Run open models in Claude Desktop
Kimi K3, Qwen3.8 Max, GLM 5.3 and DeepSeek inside the Claude Desktop app, through Keln. A 2-minute setup with your Keln API key, effort levels and web search included. Switch back to Anthropic any time.
Claude Desktop has a developer setting that lets it talk to a third-party gateway instead of Anthropic. Point it at Keln and the app keeps everything you like about it: the interface, the effort control, web search, and Claude Code in the Code tab. Only the model behind it changes to an open-source one, and you pick it from the same menu.
What you need to start
- Claude Desktop for macOS.
- A Keln API key. Sign up at app.keln.ai, then open Keys → Create key. Keln shows the key once, so copy it now.
- Credit on your Keln account. Billing → Add credit.
Connect Claude Desktop to Keln
1. Turn on Developer Mode
In the menu bar open Help → Troubleshooting → Enable Developer Mode…
Confirm with Enable. Claude restarts and a Developer menu appears in the menu bar.
2. Open the third-party inference settings
Developer → Configure Third-Party Inference…
3. Create the Keln configuration
In the setup window, open the configuration menu at the top right (it reads Default on a fresh install) and choose New configuration.
Name it Keln and press Confirm. The menu now reads Keln; Default stays as it was, so you can switch back later.
On the Connection page, set the provider to Gateway and Credential kind to Static API key. The key and auth scheme fields appear once you do. Then fill in:
| Field | Value |
|---|---|
| Provider | Gateway |
| Credential kind | Static API key |
| Gateway base URL | https://anthropic.keln.ai |
| Gateway API key | sk-keln-… your key |
| Gateway auth scheme | x-api-key |
Leave everything else at its default. Test connection is optional; a green result means the key works.
Then Apply Changes and confirm with Save & Restart; Claude relaunches on Keln.
4. Claude relaunches on Keln
On relaunch you land on the home screen, with no Claude.ai sign-in. The message box shows the current open model and effort.
5. Pick an open model and effort level
Open the model menu in the message box. The picker lists the Keln models by name, with an Effort submenu below them.
Models and features
| Model | Role | Input |
|---|---|---|
| Kimi K3 | Frontier | text + images |
| Qwen3.8 Max | Frontier | text |
| GLM 5.3 | Frontier | text |
| GLM 5.3 Flash | Flash | text + images |
| DeepSeek V4 Flash | Flash | text |
- Effort levels. Low, Medium, High, Extra and Max map to each model's reasoning levels (table below).
- Web search. Claude's search runs on Keln, server side, priced per search (pricing, details).
- Zero data retention, the same privacy contract as the API (Privacy & zero data retention).
Effort levels
Keln maps Claude Desktop's five effort levels onto three of each model's reasoning levels, so each level you pick changes how long the model reasons:
| You pick | Kimi K3 | Qwen3.8 Max | GLM 5.3 | GLM 5.3 Flash | DeepSeek V4 Flash |
|---|---|---|---|---|---|
| Low, Medium | low | low | low | low | low |
| High (default) | high | medium | high | high | high |
| Extra, Max | max | xhigh | max | max | max |
Higher levels reason longer and bill more output tokens. High is the default on every model.
Switch between Keln and Anthropic
The setup window keeps a list of saved configurations. Apply Default to go back to Anthropic; apply Keln to return to Keln. Each apply relaunches the app. You can keep several Keln configurations, for example one per workspace key.
To go back: open Developer → Configure Third-Party Inference…, pick Default in the configuration menu at the top right, and press Apply Changes.
If the app ever says "Your gateway couldn't serve …", the saved configuration refers to a model Keln no longer offers. Open the configuration, apply it again, and pick a model from the current list.
Set up a team
To set up several Macs, export the Keln configuration as a macOS configuration profile and install it on each machine with your MDM. The profile carries the key, so make it from a key meant for that team.
Claude Code in the app runs on Keln as soon as the app does. Using Claude Code in the terminal instead? It works with three environment variables and no setup window: Anthropic-compatible API. Using Codex? See Run open models in Codex.