Cline is a good agent for security work until the model behind it refuses to look at a proof of concept for a host you are paid to test. Its OpenAI Compatible provider fixes that: give it a base URL, a key and a model id, fill Model Configuration with the build’s window, output cap and prices, and a security-tuned or uncensored build plans, edits and runs commands inside VS Code, with separate builds on Plan and on Act if you want the split. The same ids work in Cursor, OpenCode and Claude Code; the condensed version of this page is at /agents/cline.
What you need
A key from the console with credit behind it, the Cline extension in VS Code, and a workspace you are cleared to work in: your own tooling, a client repository under an engagement, or a CTF challenge. Then click the settings icon (⚙️) at the top of Cline’s panel; the rest happens there. It shows the API Provider dropdown and, beneath it, the fields that provider wants.
The entry for this API is OpenAI Compatible: Cline’s generic provider for any server that speaks the OpenAI chat format at its own address. Do not pick the plain OpenAI entry; that one has its own page and no base URL to change.
Whatever you type is saved to ~/.cline/data/settings/providers.json. The VS Code extension, the cline CLI and the SDK all read that one file, and cline auth writes it from a terminal, so you enter the key a single time. The same settings area holds Use different models for Plan and Act, which lets each mode run its own build.
That switch matters because the modes do different work. Plan mode reads the code and discusses an approach with you; nothing changes on disk. Act mode writes the edits, runs the commands and reads what came back. A Plan turn is mostly input with a short reply, which a fast build handles well. An Act turn is where tool calls and edit precision decide the outcome, so the strong build pays for itself there. Leave the switch off and a single Model ID drives both.
From a blank panel to a first answer
- Make a key in the console.
- Click the Cline settings icon (⚙️) and choose OpenAI Compatible under API Provider.
- Base URL:
https://api.pentestllm.ai/v1. Keep the/v1; Cline appends/chat/completionson its own. - API Key: the full key, which begins
sk-ptl-. - Model ID: a single id, written as it appears on the models page. The three below are what Cline users on this API pick most often; the rest of the page assumes the first:
glm-5.3-uncensoredglm-5.3-flash-uncensoreddeepseek-v4.1-flash-abliterated- Expand Model Configuration under the model field and enter the build’s row from the table further down. Context Window and Max Output Tokens come straight from the catalog. Image Support stays off unless the model page shows image input. Computer Use goes on only for a build whose Tool calls column reads yes. Input Price and Output Price, per million tokens, come from the price table, which makes Cline’s per-task cost agree with the console.
- Press Verify. Cline fires one request at the base URL with your key and id and reports whether it answered.
- Optional: enable Use different models for Plan and Act in Cline Settings and assign a build to each mode. Plan reads and questions; Act edits and runs. The common split is a fast build like
glm-5.3-flash-uncensoredon Plan and a strong one likeglm-5.3-uncensoredon Act, and Cline swaps model as you swap mode.
Then put it to work: open the handler you are auditing and ask Plan mode where unvalidated input reaches a query, or hand Act mode a proof of concept for a host in scope and have it add logging before you run it. To see how the build answers before you install anything, the home page gives a few demo turns without an account, and the playground runs the same catalog in a browser.
Prefer the terminal? One CLI command writes the provider, key, model and base URL to the shared file:
cline auth -p openai-compatible -k "sk-ptl-live-…" -m glm-5.3-uncensored -b https://api.pentestllm.ai/v1From then on cline starts on that model; -m (--model) substitutes another id for one run. The extension sees the change the next time it reads the settings file.
Each setting, one line
- API Provider
- OpenAI Compatiblethe generic provider with its own base URL field; the CLI and SDK know it as openai-compatible
- Base URL
- https://api.pentestllm.ai/v1the address of every request; keep the /v1 and add nothing behind it
- API Key
- the whole sk-ptl- keythe bearer token on each call; it lives in providers.json under your home directory, never in the workspace
- Model ID
- glm-5.3-uncensoredthe id as spelled on the models page, not the display name; a provider entry holds one id at a time
- Context Window
- from the table belowthe ceiling Cline keeps the conversation under; it drops older messages before reaching it, so a wrong figure trims too soon or too late
- Max Output Tokens
- from the table belowthe longest reply Cline requests; an edit longer than this gets truncated
- Image Support
- off unless the model page lists image inputallows screenshots and pasted images in the task
- Computer Use
- only when Tool calls is yesreserved by Cline’s docs for builds with tool or function calling
- Input Price / Output Price
- per million tokens, from the price tablefeeds the cost figure Cline prints on each task; the console bills at the catalog rate regardless
- Use different models for Plan and Act
- optionala build per mode; Cline changes model with the mode and remembers both choices
Choosing the build
Every build in the catalog answers the OpenAI-format requests Cline sends. Late in a task, one turn resends the full conversation, each file Cline opened and each tool result, so input tokens are the bulk of the bill and the build with the cheapest input rate is cheapest per turn. Act mode edits and runs commands through tool calls, so only builds with a yes in that column belong in Act. The context and output columns hold the two values Model Configuration asks for.
