Connect your tools

ozeye speaks the OpenAI API, so anything that can talk to OpenAI can talk to ozeye. There is exactly one value to get right.

Base URL: https://ozeye.ai/v1. Your client appends the rest of the path itself, so do not include /chat/completions unless a tool explicitly asks for a full endpoint URL.

Why that trips people up

Most tools ask for a “base URL” and then call <base>/chat/completions under the hood. If you paste the full endpoint into that field you end up requesting /v1/chat/completions/chat/completions, which 404s. Opening https://ozeye.ai/v1 in a browser returns a small JSON document describing the API, so you can always check what a tool should be pointed at.

Per-tool guides with complete config files: Coding agents (Claude Code, Claude Desktop, Codex, Cursor, Hermes, OpenClaw, OpenCode).

Hermes

Hermes has no built-in ozeye provider, so use its custom endpoint. Either run the wizard:

export OZEYE_API_KEY=oz_your_key_here
hermes model   # choose "Custom endpoint", base URL https://ozeye.ai/v1

or write the config directly:

# ~/.hermes/config.yaml
model:
  provider: custom
  base_url: https://ozeye.ai/v1
  key_env: OZEYE_API_KEY
  default: glm-5.3

OpenAI SDKs

from openai import OpenAI

client = OpenAI(
    base_url="https://ozeye.ai/v1",
    api_key="oz_your_key_here",
)

print(client.chat.completions.create(
    model="glm-5.3",
    messages=[{"role": "user", "content": "Hello"}],
).choices[0].message.content)

The JavaScript SDK takes the same two fields (baseURL and apiKey).

Anything that reads the OpenAI environment variables

Many CLIs and editors, including coding agents and IDE extensions, pick up the standard OpenAI variables. Export these and they will route to ozeye without further configuration:

export OPENAI_BASE_URL=https://ozeye.ai/v1
export OPENAI_API_KEY=oz_your_key_here

Editors and coding agents generally

  • Look for a setting named “OpenAI compatible”, “Custom provider”, “Base URL” or “API base”.
  • Set it to https://ozeye.ai/v1 and paste an ozeye key from Dashboard → API keys.
  • Pick a model id exactly as listed on the models page, for example glm-5.3. Model ids are not HuggingFace repository paths.
  • If the tool offers an “Anthropic” mode, that works too: ozeye serves the Anthropic Messages API at https://ozeye.ai, see Coding agents. Prefer the OpenAI mode when both exist.

Checking it works

A request that reaches ozeye and is billed to your account appears in Dashboard → Logs within a few seconds. If nothing appears, the tool is not reaching us and the base URL is the first thing to check.