-
Notifications
You must be signed in to change notification settings - Fork 2
User Guide Ask Cairn
An optional natural-language question mode in the command palette. Ask "Which decisions were made at the last management review?" and get an answer that cites real records.
It is off by default, and your administrator has to enable it deliberately.
Type a question into the command palette instead of a search term. Cairn routes it to a set of read-only data tools, runs them, and returns two things:
The records. Real rows from your database, each linking to the record it came from. These are the answer.
A summary sentence, written by the configured language model and clearly labelled as AI-generated. It is a convenience over the records, not a substitute for them.
The distinction matters. Verify the summary against the records it cites, particularly for anything that will end up in a report or a decision. The records are facts from your database; the sentence is a model's reading of them.
It cannot widen your access. Every data access inside the loop runs through the ordinary read tools with your permissions and your scopes. It can only cite records you were already allowed to read. Two people asking the same question can get different answers, and that is the tenancy model working.
It cannot change anything. The tools it can reach are read-only.
Where enabled, requirement search becomes meaning-based rather than keyword-based, and works across languages : searching in French finds a requirement written in English about the same control.
This needs an index, which your administrator builds and refreshes. If semantic search seems to be missing results, a stale index is the usual cause.
Every answer has a thumbs up and thumbs down. Use them.
Administrators can review and export the feedback under Administration -> Ask Cairn -> Feedback, together with the original question and what the model returned. A thumbs down on a wrong answer is what makes the next version better; it is not a complaint that goes nowhere.
This is the part to read before asking your administrator to turn it on.
With a third-party provider (Mistral, OpenAI, Anthropic), your question text and the compact record fields used for routing are sent to that provider. That is a data protection decision, not a preference.
With Ollama, pointed at your own instance, nothing leaves your infrastructure.
The provider is a deployment choice. The assistant specification describes each option and its data-egress properties in detail.
If the assistant is disabled or its provider is unreachable, the question mode reports it plainly rather than returning a degraded answer. Ordinary search and navigation are unaffected : Ask Cairn is a layer on top, never a dependency.
The assistant's name is configurable in the company settings. "Ask Cairn" is the default; an organisation that prefers its own name for it can say so.
Built from docs/ at v0.36.0. Edits made here are overwritten by the next release : open a pull request against the source instead.
- Administration
- Ask Cairn
- Assets and suppliers
- Compliance
- The dashboard
- Finding your way
- Getting started
- Incidents
- How records move
- Organisational context
- Reports and management review
- Risks
- Trust Center
- Architecture
- Configuration
- Contributing
- The documentation system
- Installation
- Internationalisation
- Operations
- Release process
- Security
- Testing
- Adding an assistant provider
- Adding a dashboard widget
- Adding a domain entity
- Declaring a lifecycle
- Adding an MCP tool
- Adding a REST endpoint
- Adding a report
- Interface conventions
- Dashboard widgets
- Lifecycles
- MCP tools
- MCP tool parameters : Assets
- MCP tool parameters : Compliance
- MCP tool parameters : Governance and context
- MCP tool parameters : General
- MCP tool parameters : Incidents
- MCP tool parameters : Reports and management review
- MCP tool parameters : Risks
- MCP tool parameters : System and administration
- MCP tool parameters : Trust Center
- Management commands
- Models
- Permissions
- REST endpoints
- Environment variables
- MCP server
- REST API
- Assistant module (Ask Cairn)
- Module 0: User Management and Access Control
- Module 1: Context and Organization
- Module 2: Asset Management
- Module 3: Compliance
- Module 4: Risk Management
- Module 4 bis - EBIOS Risk Manager
- Module 5 : Trust Center
- Module 6 : Security Incident Management
- Management review : ISO 27001:2022 compliance (clause 9.3)