| Model | One agent turn | Tool calls | Context Window | Max Output Tokens |
|---|---|---|---|---|
| GLM-5.3 Uncensored | $0.01 | model page | model page | model page |
| GLM-5.3 Flash Uncensored | $0.0047 | model page | model page | model page |
| DeepSeek V4.1 Flash Abliterated | $0.04 | model page | model page | model page |
| DeepSeek V4 Pro Uncensored | $0.04 | model page | model page | model page |
| GLM-5.3 Abliterated | $0.09 | model page | model page | model page |
| GLM-5.3 Flash Code Uncensored | $0.25 | model page | model page | model page |
| GLM-5.3 Uncensored | Uncensored | $0.282 | $0.846 |
| GLM-5.3 Flash Uncensored | Uncensored | $0.1053 | $0.3685 |
| DeepSeek V4.1 Flash Abliterated | Abliterated | $0.8488 | $1.6975 |
| DeepSeek V4 Pro Uncensored | Uncensored | $0.97 | $1.94 |
| GLM-5.3 Abliterated | Abliterated | $2.1825 | $4.365 |
| GLM-5.3 Flash Code Uncensored | Uncensored | $6.00 | $6.00 |
A real task costs less than turns times the per-turn figure, for two reasons. First, billing is per token from prepaid credit, and when a response reports cached input tokens those bill at the build’s cached rate rather than the input rate. A Cline task resends the same system prompt, rules and early messages every turn, so a long task’s input is mostly that repeated prefix. Second, the Plan and Act split puts the fast build on the read-only turns. The price table carries each build’s input and output rate per million tokens, the pair Model Configuration wants; the cached rate sits beside them on each model page.
Writing tooling or exploit code all day? The uncensored coding AI guide lines the code builds up and prices a full day on each.
Prove it works
Ask the API which models your key may call. Whatever you typed into Model ID has to be in the list:
curl https://api.pentestllm.ai/v1/models -H "Authorization: Bearer sk-ptl-live-…"Next, send one request shaped like Cline’s. The -i flag prints the response headers, which carry the request id and, on a non-streamed reply, the charge in USD.
curl -i https://api.pentestllm.ai/v1/chat/completions \ -H "Authorization: Bearer sk-ptl-live-…" \ -H "Content-Type: application/json" \ -d '{"model": "glm-5.3-uncensored", "messages": [{"role": "user", "content": "Reply with the single word ok."}]}'Back in Cline, start a task on one of your ids, then open the console’s usage page. The request is listed with its token counts and its charge, and the cost line Cline prints under the task matches once Model Configuration holds the catalog’s two prices.
When it breaks
401 invalid_api_key, shown by Cline as Invalid API Key: wrong or revoked key, or a stray space got pasted with it. Copy the key from the console once more into the API Key field.404 model_not_found, shown by Cline as Model Not Found: no such id on the models page. Compare the spelling; the id is the slug, never the display name.- Every request answers
404although the id is right: the Base URL carries/chat/completionsor something else after/v1. Cline adds the route itself; the field stops at/v1. 402 insufficient_balance: not enough credit for this request; top up in the console.402 monthly_limit_reached: the key reached the monthly cap you gave it.403 scope_denied: this key is restricted to a set of models and the one you asked for is outside it.429 rate_limited: the key is over its rate limit. Readretry-afteron the response.- Connection Errors with no status code at all: a typo in the Base URL, or a proxy or firewall between VS Code and the API, which is common on a locked-down testing laptop. Cline’s networking page documents the proxy settings; the first curl above tells you whether the machine reaches the API in the first place.
- Act mode keeps saying You did not use a tool in your previous response: Cline wanted a tool call and the build replied in prose. Move Act to a build marked yes under Tool calls and keep prose-only builds on Plan.
- The start of a long task vanishes from the history: as the window fills, Cline trims the oldest messages, and a Context Window value below the build’s real one makes it trim early. Enter the table’s value; Cline also retries with a shorter history when the API reports the window exceeded.
Questions
https://api.pentestllm.ai/v1, ending at /v1. Cline appends /chat/completions to it on each request, so a Base URL that already includes the route gets a 404 back.
The id exactly as the models page spells it, such as the first id under the steps above, never the display name. Cline forwards the string untouched, so a misspelling returns Model Not Found; the first curl under Prove it works prints every id the key may call. See the models page.
A key from this console, beginning sk-ptl-, pasted whole into the API Key field or given to cline auth with -k. One key covers every model in the catalog, and Cline stores it in providers.json where the extension, the CLI and the SDK all read it.
Tokens at the build’s listed price, drawn from prepaid credit, no subscription. The table above prices one agent-sized turn per build at the live rate, and once the catalog prices are in Model Configuration the figure Cline prints under each task agrees with the console.
Every model above is on the uncensored model list with its live price per million tokens. Per-token pricing sets the whole catalog side by side, and the family directory groups the builds by base model.