Cline ships with first-class support for the big providers, and at ~64K GitHub stars it is one of the most-used coding agents in VS Code and JetBrains. But the built-in list is not the whole story. The “OpenAI Compatible” provider is how you point Cline at a gateway, a local runtime, an internal inference endpoint, or a model that has not landed in the directory yet.
The flow is short: choose the OpenAI Compatible provider, paste a Base URL and API key, name the model, set a couple of capability fields, and verify. The one step people skip is the Model Configuration block, which matters more for custom endpoints than it does for built-in providers.
Just want Haimaker in Cline? Skip the manual settings — run
npx -y @haimaker/connect --clineand the CLI writes Cline’s OpenAI Compatible provider config (endpoint, key, and model) for you, then exits. The walkthrough below is the general method for any OpenAI-compatible endpoint — Ollama, LM Studio, an internal gateway, or a model not yet in the directory. See the connect guide for the one-command path.
When to use a custom provider
Reach for the OpenAI Compatible provider when the model or endpoint is not already in Cline’s built-in list, or when you want to route several model families through one key.
Good examples:
- Haimaker — one API key for multiple model families through an OpenAI-compatible gateway.
- Ollama — local models at
http://localhost:11434/v1. - LM Studio — local models at
http://127.0.0.1:1234/v1. - Internal gateways — company-hosted OpenAI-compatible endpoints.
- New providers — anything that speaks the OpenAI chat completions API before Cline’s directory catches up.
If the provider already has a dedicated entry in Cline, start there. The OpenAI Compatible path is most useful when you need a custom Base URL, a gateway, or a model that the default options do not expose.
Step 1: Open settings and pick the provider
Click the gear icon in the Cline panel to open settings. Under API Provider, select OpenAI Compatible.
This switches the form to the generic OpenAI-compatible fields: a Base URL, an API Key, a Model field, and a Model Configuration section. Everything else on this screen keys off those values.
Step 2: Enter the Base URL and API key
Fill in the two connection fields:
- Base URL — the provider’s API endpoint. For Haimaker, that is
https://api.haimaker.ai/v1. Cline accepts both the version-prefixed base (https://api.provider.com/v1) and, for some providers, the full chat-completions URL. When in doubt, use the/v1base. - API Key — the secret key from your provider. Paste the raw key with no
Bearerprefix.
There is also a Use Azure Identity Authentication checkbox. Leave it unchecked for standard API-key providers like Haimaker or Ollama. It only applies to Azure deployments where you authenticate with az login instead of a key.
Step 3: Set the Model ID
In the Model field, enter the exact model identifier the provider expects in the model field of a request. Custom providers pass this string through unchanged, so it has to match the upstream API character for character.
For Haimaker, that looks like:
anthropic/claude-sonnet-4-6
openai/gpt-5.4-mini
minimax/minimax-m2.5
qwen/qwen3-coder
If you are unsure of the exact string, list the provider’s models first (see the verification step below).
Step 4: Fill in Model Configuration
This is the step that trips people up. For built-in providers, Cline already knows each model’s context window, output limit, and whether it supports tool calls. For an OpenAI Compatible provider it cannot detect any of that, so it falls back to conservative defaults that often do not match your model.
Open Model Configuration and set:
- Context Window — the model’s real input limit. If this is too low, Cline truncates context early; too high and you get upstream errors on long sessions.
- Max Output Tokens — the per-response output cap.
- Image Support — turn on only if the model is multimodal.
- Computer Use — turn this on for any model you want to drive agentic edits, file operations, and tool calls. With it off, Cline treats the model as chat-only and agentic actions silently fail.
- Input Price / Output Price — the provider’s per-million-token rates, used for Cline’s running cost estimate. These are for display only and do not affect requests, so set them to whatever your provider currently charges.
Getting the Context Window and Computer Use fields right resolves most “the model is connected but acts broken” reports.
Step 5: Verify and test
Click Verify to confirm the Base URL, key, and model resolve. Then send a small prompt in a throwaway task before pointing Cline at real code.
If Verify fails, the problem is almost always the Base URL, the key, or the Model ID, in that order. You can isolate it by hitting the API directly:
curl https://api.haimaker.ai/v1/models \
-H "Authorization: Bearer your-haimaker-api-key"
A clean JSON list back means your credentials and endpoint are good, and the issue is in the Cline form. Use the exact id values from that response in the Model field.
Full example: Haimaker gateway
This is the setup to use when you want one OpenAI-compatible endpoint for several model families:
| Field | Value |
|---|---|
| API Provider | OpenAI Compatible |
| Base URL | https://api.haimaker.ai/v1 |
| API Key | your Haimaker key |
| Model | anthropic/claude-sonnet-4-6 |
| Context Window | match the model (for example 200000) |
| Max Output Tokens | a sane cap (for example 8192) |
| Computer Use | On |
Why this works well: Cline sees one provider, while Haimaker handles access to multiple upstream model families behind a single key. To switch models, you change the Model field (or save a second profile) instead of juggling separate provider credentials. The same key gives you a wide range of models, so you can move from a budget model for routine edits to a premium model for hard refactors without re-authenticating.
Full example: Ollama local provider
Ollama exposes an OpenAI-compatible endpoint on your machine, which makes it a clean fit for the same provider type:
| Field | Value |
|---|---|
| API Provider | OpenAI Compatible |
| Base URL | http://localhost:11434/v1 |
| API Key | any non-empty string (Ollama does not validate it) |
| Model | qwen3-coder:30b |
| Computer Use | On |
Pull the model before you point Cline at it:
ollama pull qwen3-coder:30b
Confirm the exact tag with ollama list and use that string in the Model field. Local models are more sensitive to context limits and tool-call formatting, so start with one that handles agentic coding well, such as Qwen3 Coder, and keep the Context Window modest until tool calls behave.
Common errors and fixes
Verify fails immediately
The Base URL or key is wrong. Confirm the URL ends at /v1 (not /v1/chat/completions unless your provider requires the full path), and that you pasted the key with no Bearer prefix and no trailing whitespace. Test with the curl command above to isolate Cline from your credentials.
Model connects but every agentic action fails
Computer Use is off. Cline is treating the model as chat-only, so file edits and tool calls never fire. Turn it on in Model Configuration. If it is already on and tool calls still fail, the model itself may be weak at structured tool use; switch to a model known for agentic coding.
Output gets truncated or context errors appear on long tasks
The Context Window or Max Output Tokens value does not match the model. Cline used a default. Set both to the model’s real limits.
Requests hit the wrong model or 404
The Model ID does not match what the upstream API expects. Custom providers pass it through verbatim, so anthropic/claude-sonnet-4-6 and claude-sonnet-4-6 are different strings to the API. Copy the exact id from the provider’s /models response.
Cost estimate looks wrong
The Input Price / Output Price fields are display-only estimates you entered by hand. They never affect requests or billing. Update them to your provider’s current rates if you want the in-session estimate to track reality.
The practical setup
For most Cline users, a clean two-or-three provider setup covers everything:
- Haimaker for cloud models and one-key routing across families.
- Ollama for local, private work.
- One premium model kept on hand for hard debugging and multi-file refactors.
That gives you local privacy when it matters, low-cost cloud models for routine work, and a stronger model when the task is expensive in attention, all from the same OpenAI Compatible provider type.
Setting up a different agent? See OpenCode Custom Provider Setup and OpenClaw Custom Provider Setup for the same pattern in those tools. For help picking a model to point at, see Best Models for OpenClaw.