Use the OpenAI API locally just like ChatGPT.
Logs data to a local database and tracks token usage.

https://cornpip.github.io/TokenMeter/

To use Token-Meter, an OpenAI API key is needed.
Please use this as a reference — it’s a mock site.

• Environment where Docker Engine is running
• docker-compose must be available

1. Download or copy the docker-compose.yml file

2. Run the command docker compose up (docker compose pull before docker compose up is the latest update.)

3. Service URL: http://localhost/token_meter/viewer/main

By default, the service runs on port 80. To change this, modify {your_port} in the docker-compose.yml file:

// docker-compose.yml
...
    nginx:
        environment:
            <<: *common-environment
        image: cornpip77/token-meter-nginx
        ports:
            - {your_port}:80
...

• Node.js must be installed on your system
• Docker

1. git clone https://github.com/cornpip/TokenMeter.git
2. npm run start
3. run ai_server (It may take a few minutes the first time you run it.)
   1. docker pull cornpip77/token-meter-ai:latest
   2. run docker script

docker run -d --gpus all -p 7775:7775 --name token-meter-local-ai -w /app/server cornpip77/token-meter-ai bash -c "uvicorn main:app --host 0.0.0.0 --port 7775"

Service URL: http://localhost:7777/token_meter/viewer/main

default port
token_ai_server: 7775, token_server: 7776, token_client: 7777

• To back up or change the database, copy/replace/delete the ./server/database.db file
• If the database file is missing, a new one will be automatically created when the app starts.

Initial settings:
openai-api-key – You need to register your API key.
Model – Select an existing model or manually input a new one when available.
Maximum send message count – Determines how much chat history (excluding the assistant’s upcoming replies) is included in each message sent to the API.
system instruction – Add a default system instruction for GPT

Example: Maximum Send Message Count

Chat History
Q1 → A1 → Q2 → A2 → Q3 → A3

If you're sending Q3 with a message count of:
2 → History includes [A2, Q3]
3 → History includes [Q2, A2, Q3]
4 → History includes [A1, Q2, A2, Q3]

Total token usage = history + token usage for the response to Q3

Tip: As of February 2024, no automatic token optimization is available. A message count between 3–5 is recommended.

Refer to the OpenAI Pricing Page for token costs per model. OpenAI Pricing Page

After a response is generated, the token count is displayed next to each message:

• The number beside a question is the total token usage (history + current message).
• The number beside an answer represents the token usage of the response only.

Click on a token count to view the message range used and detailed token stats.

You can attach image files via drag-and-drop or by clicking the paperclip icon.

Note: As of February 2024, attachments are not saved to chat history. However, their content is included in token calculations. (You can see above — A short sentence and a file resulted in 1378 tokens)

During response generation, the input box is disabled. Real-time (streaming) responses are planned for a future update.

Click the "Token-Meter (Image)" menu to go to the image generation page.

• Enter a text prompt
• Choose image size & quality
• The "prompt refined" shows the final prompt used, which may be modified to improve results.

The mask generator uses the Token-Meter-AI, and image is created using the OpenAI API.

TokenMeter is released under the MIT License. See the LICENSE file for full terms.