Cline with an uncensored model

Where the settings are

In VS Code, the settings icon (⚙️) at the top of the Cline panel opens the provider settings: the API Provider dropdown, and under it the fields the chosen provider needs. For this API the provider is OpenAI Compatible, which is Cline’s entry for any endpoint that answers the OpenAI chat format from a base URL of its own. The official OpenAI entry is a separate provider with its own settings page.

What you enter there is stored in ~/.cline/data/settings/providers.json, which the IDE extension, the cline CLI and the SDK share. The CLI writes the same file with cline auth, so the key is entered once. Cline Settings also holds Use different models for Plan and Act, which gives each mode its own model.

The two modes are the reason that switch exists. Plan mode reads the code and talks the approach through with you before anything changes; Act mode makes the edits, runs the commands and reads their output. Plan turns are mostly input tokens and a short answer, so a fast build does them well; Act turns are where tool calls and edit quality count, so the strong build earns its price there. With the switch off, one model does both and the Model ID field is the only choice to make.

Steps

1. Create a key in the console.
2. Open the Cline settings (⚙️) and set API Provider to OpenAI Compatible.
3. Base URL: https://api.uncensoredllm.ai/v1. The /v1 stays; Cline adds /chat/completions itself.
4. API Key: the whole key, starting sk-ullm-.
5. Model ID: one id, spelled as the models page shows it. These three are the ones Cline users here run most; the first is the default this guide assumes:

6. Open Model Configuration under the model field and type the build’s values from the table below: Context Window and Max Output Tokens as the catalog lists them, Image Support off unless the model page lists image input, and Computer Use on only when the build’s Tool calls column says yes. Input Price and Output Price per million tokens come from the price table further down, so the cost Cline shows on each task matches the console.
7. Click Verify to confirm the connection: Cline sends one request to the base URL with the key and the id.
8. Optional: in Cline Settings, turn on Use different models for Plan and Act and pick one model for each mode. Plan mode reads and asks; Act mode edits files and runs commands. A fast build such as glm-5.3-flash-uncensored for Plan and a strong one such as glm-5.3-uncensored for Act is the usual split, and Cline switches model with the mode.

From a terminal, the CLI stores the same provider, key, model and base URL in one command:

After that, cline runs with that model, and -m (--model) swaps in another id for a single run. The extension picks the change up from the shared settings file.

What each setting does

API Provider

OpenAI Compatible

the provider entry that takes a base URL of its own; the CLI and SDK call it openai-compatible

Base URL

https://api.uncensoredllm.ai/v1

where Cline sends every request; the /v1 at the end matters, and nothing goes after it

API Key

the whole sk-ullm- key

sent as the bearer token on each request; stored in providers.json, not in the project

Model ID

glm-5.3-uncensored

the id exactly as the models page shows it, not the display name; one id per provider entry at a time

Context Window

from the table below

the size Cline keeps the conversation under; it trims older messages before the window fills, so a wrong value trims too early or too late

Max Output Tokens

from the table below

the longest answer Cline asks for; an edit that needs more than this is cut off

Image Support

off unless the model page lists image input

lets Cline attach screenshots and pasted images

Computer Use

only when Tool calls is yes

Cline’s docs reserve it for models with tool or function calling

Input Price / Output Price

per million tokens, from the price table

what Cline uses for the cost it shows on each task; the console bills at the catalog price either way

Use different models for Plan and Act

optional

one model per mode; Cline switches with the mode and keeps both selections

Picking a model

Cline sends OpenAI-format requests, and every build on the models page answers in that format. An agent turn carries the conversation, the files Cline read and every tool result so far, so input tokens are most of the bill and a build with a low input price costs the least per turn. Act mode runs on tool calls for its edits and commands, so the Tool calls column decides which builds belong in Act; the context and output columns are the two Model Configuration values to type for each.

40,000 input and 1,200 output tokens at the catalog prices on 2026-11-09. Tool calls, context and output as the catalog publishes them; where it does not, the model page says.

Two things make a real session cost less than the turn figure times the number of turns. Requests bill per token at the model’s price from your prepaid balance, and when a response reports cached input tokens they bill at the model’s cached rate instead of the input rate; a Cline task repeats the same system prompt, rules and early conversation on every turn, so most of a long task’s input is that repeated prefix. And the Plan and Act split above puts the fast build on the turns that only read. The price table carries the input and output rates per million tokens, the two numbers to type into Model Configuration; each model page lists the cached rate beside them.

For coding specifically, the uncensored coding AI guide compares the code builds and costs a full coding day on each.

Check it

List the models your key can call; the id in the Model ID field should be in the answer:

Then send one request the way Cline does. -i prints the headers: x-uplink-request-id names the request, and on a non-streamed answer x-uplink-charge is what it cost in USD.

In Cline, start a task with one of your ids selected. The request shows on the console’s usage page with its tokens and its charge, and the cost Cline prints under the task matches it when the two prices in Model Configuration are the catalog’s.

Errors and fixes

• 401 invalid_api_key, which Cline reports as Invalid API Key: the key is wrong, revoked or pasted with a space. Copy it again from the console into the API Key field.
• 404 model_not_found, Cline’s Model Not Found: the Model ID is not on the models page. Check the spelling against the page; the id is the slug, not the display name.
• A 404 on every request with a correct id: the Base URL has /chat/completions or a trailing path after /v1. Cline appends the route itself; the field ends at /v1.
• 402 insufficient_balance: the balance does not cover the request. Top up in the console. 402 monthly_limit_reached means the key hit the monthly spend limit you set on it.
• 403 scope_denied: the key is limited to certain models and this is not one of them.
• 429 rate_limited: the key hit its rate limit. The response carries retry-after.
• Connection Errors before any status code: the Base URL has a typo, or a proxy or firewall sits between VS Code and the API. Cline’s networking page covers the proxy settings; the first curl above shows whether the machine reaches the API at all.
• Act mode repeats You did not use a tool in your previous response: the build answered in prose where Cline expected a tool call. Switch the Act model to a build whose Tool calls column says yes, and keep the prose-only builds for Plan.
• Early parts of a long task drop out of the conversation: Cline trims older messages as the context window fills, and a Context Window value smaller than the build’s makes it trim sooner. Type the value from the table; Cline also retries with a shorter conversation when the API reports the window exceeded.

Questions