The full action-class weight table
When the hybrid consumption model ships in Q3 2026, every AI, MCP, and agent operation debits a weighted number of governance actions — not tokens. Weights reflect value delivered, so bills track outcomes rather than how chatty the model was on a given day. This page is the published weight table, automatically generated from the source-of-truth module.
Why actions, not tokens
Buyer alignment
Our buyers are CCOs, legal, risk, and compliance leaders — not ML engineers. Tokens aren't a mental model they work with.
No bill shock
Token burn varies 50x by context size and model choice. A governance action has a predictable cost regardless of what model ran it.
Decoupled from provider pricing
If Anthropic/OpenAI costs drop, your bill doesn't. If they rise, we absorb it on our margin, not your invoice.
Value over chattiness
A 200-token insight is worth the same as a 20,000-token verbose one if they deliver the same result. Action pricing reflects that.
The weight table
Grouped by category, ordered by weight. The table updates automatically when we add new actions — the source is lib/billing/action-classes.ts.
Read-side lookups
Policy, statement, glossary, and regulation reads. The cheapest class — these happen on almost every agent call.
| Operation | Actions |
|---|---|
| Policy lookup | 1 |
| Statement search | 1 |
| Glossary lookup | 1 |
| Regulation lookup | 1 |
MCP tool calls
Model Context Protocol tool invocations from external agents (Claude Code, Cursor, ChatGPT, Claude web, custom clients).
| Operation | Actions |
|---|---|
| MCP: search glossary | 1 |
| MCP: applicable policies for actor | 2 |
| MCP: search statements | 2 |
| MCP: statement guidance | 3 |
| MCP: get assembly bundle | 8 |
AI analysis
LLM reasoning over the corpus — comprehension, taxonomy, approval evaluation, gap analysis.
| Operation | Actions |
|---|---|
| Agent approval evaluation | 5 |
| Policy comprehension | 5 |
| Taxonomy decomposition | 5 |
| Light gap analysis | 8 |
| Deep gap analysis | 20 |
AI drafting
Generative operations that produce new governance artifacts. Higher cost reflects the output-heavy token profile.
| Operation | Actions |
|---|---|
| AI-drafted statement | 10 |
| AI-drafted glossary term | 10 |
| AI-drafted assembly | 15 |
Exports
Large structured exports — full bundle, ontology, audit log. Heaviest class because of I/O plus payload size.
| Operation | Actions |
|---|---|
| Full bundle export | 15 |
| MCP: ontology export | 15 |
| Ontology export | 15 |
| Audit log export | 25 |
The 30-day calibration commitment
These initial weights are estimates. When action metering launches in Q3 2026, the first 30 days are a public calibration window. We'll measure actual provider token cost per action class against the weights and adjust so the blended customer gross margin lands within a consistent band — not too thin, not a surprise markup on any specific class.
Any adjustments land in the open on this page, with a short changelog explaining what changed and why. No silent re-weightings.
Start on Community today
You already get AI credits on every tier. The Q3 2026 transition moves those credits to governance actions with the weights above — same system, better pricing mental model. Existing paying customers are grandfathered for at least six months.