AI Credits

How AI credits work

Documentation.AI uses credits for two AI features:

• The AI Documentation Agent (editor-facing — helps you write and edit docs)
• The AI Assistant or Ask AI widget (end-user facing — answers questions for your users)

Only these features consume credits. Browsing, editing, publishing, and normal site usage do not consume credits.

AI Assistant / Ask AI

The AI Assistant powers the on-page Ask AI widget for your end users.

Each end-user question costs exactly 0.1 credit, regardless of answer length or page.

For example, on the Professional plan (500 credits), your users can ask up to 5,000 questions per month.

AI Documentation Agent

The AI Documentation Agent helps you create, edit, and organize documentation. Documentation.AI recently upgraded the writer and code analysis model stack, so more capable models can improve output quality, but complex tasks may use more credits as a result.

What determines credit cost

Four practical factors affect how many credits a task uses:

1. Number of steps — Each action the agent takes, such as reading files, searching, editing, or validating, adds to the total. A small change may take 2–3 steps, while a broader task can take many more.

2. Amount of content processed — Larger files, more pages, and longer conversations require the agent to read and work through more content. A short page edit costs less than updating a large section of your docs.

3. Type of operation — Reading and searching usually cost less than writing, rewriting, or validating content. Tasks that create new pages or update multiple files generally use more credits.

4. Model usage and task complexity — Some tasks require deeper planning, code-aware analysis, or background validation. When the system uses more capable models to complete a more complex task, credit usage can be higher.

Typical credit usage

These examples reflect current agent behavior. If you saw older examples elsewhere, complex code-aware or background tasks may now be roughly 2x to 3x higher because the agent can do more planning, analysis, and validation during a request.

Worked example: "Check the latest commits and update the documentation"

This is a common task where you connect a GitHub repository and ask the agent to review recent code changes and update docs accordingly. The system may analyze connected repositories and may use different models during planning, code analysis, writing, and validation behind the scenes, which is why this kind of request often costs more than a basic edit.

Total: ~8–18 credits ($0.80–$1.80) depending on how many pages need updating

This task costs more because the agent needs to read from GitHub, search your docs, and then edit multiple pages. If only 1–2 pages need updating, it will usually land near the lower end of the range. If the commits affect many areas of your docs, it can move well beyond 10 credits.

Worked example: "Fix a typo"

Here is a step-by-step breakdown of a simple edit task — asking the agent to change a single word on a page.

Total: ~0.6 credits ($0.06)

Worked example: "Create 5 new pages from our API docs"

Here is a step-by-step breakdown of a complex, multi-page creation task.

Total: ~4.5 credits ($0.45)

Multi-turn conversations

If you send multiple messages in the same conversation (for example, asking the agent to edit a page and then asking a follow-up), each message is a separate request that uses credits independently. However, follow-up messages in the same conversation tend to be cheaper because:

• The agent already has context from previous messages
• Repeated content is cached, reducing processing costs

A typical follow-up message costs 30–50% less than the first message in a conversation.

Why the same task can cost different amounts

You might notice that similar tasks sometimes use slightly different amounts of credits. This is normal and happens because:

• Document size matters — Editing a 200-line page costs more than editing a 50-line page because the agent processes more content.
• Retries — If the agent's first edit attempt produces invalid JSON or doesn't match the file correctly, it retries automatically. Each retry adds a small amount of credits.
• Conversation length — Later messages in a long conversation process more context (the full conversation history), which increases cost slightly.

Where to view credit usage

You can view your current credit balance and usage history from Settings → AI Usage in your Documentation.AI dashboard.

The usage page shows:

• Credits remaining in your current billing cycle
• Credit usage over time
• A breakdown by feature (Documentation Agent vs. AI Assistant)

Tips to use credits efficiently

• Be specific — "Change the title of getting-started.mdx to Quick Start" costs less than "update the getting started page" because the agent doesn't need to search and read first.
• Use follow-ups — After the agent edits a page, ask for changes in the same conversation rather than starting a new one. Follow-ups are cheaper.
• Batch related changes — "Add pages for authentication, rate limiting, and webhooks" in one message is cheaper than three separate conversations.
• Use the AI Assistant for questions — If you just need to look up something in your docs, use Ask AI (0.1 credits) instead of the Documentation Agent.