Connecting tools to go2llm

Any client that lets you set your own base URL works with go2llm without modification: the protocol is either Anthropic Messages or OpenAI Chat Completions. What changes between tools is only where the setting lives and which of the two protocols the tool speaks.

Written 2026-09-12.

The three fields every tool asks for

Whatever the UI looks like, you are filling in the same three values.

Base URL. For OpenAI-compatible clients: https://go2llm.tech/v1. For Anthropic-compatible clients and Claude Code: https://go2llm.tech — without /v1, because the Anthropic SDK appends it itself. Getting this wrong produces a request to /v1/v1/messages and a routing error, which is the single most common setup failure.

API key. A go2llm key, prefixed gk_sk_. OpenAI-compatible clients send it as Authorization: Bearer; Anthropic-compatible clients send it as x-api-key. The gateway accepts both, each on its own endpoint.

Model ID. The exact public ID from the catalog. It needs no key to read:

curl -s https://go2llm.tech/v1/models

Claude Code

Claude Code needs one thing that is documented almost nowhere: use ANTHROPIC_AUTH_TOKEN, not ANTHROPIC_API_KEY. If the machine has ever been logged into a personal claude.ai account, ANTHROPIC_API_KEY can lose to the stored login and the request silently goes to api.anthropic.com instead of your gateway — where a gk_sk_… key is unknown, and you get a 401 that looks like your key is broken.

export ANTHROPIC_BASE_URL=https://go2llm.tech
export ANTHROPIC_AUTH_TOKEN=YOUR_API_KEY
export ANTHROPIC_MODEL=claude-sonnet-5
unset ANTHROPIC_API_KEY
claude

OpenAI-compatible clients

The general pattern, which covers most desktop clients, chat UIs and SDK-based code:

from openai import OpenAI

client = OpenAI(
    base_url="https://go2llm.tech/v1",
    api_key="YOUR_API_KEY",
)

resp = client.chat.completions.create(
    model="gpt-5.6-sol",
    messages=[{"role": "user", "content": "Hello"}],
    max_tokens=200,
)
print(resp.usage)  # go2llm also returns cost_usd here

The same base URL and key work in Continue.dev, Roo Code, Kilo Code, Zed, Open WebUI, n8n and anything else that exposes a custom-provider form. Per-tool walkthroughs are below; the settings themselves are identical from one client to the next. Tools without an English page yet are covered in the Russian tools section.

Codex CLI connects through the Responses API. Point model_providers at https://go2llm.tech/v1 with wire_api = "responses". Verified with a live request on 2026-09-14: codex exec 0.151.0 on model qwen3-8-flash read a file, ran its tools and made a real edit. The interactive TUI has not been run, and only that one model was tested — /v1/responses serves the same models as chat.completions. Setup and the write-up of why this did not work before are on the Codex page.

Per-tool walkthroughs

Cursor custom API: Override OpenAI Base URL with a go2llm key — Point Cursor at go2llm through Override OpenAI Base URL: the exact fields, the Custom Models entry, and the honest status — we have not run a live request through it yet.
Roo Code custom API: setting up an OpenAI Compatible provider — Point Roo Code at go2llm: pick OpenAI Compatible as the provider, fill in Base URL, API Key and Model, and why native tool calling rules out some models entirely.

What does not work, stated plainly

Cursor is unverified. We have not tested it against go2llm and will not claim it works until we have.

Not published: embeddings, audio and video endpoints. Text goes through /v1/messages, /v1/chat/completions and /v1/responses; image generation through /v1/images/generations.

We also do not configure integrations on a customer's behalf — the instructions are the product here, not a service.

Checking the connection

One request tells you whether the tool is misconfigured or the key is:

curl -s -o /dev/null -w "%{http_code}\n" https://go2llm.tech/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "content-type: application/json" \
  -d '{"model":"gpt-5.6-sol","max_tokens":16,"messages":[{"role":"user","content":"ping"}]}'

200 means the credential path is fine and the problem is in the tool's settings. Anything else is covered in the error reference.

Get a key and make your first request in 3 minutes → go2llm.tech. Registration credits $1 to your balance. Full API reference: go2llm.tech/docs.