Skip to content

Usage

ArchNGN uses a token-based system to measure AI processing. Tokens are consumed each time a workstream, connector job, or billable design check runs. Your current balance and usage history are displayed on the Usage page in the application.

ArchNGN supports three AI reasoning levels:

  • Advanced: Maximum depth reasoning for complex design and architecture tasks requiring thorough analysis, multi-step problem solving, and high-fidelity outputs.
  • Standard: Balanced reasoning suited for most workstream and connector workflows. Delivers strong results with efficient token usage.
  • Basic: Reserved for lightweight tasks such as document indexing and simple lookups.

When a job runs in advanced mode, the AI performs extended reasoning: working through the problem step by step, considering multiple approaches, and self-checking its outputs before completing. This additional computation produces more thorough and accurate results, but it is reflected in a higher token cost.

Use advanced mode for tasks where output quality matters most: generating new complex architecture designs, conducting extensive design reviews, or reverse-engineering highly complex systems. Standard mode is sufficient for routine connector syncs and lightweight metadata extraction, including basic design generation. Basic mode is reserved for lightweight tasks such as document indexing and simple lookups.

Reasoning ModelWorkstream TasksDesign ReviewConnector TasksDeep Design Check
Advanced10 tokens10 tokens4 tokens4 tokens
Standard5 tokens5 tokens2 tokens2 tokens
Basic3 tokens3 tokens1 token1 token
  • Fast design checks — the structural gate a pipeline runs on every commit. It performs no AI processing, so it costs nothing and keeps running even when your allowance is exhausted.
  • REST API and MCP reads — standards, components, the metamodel and context search. Metering architecture context would discourage exactly the behaviour the platform exists to encourage: querying the model before building. These are bounded by rate limits rather than tokens.

A deep design check is one bounded AI pass, priced like a connector task. A design review runs the full workstream pipeline and produces a document, so it is priced like a workstream.

If a deep check cannot be afforded it falls back to the free structural check, reports that it did so, and charges nothing. A design review is refused outright rather than quietly downgraded — it is an explicit request for a document, and returning something smaller would be worse than saying no.

Your token balance and aggregated usage over the last 6 months are visible on the Usage page in the application.

Contact your account administrator to adjust your token allowance.

Your allowance is checked before billable work starts, so a job that cannot be paid for is refused up front with a clear message rather than failing part-way through. Accounts with no allowance set are unlimited.