omnicall-agent
v0.1.0
Published
A tiny autonomous research agent that runs entirely on Omnicall — 248 LLMs + live web search through one keyless, pay-per-call endpoint. No OpenAI/Anthropic key, no signup.
Maintainers
Readme
omnicall-agent
An autonomous AI agent in ~150 lines — with no API keys. The brain (248 LLMs) and the tools (live web search) are one keyless, pay-per-call endpoint: Omnicall. Pay per call in USDC via x402 on Base/Solana, or with a gck_ credit key. It runs free out of the box.
npx omnicall-agent "What is the x402 protocol and who is building on it?"That's it — no OpenAI key, no Anthropic key, no signup. The agent picks a model, searches the web when it needs facts, and answers.
Why this exists
Most "agent starter kits" make you sign up for an LLM provider, paste a key, add a billing card, then wire up a separate search API with its key. omnicall-agent needs none of that. One endpoint gives an agent every model plus tools, and it pays per call — so you can ship an autonomous agent with zero accounts and pay only for what it actually does.
Two modes
| Mode | Setup | What you get |
|------|-------|--------------|
| Free | nothing | Single-shot answer on the free tier. Instant. |
| Agentic | OMNICALL_KEY | Full reason→search→verify loop over 248 premium models + live web search. |
# Free — runs immediately
npx omnicall-agent "Summarize the latest on agent payments"
# Agentic — unlock tools + premium models with a credit key (or x402 wallet)
export OMNICALL_KEY=gck_your_key # get one at https://omnicall.gocreativeai.com
npx omnicall-agent "Research USDC stablecoin volume trends and cite sources"Environment
| Var | Default | Purpose |
|-----|---------|---------|
| OMNICALL_KEY | (none) | gck_ credit key or x402-funded key. Unlocks the agentic loop. |
| OMNICALL_MODEL | auto | Model id. auto routes to the cheapest capable model. |
| OMNICALL_URL | https://omnicall.gocreativeai.com | Gateway base URL. |
How it works
The agent is a simple loop against Omnicall's OpenAI-compatible gateway:
- Send the goal to a model (
POST /v1/chat/completions,model: "auto"). - If the model asks to look something up (
SEARCH: <query>), the agent callsGET /v1/web/<query>and feeds the result back. - Repeat until the model returns a
FINAL:answer.
Every call is metered and paid per-use — no subscription, no idle cost.
Build your own on the same rails
- OpenAI SDK drop-in:
base_url="https://omnicall.gocreativeai.com/v1" - MCP server:
npx -y omnicall-mcp(orclaude mcp add --transport http omnicall https://omnicall.gocreativeai.com/mcp) - 248 models, image/video/voice/music, and live crypto/markets/web/research — all keyless, all pay-per-call.
MIT licensed.
