Work with AI. Without losing control.

One AI workspace for people. One AI control plane for the organisation. Imperium is where employees work with AI — approved models, agents, documents, memory and history in one place — and the environment the organisation uses to discover, govern, decide, control, observe, verify, optimize, protect and execute. It is not an AI proxy, a model router, an API gateway, a cost dashboard, an agent inventory or a policy engine; Imperium Sentinel independently protects it.

YOUR AI WORKSPACE

Work with AI. Without losing control. Employees should not need five different AI products, separate histories, different prompts and unmanaged credentials. Imperium gives each user one governed workspace for the AI tools, models and agents they are allowed to use — and the organisation keeps policy, identity, security and cost.

PERSONAL AI HOME

Useful to someone who is not security, DevOps, IT or an AI administrator. The manager dashboard is a different view — employees are not shown enterprise-security noise.

USE THE RIGHT AI — WITHOUT CHANGING WORKSPACE

Approved access to OpenAI, Anthropic, Google, Microsoft, local models, Ollama, vLLM, internal and specialised organisational models. The employee never manages provider credentials — Imperium brokers them per request and the authority expires.

Depending on policy a user may choose a model, let Imperium select one, compare several, use the organisation-recommended model, or use a local model for sensitive data. Ask once, compare across models, and see quality, latency, cost and policy eligibility side by side.

PERSONAL INSTRUCTIONS & MEMORY

My AI instructions — writing style, output format, language, technical depth, preferred terminology, formatting to avoid, "always include implementation notes", "prefer concise responses". Inherited organisation → department → team → user: personal preferences extend policy, they never override mandatory organisational controls.

My memory — projects, preferred formats, working context, frequently used systems, approved recurring instructions, team terminology. Memory stays visible, scoped, editable according to policy, auditable, classified and governed. This is a productivity feature, not hidden behavioural profiling.

Organisational knowledge — personal → team → department → organisational memory, always policy-controlled. A finance employee asking "explain the current purchasing approval process" gets the finance policy, the internal runbook and their approved memory — and nothing from unrelated departments.

DOCUMENTS · VOICE · COMMUNICATION

AI doesn't only read prompts. A PDF, a scan, a screenshot, an image, an email, a WhatsApp message or a voice recording all enter through the same ingestion boundary — and leave it as governed content.

Conversations are AI input too. A waveform becomes a transcript, the transcript becomes conversation nodes, and Imperium identifies PII, policy violations, quality, the agent's own responses and escalation — with WhatsApp, email, Telegram and chat feeding the same communication surface.

Drop a document. Ask questions. Keep governance. PDFs, scans, images and pasted screenshots are extracted, classified and made available to approved AI under the same policy path as ordinary text.

Where enabled, employees can upload a recording to transcribe a call, summarise a meeting, extract tasks, check it against policy and create follow-up actions. Where integrations are authorised — email, WhatsApp, Telegram, Teams, Slack, chat — Imperium can summarise a conversation, extract actions, draft a reply, identify decisions, find unresolved questions and search historical context, all within organisational access policy.

CODE WORKSPACE

Developers get approved coding models, code agents, repository context, MCP, GitHub / GitLab, prompt history, cost and token optimisation, visible agent actions and visible policy. When something is not allowed, the answer says why — and what will work instead.

POLICY SHOULD GUIDE, NOT ONLY BLOCK

Policy-aware rerouting: rather than only blocking, Imperium can route the request to an approved internal model and say so — "this request contains Confidential data; Imperium routed it to your approved internal model." Employee value and security at the same time.

UNDERSTAND WHAT YOUR AI DID

For their own permitted activity, users can inspect the prompt, the model, the tools, the sources, the agent actions, the cost, the latency, the approvals — and why a request was blocked or why a different model was selected. Transparency becomes user value, not only an administrator investigation tool.

Imperium Judge works for the employee too: compare answers, check quality, ask for a second opinion, verify against sources, check policy. Two model answers are compared on factual support, completeness, policy, source quality and cost — and the user makes the final choice.

