A coding agent that can prove it never stores your data
Security teams veto coding tools that hold code on someone else's server. WorkspaceGPT is built so there is nothing to hold: indexing and embeddings run on your machine, the index is local files, and in Local mode with Ollama no request leaves your computer at all.
Privacy as architecture, not a policy toggle
A privacy mode is a promise about what a vendor does with data it already received. WorkspaceGPT's approach is to never receive it: your documents and the vector index built from them stay on your device in both modes.
That makes it easier to get through a security review, and easier to use on the repositories and internal documentation you wouldn't paste into a hosted chat.
How it works
1. Local mode — free
Run the chat model on your own hardware with Ollama, or use your GitHub Copilot plan or your own provider key. No account is needed, and with a local model it runs offline.
2. Remote mode — managed model
We host the inference so there is nothing to configure. Only your question and the snippets retrieved for it reach our endpoint, are processed in memory, and are discarded.
3. Same agent either way
Search, multi-file edits, test runs and Confluence, Jira and Azure DevOps grounding work the same in both modes.
What you get
On-device indexing
Embeddings are generated locally and the vector index is written to local files. Your documents are not uploaded to build it.
Zero data retention
No prompts, answers or documents are stored, not even in logs. Usage analytics count features and never carry content.
Never used for training
Nothing you ask is used to train a model, sold or shared.
Secrets in your keychain
Credentials are kept in your host's secret storage, not in settings files.
Frequently asked questions
- Is there an AI coding agent that works offline?
- Yes. In Local mode with Ollama there are no remote APIs at all, so the whole agent loop runs on your machine with no account.
- Does WorkspaceGPT store my code or prompts?
- No. It retains no prompts, answers or documents. In Remote mode a request is processed in memory and discarded.
- Is it suitable for regulated or security-conscious teams?
- It is designed for them: the knowledge index never leaves the device, there is a zero-retention policy you can read in full on the privacy page, and Local mode needs no external service.
- What does Remote mode send to your servers?
- Only your question and the snippets retrieved for it. Inference is performed by an upstream model provider under its own policy, which the privacy policy documents.