Documentation / quickstart

Build before breakfast.

SafeAPI speaks the protocols your tools already use. Oh My Pi is our recommended coding CLI, with automatic model discovery and Responses support.

01 / Recommended: Oh My Pi

Oh My Pi installs the omp command. This setup uses openai-responses and discovers your available model IDs directly from SafeAPI. Your API key belongs in the environment variable—not in the YAML file.

Terminal
curl -fsSL https://omp.sh/install | sh
mkdir -p ~/.omp/agent
cat > ~/.omp/agent/models.yml <<'YAML'
providers:
  safeapi:
    baseUrl: https://safeapi.cc/v1
    apiKey: SAFEAPI_API_KEY
    api: openai-responses
    discovery:
      type: openai-models-list
YAML

export SAFEAPI_API_KEY='sk-safe_...'
omp models refresh safeapi
omp -p --model safeapi/gpt-6.2-astra "Reply with only OK"

For Chat Completions instead, change api: openai-responses to api: openai-completions. For Anthropic Messages, use api: anthropic-messages with auth: apiKey.

02 / OpenAI-compatible request

The base URL is https://safeapi.cc/v1. Authenticate with Authorization: Bearer sk-safe_....

curl
curl https://safeapi.cc/v1/chat/completions \
  -H "Authorization: Bearer $SAFEAPI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-6.2-astra","messages":[{"role":"user","content":"Hello"}]}'
JavaScript
import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://safeapi.cc/v1",
  apiKey: process.env.SAFEAPI_API_KEY,
});

const response = await client.chat.completions.create({
  model: "gpt-6.2-astra",
  messages: [{ role: "user", content: "Hello" }],
});

03 / Streaming

Set stream: true. SafeAPI forwards SSE events as they arrive and cancels the upstream request if your client disconnects. For Chat Completions, SafeAPI requests the final usage chunk automatically so billing does not depend on your SDK configuration.

Stream with curl
curl -N https://safeapi.cc/v1/chat/completions \
  -H "Authorization: Bearer $SAFEAPI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-6.2-astra","stream":true,"messages":[{"role":"user","content":"Hello"}]}'

04 / Claude Code and Anthropic SDKs

Anthropic clients use the base URL without /v1; the client appends the route itself.

Claude Code
export ANTHROPIC_BASE_URL=https://safeapi.cc
export ANTHROPIC_AUTH_TOKEN=sk-safe_...
claude

05 / Other coding agents

ClientProvider / protocolBase URL
Cline / Roo CodeOpenAI Compatiblehttps://safeapi.cc/v1
Codex CLIResponseshttps://safeapi.cc/v1
ContinueOpenAIhttps://safeapi.cc/v1
Claude CodeAnthropic Messageshttps://safeapi.cc

06 / Errors and retries

401 means a missing, invalid, or revoked key. 402 means insufficient available balance. 429 means a user or capacity limit; honor Retry-After. Retry upstream 5xx errors with exponential backoff, but never automatically repeat a stream after output has started.

07 / Privacy

SafeAPI logs request metadata needed for billing and reliability: account, model, token counts, timings, status, and cost. It does not log prompts, completions, authorization headers, or API keys.