Token4S Model API

Token4S provides a model API. Connect it as a custom model in Ai4Scholar, or use it in your own scripts to organize academic search results from the Ai4Scholar API.

Choose your integration

The two sites manage accounts, API keys, and balances separately. Your Ai4Scholar key is for academic data and tool APIs; your Token4S key is for model APIs. With a custom model in Research Mode, the provider charges for model usage, while tools still use Ai4Scholar credits. Ai4Scholar built-in models do not require a Token4S account.

Connect to Research Mode

1. Visit Token4S, sign in, create an API key, and check its available models and balance.
2. Return to Ai4Scholar and open Settings → Model settings → Token4S.
3. Click Connect Token4S. The provider name, Base URL, and API format are already filled in; expand Advanced settings to change them.
4. Enter your own Token4S API key and fetch the available models. Select a model that supports tool calling, or enter its model ID manually.
5. Save the provider and test the connection. After a successful test, select the model for a new conversation.

The preset only fills in connection details. It does not create an account, purchase credits, or make a model request. A model appearing in the list still needs to pass the Research Mode tool calling test. Connection tests incur a small amount of model usage.

Academic search + model summaries

A typical workflow: search with Ai4Scholar → extract titles and abstracts → summarize with Token4S → retain paper IDs and links for verification.

This example sends only the paper information returned by search. Install requests, then set AI4SCHOLAR_API_KEY, TOKEN4S_API_KEY, and TOKEN4S_MODEL in your environment. Choose a currently available Token4S model that supports Chat Completions.

import json
import os
import requests
 
# Academic search uses the Ai4Scholar key.
search = requests.get(
    "https://ai4scholar.net/graph/v1/paper/search",
    headers={"Authorization": f"Bearer {os.environ['AI4SCHOLAR_API_KEY']}"},
    params={
        "query": "retrieval augmented generation evaluation",
        "limit": 5,
        "fields": "paperId,title,abstract,url,year",
    },
    timeout=60,
)
search.raise_for_status()
papers = search.json().get("data", [])
if not papers:
    raise SystemExit("No papers found. Try a different query.")
 
# Model inference uses a separate Token4S key.
completion = requests.post(
    "https://token4s.com/v1/chat/completions",
    headers={"Authorization": f"Bearer {os.environ['TOKEN4S_API_KEY']}"},
    json={
        "model": os.environ["TOKEN4S_MODEL"],
        "messages": [
            {
                "role": "system",
                "content": (
                    "Summarize research themes using only the supplied abstracts. "
                    "Cite paper IDs and URLs. Flag missing abstracts; "
                    "do not infer findings from titles alone."
                ),
            },
            {"role": "user", "content": json.dumps(papers, ensure_ascii=False)},
        ],
    },
    timeout=120,
)
completion.raise_for_status()
print(completion.json()["choices"][0]["message"]["content"])

Academic search and model inference are billed separately. Check summaries against the original papers; a model response is not a substitute for evidence. Consult each site's console for available models, pricing, and balances.

Continue integrating

• Ai4Scholar Open API: academic data endpoints and authentication.
• Plugins & Integrations: connect Dify, MCP, OpenClaw, and other tools.
• Explore Token4S: view model services and your usage.