Self-hosting changes who sees the data. Sensitive fields still enter the model context in plaintext. They can persist in your request logs, shell history and transcripts that get shared, synced or backed up. And the same people often use ChatGPT, Claude or Gemini in the browser too.
Locke works on the data, wherever the model runs. Point your Ollama client at its proxy and sensitive values are masked before they reach the context or your logs. The cloud tools Locke covers get the same treatment.
The risk with Ollama
Running a model locally removes the provider's access to your prompts. Sensitive data still enters the model context in the clear and can persist in your logs, shell history and transcripts. Any cloud AI tools your team also uses stay exposed.
How Locke helps
Locke masks sensitive fields before they reach Ollama or your logs. The Early Access desktop app screens a local model server when you point that client at Locke's proxy explicitly. That is a deliberate choice: Locke does not screen loopback traffic behind your back.
Locke is free for one person
It runs on your device and masks sensitive data before a prompt is sent, on the AI tools it covers. Teams starts at $14.99 per seat a month.