YOUR AI USAGE

Shown where policy allows, framed as awareness — not punishment. Token optimisation is visible value: the same task, lower latency, lower cost.

MY AI HISTORY & SAVED WORK

Search your own AI work by conversation, project, agent, model, tool, document or date. Nobody should have to remember whether something was in Claude, ChatGPT or Gemini — Imperium is the unified organisational history layer. Saved prompts, workflows, agents, reports, documents, searches and views live beside it.

ROLE-SPECIFIC WORKSPACES

Same control plane. Different work experience.

TRANSPARENT GOVERNANCE

Employees can see what is inspected, what is retained, which policy applies, why something was blocked, what their own AI activity costs and which data classification applies. Imperium does not imply — and does not default to — screen recording, keystroke monitoring or secret employee profiling.

AI traffic inspection and conversation retention are separate things. Inspection is what the gateway evaluates in order to decide; retention is what is kept afterwards, configurable by application, department, classification, purpose and policy. Regulated organisations can keep more; everyone else does not have to.

Imperium is not another AI chatbot. The workspace is the employee interface to the wider control plane — the same governed platform powers employee AI, agents, models, MCP, actions, data, security, cost and evidence.

One view of the AI estate.

Models, agents, providers, usage, security, policy and cost — in one operational view.

PERSISTENT SESSIONS

A session belongs to the organisation, not to the window it was opened in. It survives a closed laptop, a dropped connection, a different machine and the next morning.

An incomplete session stores its summary, completed work, open work, files, artifacts, tools, decisions and next steps — the resume package. Reopening it anywhere restores the same context and history. See it move →

CROSS-MODEL HANDOFF

The work is not the model. A session can be handed from one model or agent to another — Claude to GPT to a DevOps agent — and the context package follows it, along with identity, policy, trust and evidence. The model may change. The session does not.

IDENTITY BEYOND USERS

Every actor has an identity — human, machine, service, agent, sub-agent, workflow, MCP — and every identity carries an owner, a scope, permissions, data access, tools, actions and a budget.

DELEGATED AUTHORITY

Agents should not inherit everything. Authority narrows at every hop, and the narrowing is enforced, not advisory:

Delegate only what is required. See the chain of authority →

WHEN AN AGENT CREATES ANOTHER

An agent does not simply spin up a helper of its own accord. Before a sub-agent exists, the organisation's controls apply to it too — what it is for, what it may reach and whether a person needs to approve it. The result is that a chain of agents cannot quietly acquire more authority than the person who started the work.

ACCESS FOR A TASK, NOT FOREVER

Agents should not hold permanent secrets. Access is granted for the work in hand and withdrawn afterwards, and provider credentials stay with the organisation rather than being handed to clients or written into prompts.

WHAT CAN IT SEE? WHAT CAN IT DO?

The two questions every security review actually asks, answered per agent — data sources on one side, action verbs on the other, each marked allowed, approval-required or denied. See the split →

FLOW INTELLIGENCE · AI FLIGHT RECORDER

Don't just know that something happened — know how. The flight recorder reconstructs an execution as a graph and replays it through time: nodes appear and disappear, authority changes, data paths reverse. Seven lenses read the same timeline differently.

EXPECTED VS ACTUAL FLOW

Every governed execution has an expected shape. When the actual path deviates — an unknown sub-agent, an unapproved MCP, an external API that was never in the plan — the deviation is the finding, not a log line to grep for later.

FLOW DEVIATION · RISK HIGH. Open the flight recorder →

SAME CONTROL PLANE. DIFFERENT WORK.

Imperium understands different organisational flows without becoming a department-specific product.

FROM DECISION TO EXECUTION

A decision is not the end. Governance continues after it — through a durable workflow that survives waiting, approval, failure and rollback.

WORKFLOW DESIGNER

Build a governed workflow visually: the steps, who approves what, and what happens if something fails. The same workflow can be kept alongside the rest of your infrastructure definitions.