Your own model.
Permission-checked agents.
Covan agents use the customer’s own model provider: AWS Bedrock, Azure OpenAI, Anthropic, OpenAI, or local Ollama / vLLM. No data leaves the network unless the customer chooses so.
Design configuration by asking.
A managed SM Admin agent can design surveys, service catalog entries, request templates, custom fields and workflows. Every write it proposes is shown for confirmation first, then runs through the same permission checks as a person.
- Other specialists cover service desk, CMDB, asset, change, security and reporting work.
- A coordinator agent routes each question to the right specialist.
Eight specialists, one surface.
Each agent is specialised and can only call the tools it is permitted to use in that domain. A coordinator routes each incoming question to the right specialist.
SM Admin
Designs surveys, catalog items, request templates, custom fields and workflows. Every proposed write is shown for confirmation first and runs through the same permission checks.
Service Desk
Summarises records, finds similar incidents, suggests knowledge answers and routes to the right team.
CMDB
Traverses CI relationships, explains impact analysis and flags missing links.
Asset
Answers custody, warranty, stock and license compliance questions; verifies CI matches.
Change
Summarises change risk, colliding windows and affected services.
Security
Reports open vulnerabilities, patch status and access changes together with the audit trail.
Reporting
Builds a safe query over any field of any object; row and field security is applied inside the query.
Coordinator
Routes each question to the right specialist, orchestrating the team from one conversational surface.
AI cannot step outside permissions.
Agents are not a magic back door. Every tool call is permission-checked, every write is confirmed and every action lands in the audit trail.
Permission-checked tools
Agents can only call the permitted subset of 176 tools. An agent can never exceed the permissions of the person it acts for.
Confirmation before writes
Every proposed write is shown for confirmation first, then passes the same permission checks. An agent cannot change data silently.
Inside your network
The model provider is chosen by the customer. When no cloud model is used, no data leaves the network.
Auditable
Every agent action is written to the audit trail and record history exactly like a human action.
Tools bind to the record engine.
Record reads, CI relationships, workflow design, report queries, SLA analysis and more. Every tool is filtered by the calling user’s role and company scope.
You choose the model, your data stays put.
Cloud providers or fully local models. The same agent definition, the same tools and the same permissions; only the model endpoint changes.
Fully air-gapped with Ollama and vLLM.
Model weights run on your own hardware. Agents use the same 176 tools; the only difference is that inference never leaves the network.
Which agent, which tool, which model.
Model assignment and tool permissions per agent are set by an administrator. For sensitive functions the AI step can be turned off entirely.
The boundaries are explicit.
An agent can never exceed the permissions of the person running it. Agents do not change records silently in the background; they propose, get confirmation, then apply.
- No bulk updates without confirmation.
- Agents cannot decide on a user’s behalf; they present recommendations.
- Model calls appear in the audit trail.
- Data never leaves the network unless the customer chooses so.
This approach makes AI usage auditable in regulated sectors and public bodies.
What the agents change in practice.
AI here is not a separate product but a user bound by the same permission model. The result: faster first response, faster configuration and an auditable decision trail.
Faster first response
The agent summarizes records, finds similar incidents and suggests knowledge answers, so specialists start from the right record.
Faster configuration
The SM Admin agent drafts surveys, fields and workflows; an administrator confirms and the same permission checks apply.
Auditable AI
Every tool call and write lands in the audit trail; an agent can never exceed the user’s permissions.
See the agents on your own model.
In the demo we can show the SM Admin agent designing a workflow, asking for confirmation and passing the permission check.