How Gluon works
Gluon is the control layer for AI. One governed model sets what AI may read, say and do. Gluon applies it to every question and every action, and records why. The six steps below are the parts of the platform.
01Model
02Connect
03Decide
04Act
05Prove
06Improve
One governed model of what AI may use and do, designed, connected, decided, acted on, proven and improved in one lifecycle, under one permission path.
Lifecycle
Every organisation already decided which version is in force, who may see what and who signs above which amount. Gluon turns those decisions into a model, applies them to every question and every action, and records why.
Each step has its part of the product. Every part uses the same decision point and writes to the same audit log. AI keeps to what you model, and you can prove what AI did.
01 Model
02 Connect
03 Decide
04 Act
05 Prove
06 Improve
Components
Chat, the API, MCP, Gluon Desktop and the designer all ask the same decision point. The channel is an input to a policy, never a separate code path.
- Model · 01OntologyGOL and the designer: your business as a model

- Connect · 02Gluon ConnectYour sources, read where they are

- Connect · 02Gluon DesktopOne person’s Gluon, on a Mac

- Decide · 03The enginesDecide what applies and who may see it

- Decide · 03Tools & GatewayAuthorised tools, models you choose

- Decide · 03ChatAnswers that show their work

- Act · 04Action controlA verdict before an agent acts

- Act · 04API & MCPGenerated from your model

- Prove · 05EvidenceCitations, traces and an audit log you can verify

- Prove · 05ConsoleRun, secure and audit the service

- Improve · 06DriftDo the answers stay right?

Same controls everywhere
- Gluon chatChat
- Gluon DesktopDesktop
- CopilotMCP
- ChatGPTMCP
- ClaudeMCP
- Your appsAPI
- Your data centreOn-premises, next to your sources
- Your cloudPrivate or sovereign, in Europe
- Hosted by usManaged by Gluon, in Europe
- PoliciesWritten as sentences, with who gains access before one widens.
- MarkingsSensitive fields masked by policy, the reason shown.
- AuditAppend-only and hash-chained, streamed to your SIEM.
- Entra IDConditional Access and named locations
- Active DirectoryLDAP and Active Directory groups
- SCIMProvisioning, with break-glass accounts
- OAuth 2.1PKCE for people, API keys for services
- Azure OpenAI
- Amazon Bedrock
- Google Vertex AI
- Mistral AI
- Ollama
Models you chooseEach one sees only the markings you allow it.
Console
The console shows what is healthy and what needs you. Each view names its next step, and every change in it is audited.
Run
Secure
Audit
Sign in withEntra ID
Active DirectoryStreams to
Microsoft Sentinel
Splunk
Employees use the chat or the assistant they already have.
In the manualHealth and usage Investigations
gluon-gol
gluon
gluon-miner