Your computer
Keep your local workspace's documents, ontology, and search index on your device.
Data storage and AI model options
Choose where to store your workspace and which AI models to use, from your own computer to private servers and the cloud.
Where your data is stored
On-premises
Use storage on your own server, with models and access policies set by an administrator.
Cloud workspace
Store data on Consilience servers and connect from your browser or desktop app.
AI model connections
Local LLM
Run Ollama, LM Studio, or the bundled engine on your device or internal server.
API · Endpoint
Connect to your chosen model with your own API key and endpoint.
Consilience relay
Model provider
Connect to a model provider through Consilience’s relay server.
Agent permissions and execution
Approval modes
Change-making tools are restricted while the agent investigates and proposes a plan for approval.
Saved permission rules apply, with approval requests for file changes and actions that need attention.
Routine edits and commands are approved automatically; sensitive changes follow permission rules.
Routine approval prompts are skipped while isolation boundaries and additional safeguards remain.
Ontology change review
Preview a change and apply the exact contents of the saved proposal.
Workspace permissions
Server graph management is available to the owner of a workspace that allows edits.
Command execution boundaries
On supported Mac systems, commands have restricted file access and blocked network access.
Design previews
View generated designs in a preview that blocks script execution.
Update file verification
In-app updates check the package signature before installation to detect tampering.
Settings explained
Can I use it without internet access?
Fully offline mode blocks the app's external requests and online subprocesses. Prepare the local models you need in advance, and check your embedding and voice transcription settings too.
Do chat, search, and voice use the same model?
Models for chat and extraction, search embeddings, and voice transcription are configured separately. Even when documents are stored locally, features using an external model send the inputs they need through that connection.
Does every action need approval?
It depends on the approval mode and permission rules. Full access skips some approval prompts, but proposals cannot be applied if the graph has changed since review.
How are external content and tools handled?
Documents, web content, and tool results are treated as external input, separate from user instructions. Web search and control of a signed-in browser have separate settings, so you can adjust the available tools and permissions.
Can I check and stop work in progress?
The Activity Center shows progress and results for extraction and graph management jobs and lets you request a stop. The agent continues its response after receiving a completed, failed, or cancelled result.
What does on-premises support include?
Version 2.0 is in the deployment evaluation stage before public release. It supports administrator-selected models, deployment configurations that block internet access, and email OTP for the managed web app. Enterprise OIDC and selecting an internal server in the native app are in development.