ANIMICA/reference

Reference · verified 2026-08-14

Free AI inference API — no key required

Animica runs a free, OpenAI-compatible AI inference API at https://animica.dev/v1. There is no API key, no signup, and no billing: send a standard chat-completions request and you get a response, rate-limited to 30 requests/minute per IP. Compute comes from community GPU workers on the Animica network, so capacity is best-effort — this page documents exactly how to call it and what its real limits are.

Quickstart

curl -s https://animica.dev/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"kimi-k3","messages":[{"role":"user","content":"Hello"}]}'
from openai import OpenAI

client = OpenAI(base_url="https://animica.dev/v1", api_key="not-needed")
resp = client.chat.completions.create(
    model="kimi-k3",
    messages=[{"role": "user", "content": "Hello"}],
    timeout=600,   # completions typically take 1–3 minutes; prefer "stream": true
)
print(resp.choices[0].message.content)

Streaming uses the standard OpenAI wire format — add "stream": true and read Server-Sent Events.

Models

GET https://animica.dev/v1/models returns the live list. As of 2026-08-14:

Model idDescription
kimi-k3Default. Flagship coding & chat, served by the Animica miner network. May emit <think>…</think> reasoning blocks — strip them if you need clean output.
animica-chatStandard-tier chat routed through registered miners.
animica-chat-smallSmall tier — lighter, faster when serving.
animica-chat-flagshipFlagship tier — highest quality when serving.
animica-knowledgeENA — collaboratively trained by the network (LoRA head over Qwen2.5-1.5B).

Each model entry includes a boolean "serving" flag: whether a worker is live for that model right now. Treat it as the source of truth before sending work.

Honest limits

FAQ

Do I need an API key to use Animica's free AI inference API?

No. The API at https://animica.dev/v1 is keyless: there is no signup, no billing, and no Authorization header required. OpenAI SDKs demand a non-empty key string, so pass any placeholder such as "not-needed".

What models are available on the free API?

GET https://animica.dev/v1/models lists them live: kimi-k3 (default, flagship coding and chat), animica-chat, animica-chat-small, animica-chat-flagship, and animica-knowledge (ENA, trained collaboratively by the network). Each entry carries a boolean serving flag telling you whether a worker is currently serving that model.

What are the rate limits?

30 requests per minute per IP address, with a burst allowance of 12. Exceeding the limit returns HTTP 429. There is no paid tier on this endpoint and no published daily quota.

Why do requests sometimes hang or return 503?

Capacity is provided by community GPU workers on the Animica network, not a fixed server fleet. When a model's serving flag in /v1/models is false, requests may queue for a long time or fail instead of completing promptly — at the 2026-08-14 audit all models showed serving: false and a test request hung past 60 seconds. Check the flag first and set a generous client timeout. Media endpoints (/v1/images, /v1/videos, /v1/audio) return 202 {status:"queued", job_id, poll_url} when no GPU miner is online — queued jobs persist until a miner connects; a 503 no_media_miner means the dispatcher itself was unreachable and the submission did not enqueue — resubmit.

Can I use the official OpenAI SDK?

Yes. Any OpenAI-compatible client works: set base_url to https://animica.dev/v1 and pass a placeholder api_key. Chat completions, model listing, and streaming via "stream": true (Server-Sent Events) follow the OpenAI wire format.

Who pays for the compute, and is there an SLA?

Inference is served by community GPU workers on the Animica network, which earn ANM (the chain's native coin), with usage funded so that end users pay nothing. There is no SLA on the free endpoint: it is best-effort capacity, suitable for agents and development rather than latency-critical production.

Canonical docs