Jan
A local-first AI desktop client for managing local models, connecting compatible APIs, saving conversations and extending workflows through an OpenAI-compatible interface.
What Jan provides
Jan places local models and compatible API providers in one desktop workspace for conversation, prompt organization and model comparison. Local inference can reduce requests leaving the device, while model downloads, updates, online providers and extensions still have distinct network and permission boundaries.
Workspaces and provider separation
A compatible API shares a request format, not the provider's model ability, retention policy, billing or tool support. Give local, test API and production API workspaces explicit names, use revocable limited keys, and keep customer data and internal endpoints out of reusable prompts.
Maintenance note
This page reviews Jan 0.8.4, three-platform installation, local models, compatible APIs, workspace data paths, extensions and AGPL-3.0 boundaries. Content review date: 2026-08-23.
Save to your cloud drive
Open the cloud drive to get the file directly, or save it for convenient access on another device.
Quark Cloud Drive
RecommendedSave Jan to this cloud drive
Baidu Netdisk
Save Jan to this cloud drive
Jan 0.8.4 local model and API configuration guide
Create an isolated test workspace, verify a local model, then add one compatible API provider with a limited key before moving real conversations into a long-term setup.
Before you start
- Prepare the Windows x64, macOS universal or Linux x64 package and reserve space for model files.
- Use non-sensitive content for local tests and a revocable, limited test key for online providers.
- Record model source, quantization, API address, model ID and data-retention policy separately for each provider.
Installation steps
- 01
Install the platform build
Install the Windows, macOS or Linux package, then verify version 0.8.4, workspace settings and actual model and conversation directories.
- 02
Initialize a local workspace
Keep online providers empty, load a small local model, send a test message and observe loading, context and generation behavior.
- 03
Add one compatible API
Enter the base URL, model ID and limited test key for one provider, send a minimal request and save it only after the response is verified.
Quick start
- 01
Compare local and online paths
Use the same harmless questions in local and online workspaces, recording response speed, context, cost notices and network behavior.
- 02
Organize prompts and sessions
Keep local, test API and production API workspaces separate, export important prompts and exclude keys, customer data and internal URLs from shareable templates.
- 03
Pin and back up configuration
Record model files, provider IDs and app version, export required data before upgrades and retain the previous configuration until recovery is tested.
Usage tips
- OpenAI compatibility describes a request shape; model capability, tools, context and billing remain provider-specific.
- Local speed depends on memory, CPU, GPU, quantization and context, not merely the size of the application installer.
- Extensions and tools may read files, environment variables or networks; enable only the components required for the current task.
Troubleshooting and uninstall
Why will a local model not load?
Check model format, file integrity, architecture and available memory, then try a smaller model and inspect logs instead of renaming an incomplete download.
Why does an API connect but not generate?
Check the base URL path, model ID, key scope, balance and response format with a minimal request, separating provider errors from client configuration.
- Export and revoke dataExport required conversations and prompts, delete or rotate API keys, and record local model and workspace locations.
- Uninstall the app and handle models separatelyQuit Jan and uninstall it through the operating system; clear models, caches and extensions only after confirming that original data is backed up.
Frequently asked questions
Can Jan use only local models?
Yes. Load a downloaded local model and leave online providers disabled, but check model downloads and updates separately because they can still use the network.
Does an OpenAI-compatible API behave like OpenAI in every way?
No. The request format may match while model capability, tools, context limits, billing and retention differ by provider.
Where does Jan store models and conversations?
The exact paths depend on the operating system and settings. Confirm them after first launch and record them before migration or uninstall.