Ship AI you can actually prove.
OpenAI-compatible inference on a post-quantum chain. Every response returns a signed, replayable proof-of-inference receipt — verifiable offline, by anyone.
# OpenAI-compatible — every response carries X-Animica-Receipt
curl https://animica.dev/v1/chat/completions \
-H "Authorization: Bearer $ANIMICA_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "qwen2.5:7b",
"messages": [{"role":"user","content":"Explain post-quantum signatures."}]
}'# Drop-in: the OpenAI SDK, pointed at Animica
from openai import OpenAI
client = OpenAI(base_url="https://animica.dev/v1", api_key="$ANIMICA_API_KEY")
r = client.chat.completions.create(
model="qwen2.5:7b",
messages=[{"role":"user", "content":"Hello, Animica"}],
)
print(r.choices[0].message.content)
# r.animica_receipt → signed proof-of-inference# Verify + replay a receipt offline — no server, no trust
pip install animica
animica ai receipt verify receipt.json
# ✔ receipt a7ce980a… valid=True signature_ok=True
animica ai replay receipt.json --prompt "the original prompt"
# reproducible: verified match=True backend=deterministic# Run your own gateway — receipts on by default
pip install "animica[backend]"
animica ai serve --host 0.0.0.0 --port 8080 --api-key $KEY
# OpenAI-compatible /v1 + /v1/verify + /v1/signer + /v1/router/statusProof, not promises.
Most AI APIs ask you to trust the black box. Animica hands you a receipt. Built on a post-quantum chain, every layer is designed to be checked — by you, offline, forever.
Post-quantum by default
Accounts and receipts are signed with ML-DSA-65 (FIPS-204) — lattice cryptography that stands up to quantum attack.
Proof-of-inference
A content-hashed, signed record of model, prompt, output and seed — replayable and verifiable with zero trust in the server.
Quantum-seeded sampling
Sampling seeds come from the network randomness beacon — attested hardware QRNG when available — and are recorded on the receipt.
Drop-in compatible
It's just the OpenAI API. Keep your SDK, your prompts, your tooling — point the base URL at Animica and get proofs for free.
Every receipt tells the truth about itself.
Animica never overclaims. Each receipt declares exactly how far it can be reproduced — so an auditor knows what a green check really means.
Deterministic or seed-honoring in-process backends: the output is re-derived and matched, hash for hash.
Remote or GPU backends drift across kernels. The signature still proves origin; replay is not asserted.
The backend offers no path to reproduction. The receipt says so plainly, rather than pretending.
{
"kind": "inference",
"model_id": "qwen2.5:7b",
"provider_key": "ollama",
"prompt_hash": "9f21a4c7…e0b3",
"output_hash": "b8d0f16a…7cc2",
"quantum_seed": { "beacon_round": 184402,
"attested": true },
"signature_alg": "ml_dsa_65",
"signature_domain": "animica.ai.proof-of-inference.v1",
"signer_address": "anim1zqp…q5tqgv",
"chain_id": 1
}One chain, running now.
Everything Animica, one map.
From the RPC that powers your node to the apps people play with — the full stack, cross-linked.
Build
Network
Wallets & assets
Apps
Build on proof.
Grab a key, point your OpenAI client at Animica, and start shipping inference you — and your users — can verify.