You have a gateway key and an editor extension. The extension wants a base URL, a key and a model name. Two of those are easy. The third, the model name, is where most "it does not work" reports start, because gateways list models with provider prefixes and the tool does not tell you the right string. This post gives the exact settings for Continue and Cline from their docs, then a script that looks up the model ID in AICredits' public catalog.
Disclosure: AICredits links on this site are affiliate links. I had no AICredits API key, so I did not run Continue or Cline against it. The settings below come from each tool's documentation, read on 2026-10-16. The catalog data comes from a live request to AICredits' public models endpoint, which needed no key.
How do you set a custom base URL in Continue?
Continue's OpenAI provider page says to keep provider: openai and change apiBase. Its example, with the placeholders as written in the docs:
models:
- name: <OPENAI_API_COMPATIBLE_PROVIDER_MODEL>
provider: openai
model: <MODEL_NAME>
apiBase: http://localhost:8000/v1
apiKey: <YOUR_CUSTOM_API_KEY>
For a hosted gateway, replace apiBase with the gateway's /v1 URL. AICredits' site gives its base URL as https://api.aicredits.in/v1. Put model as the exact catalog ID, such as openai/gpt-6-luna, and keep the key in an environment variable or secret store, not in a file you commit.
How do you set it in Cline?
From Cline's OpenAI Compatible provider page:
- Set API Provider to "OpenAI Compatible".
- Base URL: the page says to enter the base URL provided by your chosen provider and calls this a crucial step. It should not be the official OpenAI URL.
- API Key: your key from the provider.
- Model: choose or enter the model ID.
- Optionally open the model configuration section: max output tokens, context window, image support, a computer-use option described as for models with tool or function calling, and input and output prices.
That last section matters. If the context window you enter is larger than the model really supports, you get truncated or failed requests. If you enter it too small, the tool compacts your context sooner than needed. Take the number from the catalog, as below.
How do you find the exact model ID?
AICredits serves its catalog at GET https://api.aicredits.in/v1/models. I fetched it on 2026-10-16 without a key. It returned 432 entries, each with an id, context_length, mode, supported_apis, and prices in USD and INR per token. The script below searches it:
def find(models, needle, mode="chat"):
return [m for m in models
if needle.lower() in m["id"].lower() and m.get("mode") == mode and m.get("is_active")]
def fmt(m):
i = (m.get("input_cost_per_token_inr") or 0) * 1e6
o = (m.get("output_cost_per_token_inr") or 0) * 1e6
return f"{m['id']:<38}{m['context_length']:>9} Rs {i:>8.2f} in / Rs {o:>9.2f} out {','.join(m['supported_apis'])}"
Run with python find_models.py "sonnet-4.6,sonnet-5.5,gpt-6-luna,deepseek-v4.1,qwen3-coder". Output on 2026-10-16:
-- sonnet-4.6
anthropic/claude-sonnet-4.6 1000000 Rs 299.57 in / Rs 1497.83 out messages,chat_completions
-- sonnet-5.5
anthropic/claude-sonnet-5.5 1000000 Rs 199.71 in / Rs 998.55 out chat_completions
-- gpt-6-luna
openai/gpt-6-luna 1050000 Rs 9.99 in / Rs 49.93 out chat_completions
-- deepseek-v4.1
deepseek/deepseek-v4.1-flash 1048576 Rs 29.96 in / Rs 119.83 out chat_completions
-- qwen3-coder
qwen/qwen3-coder 262144 Rs 21.97 in / Rs 179.74 out chat_completions
qwen/qwen3-coder-30b-a3b-instruct 262144 Rs 6.99 in / Rs 26.96 out chat_completions
qwen/qwen3-coder-flash 1000000 Rs 19.47 in / Rs 97.36 out chat_completions
qwen/qwen3-coder-next 262144 Rs 11.98 in / Rs 79.88 out chat_completions
Prices are per million tokens in rupees, as the catalog listed them that day. They move with the exchange rate (the catalog listed 95.10014), so re-run before you budget. This list says nothing about which model is best for coding. It only tells you the string to paste and what it costs.
Two details worth noticing:
- Prefixes are part of the ID.
openai/gpt-6-luna, notgpt-6-luna. supported_apisdiffers.claude-sonnet-4.6listsmessagesas well aschat_completions.claude-sonnet-5.5listed onlychat_completions. An OpenAI-compatible tool such as Continue or Cline in this mode uses chat completions, so both work there. A tool that speaks the Anthropic Messages API natively would needmessages, so check this field before pointing such a tool at a gateway.
What goes wrong most often?
| Symptom | Likely cause | Fix |
|---|---|---|
| 401 on every request | Wrong or missing key. AICredits returns {"error":{"code":401,...,"type":"authentication_error"}} | Re-copy the key; check for trailing spaces |
| Model not found | Missing provider prefix, or a retired ID | Search the catalog with the script |
| Works, but answers get cut off | Context window or max output set too low in the tool | Use context_length from the catalog |
| Tool or agent mode loops | The model returns tool calls badly through this route | Run the conformance test on that exact model |
| Surprise bill | Large context on an expensive model | Check cost per request before long agent sessions |
Which tools did I not cover, and why?
Cursor and Open WebUI both offer ways to point at OpenAI-compatible endpoints, but the documentation pages I tried returned a 404 or redirected away, so I have no verified steps to give. I would rather leave them out than guess at setting names. If you use them, read their current docs and apply the same two rules: use the gateway's /v1 URL, and use the catalog's exact model ID.
To get an account, use the AICredits signup (affiliate link). For setting this up from Python instead of an editor, see using the base URL in Python frameworks. For the broader comparison, read the AICredits review.
What I could not verify
- Continue and Cline running against AICredits. I did not have a key.
- Setting names in Cursor and Open WebUI.
- Whether every catalog model works well in agent mode. The catalog says which APIs a model supports, not how reliable tool calls are.
