Secure, sovereign AI for search and agents.
Oak gives your teams chat, search and agents over your own documents, inside your own environment. Every answer is permission-checked, cited and recorded.
- Your cloud
- Your servers
- Fully air-gapped
One question. Every source you’re cleared to see.
- SharePoint
- Google Drive
- Confluence
- Slack
- Jira
- Teams
- GitHub
- Gmail
- Box
- Salesforce
- Canvas
Spend you set, not spend you discover.
Open-weight models on hardware you control turn a per-token bill into a fixed, plannable line. Budgets cap tokens and cost wherever you need them.
- Workspace
- One ceiling for the whole deployment.
- Groups
- Separate budgets for Legal, Finance or any group you define.
- People
- A cap for each person, with usage reports that show who used what.
Measured on your questions before anyone relies on it.
We test Oak on a representative sample of your team’s real questions, in an approved evaluation harness, before it goes live.
- Before go-live
- Scored on a sample of the questions your team actually asks.
- Before a change
- A new model or setting runs the same sample before it reaches production.
- Your call
- You see the results, so the decision to switch stays with you.
Your data stays inside your boundary.
- Documents, indexes and chat history stay in your environment.
- Models can run on your own GPUs, with no outside calls.
- Access is checked every time someone asks.
- Every query is recorded and can be exported.
- Oak appChat, search and agents in the browser and desktop app
- API and workersIndexing, permission sync and agent actions
- Search indexKeyword and vector search over your documents
- DatabaseChats, settings, query history and logs
- Model serverOpen-weight models served on your GPUs
- ConnectorsPull from your sources on a schedule you set
Three ways to run Oak. All of them yours.
Your cloud
Oak runs in your own cloud account, next to the data it reads. Identity is wired to your IdP, and we document the architecture for your security review.
- Deployed in your account, not ours
- Sign-in through your identity provider
- Architecture documented for your review
Your servers
Run Oak on hardware you own. We size the GPUs to your users and documents, so the cost is the hardware, not the token.
- The same release as the cloud build
- GPUs sized to your workload
- No cloud account needed
Fully air-gapped
Install Oak with no outside connection at all. Models, search, logs and every answer stay on your network.
- No outbound internet needed to run
- Open-weight models served locally
- Logs and audit records kept on site
Built for your security review.
Fits the stack you already run.
Open and closed models
Run the top open and closed models in your environment, and switch later without rebuilding workflows. Your tools can call Oak through OpenAI- and Anthropic-compatible APIs.
50+ connectors
SharePoint, Google Drive, Confluence, Slack, Jira, GitHub, Salesforce and more, synced on your schedule. Permissions sync from 11 of them.
Where you already work
Use Oak in the browser or the desktop app, ask it in Slack, or give your own agents its search through the MCP server.
We deploy it and run it with you.
Northwood isn’t a login and a help-desk queue. We set Oak up in your environment, train your team, monitor it in production and extend it to new teams on your schedule.
- Pilot
- One team and one workflow, live on your own documents.
- Go-live
- Training, runbooks and a production handover.
- Ongoing
- Monitoring, tuning and new teams when you’re ready.




