LibraOS model API

The LibraOS model API serves models from several providers behind one key. Use it from the self-hosted Runtime, from your own code, or from a coding agent such as Claude Code or Codex.

	Base URL
OpenAI-compatible	https://api.libraos.com/v1
Anthropic-compatible	https://api.libraos.com

The endpoints are chat completions (POST /v1/chat/completions), embeddings (POST /v1/embeddings), the model list (GET /v1/models), the OpenAI Responses API (POST /v1/responses) and the Anthropic Messages API (POST /v1/messages). The model API does not provide web search; see Web search.

Get a key

1. Sign in at console.libraos.com.
2. Open API keys and choose Create a key under Model API keys.
3. Give it a name. Optionally limit it to some models (comma-separated; empty allows every model) and set its requests per minute.
4. Copy the secret. It starts with lk_ and is shown once; LibraOS keeps only a fingerprint of it.

Send the key as Authorization: Bearer lk_… or x-api-key: lk_…. Keep keys on servers: never put one in a web page or a mobile app. Manage on a key rotates it (the old secret stops at once) or revokes it.

Model names

Name a model with its provider prefix, as on the price list on the Console's Credits & billing page, for example:

• anthropic/claude-opus-5-5
• openai/gpt-6.1-sol
• openai/gpt-5.6-sol
• Qwen/Qwen3-Embedding-8B (embeddings)

A bare name that matches exactly one model the key may call also works. GET /v1/models lists the models a key may call:

curl https://api.libraos.com/v1/models \
  -H "Authorization: Bearer $LIBRAOS_API_KEY"

Clients

These are the Console's own examples. Replace lk_… with your key, or read it from an environment variable.

curl

curl https://api.libraos.com/v1/chat/completions \
  -H "Authorization: Bearer $LIBRAOS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"openai/gpt-5.6-terra","messages":[{"role":"user","content":"Hello"}]}'

OpenAI SDK and other OpenAI-compatible clients:

from openai import OpenAI
client = OpenAI(base_url="https://api.libraos.com/v1", api_key="lk_…")

Anthropic SDK:

from anthropic import Anthropic
client = Anthropic(base_url="https://api.libraos.com", api_key="lk_…")

Claude Code, with an anthropic/… model:

export ANTHROPIC_BASE_URL=https://api.libraos.com
export ANTHROPIC_API_KEY=lk_…
claude --model anthropic/claude-sonnet-5-5

Codex, with an openai/… model. In ~/.codex/config.toml:

model_provider = "libraos"
model = "openai/gpt-5.6-terra"
[model_providers.libraos]
name = "LibraOS"
base_url = "https://api.libraos.com/v1"
env_key = "LIBRAOS_API_KEY"
wire_api = "responses"

then export LIBRAOS_API_KEY=lk_….

The self-hosted Runtime

Point the Runtime's model settings at the OpenAI-compatible base URL:

export OPENAI_API_BASE=https://api.libraos.com/v1
export OPENAI_API_KEY='lk_...'
export OPENAI_MODEL='anthropic/claude-opus-5-5'

With no embedding settings, the Runtime also embeds through this endpoint with Qwen/Qwen3-Embedding-8B. See Model settings for the other model slots, and Getting started for a full first run.

Any other OpenAI-compatible gateway, or a local model server, also works with the Runtime; the LibraOS model API is one choice.

Billing and limits

Calls are billed to the key's account, in LibraOS credits or against a Token Plan; see Credits and billing. Calls with your own provider key or a local model are not charged by LibraOS.

Response	Meaning
401 invalid_api_key	No key, or a key that was revoked or rotated.
402 insufficient_credits	The account's credits are used up. Buy credits on Credits & billing.
403 model_not_allowed	The key may not call this model. Check GET /v1/models and the key's model list.
403 account_suspended	The key's account is suspended or closed.
429 rate_limit_exceeded	The key's requests-per-minute limit.
429 concurrency_limit_exceeded	Too many calls from the account running at once.
429 plan_limit_reached	A Token Plan limit is reached and Use credits after the plan limit is off.
503 service_unavailable	The API is starting. Retry in a few seconds.