1
Visibility
“How many AI tools do we run?”
Shadow AI nobody approved. Desktop chat apps, coding assistants, local models and MCP servers are installed without a ticket — with access to code and customer data.
Aegis AISPM
Say yes to AI.
Keep control of it.
The self-hosted control plane that finds every AI tool in your organisation, gives each user a key you can revoke, inspects every prompt and reply, and keeps the evidence an auditor asks for.
Why · The Problem
AI arrived before the controls did
Staff adopted AI one browser tab, IDE plugin and API key at a time. Security now faces four gaps.
2
Access
“Who spent this, and how do we stop it?”
Shared master keys sit in notebooks, CI jobs and laptops. A leak means rotating one key and breaking every team. Spend has no ceiling.
3
Data
“Could we have blocked it in time?”
Customer exports pasted into chatbots. Hidden instructions inside documents. Agents calling tools they should never reach.
4
Accountability
“Show me what this model was asked.”
ISO 42001 and the EU AI Act ask for an inventory, risk decisions and logs. Teams rebuild them from spreadsheets for weeks.
Why Aegis
Built for organisations that keep AI data in-house
Banks, insurers, hospitals and government need AI controls they run themselves and can prove to a regulator.
Sovereign
Self-hosted on your servers or Kubernetes, and air-gap capable. Prompts, replies and logs never leave your environment.
Inline
Blocks or redacts on every request before it reaches a model. Posture scanners can find risk — Aegis can stop the prompt.
Two Planes
An endpoint agent finds desktop apps, IDE tools, local models and MCP servers that never pass through any gateway or cloud log.
Provable
Signed, readable policy rules, a write-once audit trail, and an evidence pack mapped to ISO 42001 and the EU AI Act.
What · The Product
One control plane, four pillars
Aegis sits between your people, applications, agents and AI providers. Every pillar works on the same request path.
I
Discover
AI inventory and bill of materials
- Shadow AI found on every laptop
- Approval before any new model or tool is used
II
Control
Keys, grants and budgets
- Provider keys locked in a vault
- A personal, revocable key per user
- Model grants and daily or monthly budgets
III
Protect
Inline prompt and reply inspection
- Checks for personal data, secrets and injection
- Block, redact, coach or audit per model
- Agent and MCP tool control
IV
Prove
Audit trail and compliance evidence
- Write-once audit trail and session replay
- Risk register and impact assessments
- One-click compliance evidence pack
How · The Workflow
Every AI request takes the same four steps
From a person in a browser, a developer in Cursor or Claude Code, or a service in CI. The caller never holds a provider key.
Identify
The tool and device are in the approved inventory. The user's virtual key resolves to a person and a scope.
Authorise
The person holds a grant for this model and has budget left. If not, the request stops here.
Inspect
Detectors check the prompt. Policy allows, redacts or blocks, with a reason. The reply is checked before it returns.
Record
Request, decision and reason are written to the audit trail, your SIEM and write-once storage.
If the policy engine is unavailable, requests fail closed. Logging never slows the request, and a leaked virtual key can be revoked for the whole organisation in one action.
What You Get
What changes in the first quarter
Outcomes your security, finance and compliance teams can each point to.
Security
A complete AI inventory
Every AI tool, model, agent and key, with an owner and a risk tier. Shadow AI becomes a list you can act on.
Answer the board on day one
Operations
Leaked keys fixed in seconds
Revoke one person's key everywhere at once. No master-key rotation, no outage for other teams.
A drill that took hours now takes one click
Data Protection
Leaks stopped in the request
Personal data and secrets are redacted or blocked before they reach a provider, with a message telling the user why.
Prevention instead of breach reports
Compliance & Finance
Audits and spend in check
Audit evidence generated from production in minutes. Budgets per person and team stop surprise AI bills.
Faster audits, predictable cost
Agent Transparency
What the endpoint agent detects — and what it will not do
The agent looks for AI, and only AI. It reports only to your Aegis server — never to a vendor analytics service or third party.
Detects
Apps
Desktop apps such as ChatGPT, Claude and Cursor, plus IDE extensions like Copilot and Continue.
Local AI
Runtimes such as Ollama and LM Studio, large model files, and AI packages in Python and Node projects.
Agents
MCP server configurations, browser AI extensions, and AI provider key names in environment files.
Traffic
On connections to known AI services only — checks for personal data, secrets and prompt injection.
Will not
Will not read non-AI traffic
Banking, email and ordinary browsing pass straight through. Only known AI services are checked.
Will not upload your secrets
API key values and passwords are never sent. Sensitive matches are replaced with markers.
Will not watch the user
No screenshots, keystroke logging, webcam or microphone. It reads AI settings and package lists only.
Will not act out of sight
Every inspected and skipped connection is logged on the device. Admins can pause inspection instantly.
Where · Infrastructure
Deployed inside your own environment
Aegis server, AI gateway, admin console, key vault and audit stores run in your perimeter — with connectors to your SIEM, SSO and key management.
Single server · Docker Compose Kubernetes · Helm, scales out Private cloud · your VPC or tenancy Air-gapped · no internet needed
Next Steps
From demo to protected in eight weeks
Start in report-only mode so nothing breaks, then tighten controls as the data comes in.
- 1
Weeks 1–2
Discover
Install Aegis, enrol a pilot group of laptops, connect providers. Shadow AI is reported — nothing is blocked.
- 2
Weeks 3–4
Control
Move teams from shared master keys to personal virtual keys with grants and budgets.
- 3
Weeks 5–8
Protect
Run detectors in audit mode, tune with dry runs, then switch the critical ones to block.
- 4
Week 9 on
Prove
Generate the first evidence pack and report AI posture to the board each quarter.
Get Started
Ready for a 30-minute demo?
See posture and inventory, shadow AI on laptops, a live blocked leak, and an ISO 42001 evidence pack — then start a report-only pilot on your own AI traffic.