Features

• No API key required — uses your browser session token
• Persistent sessions — messages stay in the same conversation automatically
• Streaming support — yield reply chunks in real time
• Web search — toggle Kimi's built-in search on any message
• Multi-turn conversations — full context preserved across calls
• Image upload — send up to 20 images at once
• File upload — send documents, PDFs, code files (max 100 MB each, 1,000 files, 10 GB total)
• Built-in Models — switch models easily with Models.K2D6, Models.K2D6_AGENT, etc.
• Lightweight — only depends on requests

Install

pip install kimi-nocost

Quick Start

from kimi_nocost import KimiClient

client = KimiClient(token="your-kimi-bearer-token", model="kimi")

reply = client.chat("What is the capital of France?")
print(reply)

model is required. Pass it as a plain string — "kimi", "k1", "k2d6", etc. Omitting it raises a TypeError.

Models

Use the built-in Models class to switch between Kimi models:

from kimi_nocost import KimiClient, Models

client = KimiClient(token="your-token", model=Models.KIMI)

reply = client.chat("Explain transformers")
print(reply)

client.model = Models.K2D6
reply = client.chat("Solve this step by step: 2x + 5 = 13")
print(reply)

Multi-Turn Conversation

client = KimiClient(token="your-token", model="kimi")

reply1 = client.chat("My name is Alex.")
reply2 = client.chat("What is my name?")
print(reply2)

Streaming

from kimi_nocost import KimiClient

client = KimiClient(token="your-token", model="kimi")

for chunk in client.stream("Explain quantum computing in simple terms"):
    print(chunk, end="", flush=True)

With Web Search

reply = client.chat("Latest news on AI today", use_search=True)

Image Upload

Send up to 20 images in a single call:

from kimi_nocost import KimiClient

client = KimiClient(token="your-token", model="kimi")

file_ids = client.upload_images([
    "photo1.jpg",
    "photo2.png",
    (open("screenshot.png", "rb"), "screenshot.png"),
])

reply = client.chat("Describe these images", refs=file_ids)
print(reply)

File Upload

Send documents, PDFs, or code files (max 100 MB each, 1,000 files per session, 10 GB total):

file_id = client.upload_file("report.pdf")

reply = client.chat("Summarize this document", refs=[file_id])
print(reply)

You can also pass raw bytes with a filename:

import io

content = b"def hello(): print('world')"
file_id = client.upload_file(io.BytesIO(content), "hello.py")

reply = client.chat("What does this code do?", refs=[file_id])
print(reply)

Starting a New Chat

client.new_chat(name="fresh-start")
reply = client.chat("Let's start over.")

Manual Chat ID

chat_id = client.new_chat(name="my-session")

reply1 = client.chat("My name is Alex.", chat_id=chat_id)
reply2 = client.chat("What is my name?", chat_id=chat_id)
print(reply2)

API Reference

KimiClient(token, model, timeout=60)

Methods

Attributes

REST API Server

The included Flask server exposes HTTP endpoints:

Pass your token via Authorization: Bearer <token> header.

Chat with files:

{
  "message": "Summarize this",
  "file_ids": ["id1", "id2"],
  "model": "k2d6-agent"
}

Upload images (multipart, field name images):

curl -H "Authorization: Bearer <token>" \
  -F "images=@photo1.jpg" \
  -F "images=@photo2.png" \
  http://localhost:8080/upload/images

Upload file (multipart, field name file):

curl -H "Authorization: Bearer <token>" \
  -F "file=@report.pdf" \
  http://localhost:8080/upload/file

Getting Your Token

1. Open kimi.moonshot.cn in your browser
2. Open DevTools → Network → any request
3. Copy the value of Authorization: Bearer <token>

Or from iOS/Android using a MITM proxy (e.g. mitmproxy, Charles).

The token is a JWT that lasts ~30 days.

Demo

See DEMO.md for a full working Telegram bot with image & file upload support.

Community

Join for Python tutorials, projects, and updates → t.me/pythontodayz

License

MIT — made by addy