API Access

OpenAI-compatiblemodel: awareliquid-m1streaming

M1 is a 1.1B hybrid model with an MT-v2s liquid-recurrence adapter (fast-weight episodic memory + selective decay — the architecture documented in our benchmarks). You can call it with any OpenAI SDK or plain HTTP.

1 Get a token / 获取令牌

Sign in at relay.awareliquid.ai and create an API token (console → Tokens). Quotas and usage are shown per token. 在控制台注册并创建令牌,配额与用量按令牌统计。

2 Call the model

curl https://relay.awareliquid.ai/v1/chat/completions \
  -H "Authorization: Bearer $YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "awareliquid-m1",
    "messages": [{"role": "user", "content": "用一句话介绍你自己"}],
    "max_tokens": 128,
    "stream": true
  }'

Or with the OpenAI Python SDK:

from openai import OpenAI
client = OpenAI(base_url="https://relay.awareliquid.ai/v1",
                api_key="YOUR_TOKEN")
r = client.chat.completions.create(
    model="awareliquid-m1",
    messages=[{"role": "user", "content": "Hello!"}],
)
print(r.choices[0].message.content)

Parameters

field	notes
`model`	`awareliquid-m1`
`messages`	standard OpenAI roles; your `system` message is honored verbatim
`max_tokens`	capped at 512
`temperature`, `top_p`	`temperature: 0` = greedy
`stream`	SSE chunks in OpenAI format, ends with `[DONE]`

Notes / 说明

Each request runs with a fresh recurrent state; the API does not use the chat UI's persistent memory. 每次请求使用干净的递归状态,不注入网页版的记忆。
CPU-served; expect ~seconds-per-response latency. Streaming is recommended. 当前为 CPU 推理,建议开启流式。
Model card, honest benchmarks and retractions: BENCHMARKS.md.