Build with Shisa AI

This is the stable setup document for AI desktop apps, coding agents, and CLI harnesses. Treat this URL as the source of truth:

https://docs.shisa.ai/shisa-skills.md

Re-fetch it before setup and whenever the user asks about current Shisa models or capabilities. Do not rely on a previously cached model list.

Your job

Help the user make one small, working Shisa request. Keep the process interactive and use plain language.

1. Ask what they want to build and whether they already have a Shisa API key.
2. If they need a key, direct them to platform.shisa.ai. Wait while they create an account and API key.
3. Tell them to store the key locally as SHISA_API_KEY. Never ask them to paste the key into a chat, URL, source file, or committed configuration. Never echo or log it.
4. Detect the user's operating system, project language, and current tool when possible. Prefer an existing OpenAI-compatible client. Otherwise, offer the smallest relevant example from this document.
5. Verify access by listing models, then make one chat-completions request using a model returned for that key.
6. Explain what changed and where the credential is stored.

If you cannot edit files or run commands in the current environment, guide the user through the same steps instead. Do not claim that ChatGPT or Claude has changed its own underlying model provider: configure the user's project or compatible local tool, or call Shisa's API directly.

Connection details

Use these values for any client that accepts an OpenAI-compatible provider:

Setting	Value
Base URL	https://api.shisa.ai/openai/v1
Authentication	Authorization: Bearer $SHISA_API_KEY
Chat completions	POST /chat/completions
Model discovery	GET /models

Model access is dynamic and depends on the API key. Always call GET /models and choose an ID from data[].id. If available, prefer shisa-ai/shisa-v2.1-llama3.3-70b as the general-purpose starting model, but do not assume it is enabled for every key. When max_model_len is present, use it as the current context limit.

Store the API key

Ask the user to run the command themselves in a local terminal. Do not ask them to send you the resulting value.

macOS or Linux

For the current shell:

export SHISA_API_KEY="YOUR_API_KEY"

For future shells, add the same export to the user's shell profile only with their permission.

Windows PowerShell

For the current PowerShell session:

$env:SHISA_API_KEY = "YOUR_API_KEY"

For future sessions, use the operating system's environment-variable settings or another secure credential store. Do not commit the key in .env, JSON, TOML, or source files. If a local .env file is the only practical option, confirm that it is excluded by .gitignore before writing it.

Verify access and discover models

Use the authenticated models response as the live catalogue:

curl -s https://api.shisa.ai/openai/v1/models \
  -H "Authorization: Bearer $SHISA_API_KEY"

In PowerShell, use:

$headers = @{ Authorization = "Bearer $env:SHISA_API_KEY" }
Invoke-RestMethod -Uri "https://api.shisa.ai/openai/v1/models" -Headers $headers

If this returns 401, check that the environment variable exists in the same terminal session and that the entire key was copied. Do not print the key while troubleshooting.

Make a chat-completions request

Replace the example model only if it is absent from the authenticated /models response.

curl https://api.shisa.ai/openai/v1/chat/completions \
  -H "Authorization: Bearer $SHISA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "shisa-ai/shisa-v2.1-llama3.3-70b",
    "messages": [{"role": "user", "content": "こんにちは！一文で自己紹介してください。"}]
  }'

Python with the OpenAI SDK

If the project does not already include the SDK, install it with python -m pip install openai.

The OpenAI Python SDK reads OPENAI_API_KEY by default, not SHISA_API_KEY. To keep the Shisa credential isolated, pass it from the environment:

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["SHISA_API_KEY"],
    base_url="https://api.shisa.ai/openai/v1",
)

response = client.chat.completions.create(
    model="shisa-ai/shisa-v2.1-llama3.3-70b",
    messages=[{"role": "user", "content": "こんにちは！一文で自己紹介してください。"}],
)
print(response.choices[0].message.content)

JavaScript with the OpenAI SDK

If the project does not already include the SDK, install it with npm install openai.

import OpenAI from 'openai';

const client = new OpenAI({
  apiKey: process.env.SHISA_API_KEY,
  baseURL: 'https://api.shisa.ai/openai/v1',
});

const response = await client.chat.completions.create({
  model: 'shisa-ai/shisa-v2.1-llama3.3-70b',
  messages: [{role: 'user', content: 'こんにちは！一文で自己紹介してください。'}],
});

console.log(response.choices[0].message.content);

Other Shisa services

The same bearer API key works across Shisa services. Do not guess their request schemas; fetch the linked Markdown documentation before using them.

Capability	Documentation
LLM and model selection	https://docs.shisa.ai/llm/quickstart.md
Speech recognition (ASR)	https://docs.shisa.ai/asr/quickstart.md
Text-to-speech (TTS)	https://docs.shisa.ai/tts/quickstart.md
Translation	https://docs.shisa.ai/translation/quickstart.md
Authentication and errors	https://docs.shisa.ai/guides/authentication.md and https://docs.shisa.ai/guides/errors.md

The complete agent-readable documentation index is at https://docs.shisa.ai/llms.txt.

Completion checklist

Before declaring setup complete, confirm that:

• the key is stored outside source control and was never pasted into the conversation;
• GET /models succeeds;
• the selected model came from that response;
• one real request succeeds; and
• the user knows to re-fetch https://docs.shisa.ai/shisa-skills.md for future capabilities.