutable versions with rollback and run lineage
  • Streaming, resumable and scheduled runs
  • A curated catalogue of MCP tools, governed centrally
  • Human approval can be required before sensitive actions
03 · Run processes · Orion Digital Assembly Line

Agentic Workflows

Long-running processes where agents and people hand off and approve. Processes are defined declaratively, run durably and pause for human decisions — surviving restarts and failures.

Explore Agentic Workflows
Survives failureState is persisted at every step, so a restart or crash resumes the case instead of starting over.
Waits as long as neededA case can pause for hours or days for a person or an external signal, using no compute while it waits.
Safe to retrySteps are written to be repeated safely; retries are visible and governed by policy.

Example process Illustrative

  1. 01TriggerTriggerA new item arrives by event, webhook or schedule.
  2. 02AgentExtractAn agent reads documents and pulls out structured data.
  3. 03DecisionDecideConditions and rules route the case.
  4. 04HumanApproveA person reviews and approves, rejects or amends.
  5. 05ActionWrite backResults are written to the system of record.
  6. 06NotifyNotifyStakeholders are informed; the case is closed.
  • Visual designer with step palette and validation before publish
  • More than a dozen step types: agents, decisions, loops, parallel branches, human review
  • Event, schedule, signed-webhook and API triggers
  • Versioned definitions — running cases keep the version they started on
Shared Foundation

The power behind every module.

One set of models, tools, identity and cost controls — deployed once, or embedded when you run a single module.

Model gatewayRouting, failover, budgets and usage for every model call.
Tool gateway (MCP)Governed access to MCP servers, APIs and other agents.
Identity & secretsShared identity; credentials injected at the gateway, never in agent definitions.
Cost & budgetsToken and spend usage per model and route, capped by budgets.
Capabilities

What Orion can do.

Six things Orion does for your organisation, from the first answer to the finished process.

Answer

Get cited answers from your own documents, email and data — and a clear “not found” when the sources don’t say.

Knowledge Hive

Delegate

Put AI agents to work on routine tasks, using your tools and knowledge, within guardrails you set.

Agent Hive

Automate

Run multi-step processes end to end. They wait for people, retry on failure and resume after a restart.

Digital Assembly Line

Approve

Bring the right person in at the right moment, with a complete record of every decision.

Digital Assembly LineAgent Hive

Connect

Read from and write back to the systems you already run, safely and with approval.

Digital Assembly LineAgent Hive

Prove

Measure answer quality, trace every run step by step, and track cost by model.

Across the platform
Flexible by design

Models keep overtaking each other. Orion lets you move with them.

CommercialOpenAI · Anthropic Claude · Google Gemini · Mistral · xAI
Open-source, hostedOpen models run for you by a hosting provider
Self-hostedOpen models on your own GPUs or in your data centre
Orion model gateway One governed route to all of them, with budgets
One model, whole platformSimple and consistent
Hybrid, by taskLighter models for routine work, stronger ones for hard reasoning
Trust

Built for governed, regulated work.

Identity, access, privacy and audit are part of the platform — applied to every request and every run, not left to individual agents.

Your data stays yours

Orion runs in your environment. Documents, data and model traffic are not sent to Orion or shared across customers.

The right people see the right content

Sign in with your company identity. Access to documents is denied by default and follows the permissions of the source system, such as SharePoint.

Sensitive data is protected

Personal data is detected and masked on the way in. Guardrails block prompt-injection and out-of-scope questions.

People stay in charge

Sensitive actions wait for human approval, and every decision is recorded. AI never writes to your systems unchecked.

Quality you can prove

Every answer is checked against its sources and scored. Evaluation runs on your own data; every run is traced step by step.

Costs under control

Spend is tracked per model and capped by budgets.

Why not just use a public AI assistant?

You needA public AI assistant typicallyOrion
Data that stays in-houseSends your content to the vendor’s cloud.Runs in your environment.
Answers you can verifyFluent, but often without sources.Cited, and checked against sources.
Work that gets doneAnswers questions, then stops.Runs processes, with approvals.
Freedom to changeTied to one vendor’s model.Your choice of models, one gateway.
Deployment & fit

Fits how you run.

DeploymentA dedicated instance for your organisation, in your own cloud account (AWS, Microsoft Azure, Google Cloud, Oracle Cloud) or data centre.
ModelsThree ways to run them: commercial (OpenAI, Anthropic Claude, Google Gemini, Mistral, xAI), open-source models hosted for you, or models you host yourself. Use one across the platform or mix them by task, and change as the field moves.
Your systemsEmail and PST archives, SharePoint, file and object storage, databases, and enterprise applications through APIs, webhooks and MCP.
Sign-in and accessSingle sign-on with your corporate identity provider. Role-based and document-level permissions.
ChannelsWeb app and chat, REST / SSE API, and Slack for approvals.
OperationsTracing, quality scoring, cost tracking and dashboards. Configure through an admin console — no code.

Dedicated instance

A single-tenant installation for one organisation, with its own data, indexes and configuration.

Your cloud account

AWS, Microsoft Azure, Google Cloud or Oracle Cloud Infrastructure, in the region you choose.

Your data centre

On-premises on your own hardware, with self-hosted models for fully private operation.

Getting started

Explore. Experience. Execute.

From first conversation to production — three steps, one platform.

  1. Explore 01

    Choose the use case

    A short workshop to explore where AI pays back first: one process or knowledge area, and the systems it touches.

  2. Experience 02

    Prove it on your data

    Experience Orion working on your own documents and rules, with quality scored — before you commit further.

  3. Execute 03

    Run it and expand

    Put it into production, then add agents, workflows and more sources on the same foundation, module by module.

See Orion on your own data

Ask for a demonstration or a proof of value.

Start with one process or knowledge area. We prove it on your own documents and rules, with quality scored, before you commit further.