Tencent Cloud Agent Development Platform (ADP) uses Credit resources as the unified settlement unit for certain billing scenarios. These can be utilized across model services, plugins, knowledge base exceeding storage limits, and platform service scenarios. Deduction standards for different services can be viewed in Billing Overview (Package Subscription).
Model Services
Model services are used across multiple scenarios, with charges only applicable to platform built-in models. Including:
1. Dialogue scenario
In the following dialogue entries, using the model will incur charges:
Dialogue test: Execute in the app settings dialogue webpage or workflow debugging page.
Application evaluation: Use through the evaluation feature.
Experience user end and channels: Use via three types of entries - WeChat, WeCom, and trial user end.
API: Call the dialogue API, workflow asynchronous task interface, and knowledge retrieve API.
Charging Scenario Description:
In standard mode, charges apply when the corresponding generative models, thinking models, Prompt rewrite models, Embedding models, Reranker models, nl2sql models, document parsing models, and so on are selected.
Generative model/Thinking model:
In standard mode, select the generative model in application settings. The model output in dialogue results is QA reply, document reply, direct reply from large model, and database reply scenario.
In standard mode, select the thinking model in application settings. Use the thinking model to judge dialogue intent.
Select the model in the workflow's model node and call the related node in dialogue.
Select the model in the knowledge base QA plug-in and call the plug-in in dialogue.
In Multi-Agent Mode, report the amount by Agent model when calling an Agent.
Enable the long-term memory function during the dialogue process. Measure the data writing based on the selected model for writing, updating, and retrieval operations of the memory content.
Prompt rewrite model:
Enable "model context rewrite" in app settings to report the amount of rewritten user questions.
Enable "retrieve knowledge base to answer image questions" in app settings. In the scenario where a user asks a question containing images, use image search to retrieve the knowledge base for a response, and call the rewrite model to handle the issue and understand the image content.
Embedding model:
In standard mode, the amount used to retrieve knowledge base results for reply is reported by the selected vector model.
Call the large model knowledge q&a and knowledge retrieve nodes in the workflow during the dialogue.
Call the knowledge base QA plug-in and knowledge retrieve plug-in in dialogue.
Reranker model:
In standard mode, retrieve knowledge base results for reply.
Call the large model knowledge q&a and knowledge retrieve nodes in the workflow during the dialogue.
Call the knowledge base QA plug-in, knowledge retrieve plug-in, and reorder plug-in in dialogue.
nl2sql model:
In standard mode, when enabling the database or enhanced feature for Excel retrieval, the scenario where intelligent spreadsheets in the knowledge base are successfully retrieved will report the amount by the selected nl2sql model.
Call the database text2sql plug-in in dialogue.
Document parsing model:
Users input documents in conversation, and the document parsing model reports the amount by page.
2. One-click optimization scenario
Use the one-click optimization feature for prompts, welcome words, and handover descriptions in the app settings.
Use the one-click optimization feature in workflow nodes and workflow description.
3. App Operations - Intelligent Classification of Dialogue Records
Click the intelligent classification feature, and report the amount based on the selected generative model and vector model.
4. Generate Widget
In Widget development, use natural language generation to create widgets.
5. Knowledge base usage scenario
Documentation generation Q&A pair, similar questions generation, generate model report amount based on selected Q&A pairs.
Generate knowledge base Schema, generate model report amount based on selected knowledge base Schema.
Model usage can be viewed in Data Reports - Resource Dashboard - Model Usage:
Plugin
In addition to when the Knowledge Base QA plugin, Knowledge Retrieval plugin, Database text2sql plugin, and Reranker plugin are invoked, which generate model usage, all other plugins are billed based on usage when invoked. Invocation scenarios include when added as Agent tools and when called in workflows. Usage can be viewed in Data Reports - Resource Dashboard - Plugin Usage:
Knowledge Base Excess Storage
Use Case: Upload documents, QA, and database in the knowledge base require charging for used capacity, effective at the enterprise dimension, count the sum of usage within the enterprise, and use the knowledge base capacity in the package, Credit resources, and purchase additional prepaid resource packages. Deduction priority: knowledge base capacity in the package > Credits in the package > prepaid resource package.
Charge logic: Uploaded knowledge will preferentially use the knowledge base capacity in the package. The part that exceeds the knowledge base capacity limit of the package is called excess usage. Fees will be charged for the knowledge base excess usage until the excess part is actively deleted. The billing formula for knowledge base excess usage is as follows:
Excess usage charge = (Total knowledge base usage - Package limit) × Excess time × Unit price (Measurement unit: USD).
Excess time is the usage duration of the part that exceeds the free quota. Billing stops when the used capacity is restored to below the package limit after user deletion of the knowledge base file. Usage is reported once per hour.
Knowledge base excess usage will be divided into "storage usage" and "computation usage", calculated by different prices and deducted using Credit resources.
Storage usage: Includes charges for COS usage (storage usage will be reported when customers choose to import non-COS files or import Q&A pairs beyond the allocated quota).
Computation usage: Charges apply to vectorized knowledge data.
Excess storage and computation: When beyond package capacity (free 1 GB, starter 5 GB, Team Edition 100 GB, enterprise edition 1 TB), storage is billed at 0.2 Credit/GB/hour, and vectorized computation at 12.4 Credit/GB/hour.
Knowledge Base usage can be viewed in Data Reports - Resource Dashboard - Knowledge Base Capacity:
Platform service
1. Long-term memory
When using long-term memory, fees are charged per hour based on the number of stored dialogue entries submitted. For example, if 1 entry is stored at 12:00 and 5 entries at 12:30, a total of 6 entries will be reported for the interval 12:00–13:00 at 13:00.
2. Content security
Use the security feature with custom policy content, charging by the count of QA and answer calls. For example, one round of questions and answers is counted as 2 calls ("question + answer").
3. Application call count
Calling the created Intelligent Agent through any means to perform one input and output, for example, via API, where the agent is called for one round of conversation (one query input, one output), is counted as one call.
It can be viewed in Data Reports > Resource Dashboard > Platform Usage:
