What a plan bounds

Projects held

How many may exist at once. This bounds storage.

Analyses per day

How many runs may start. This bounds worker contention.

Tokens and spend per month

Both, because tokens alone do not bound money once model choice is yours — the same analysis costs several times more on the most capable model than on the cheapest.

Model access

All catalogued models, or the cheapest tier — see Models and routing.

Graph capability

Viewing a precomputed graph is cheap; the symbol level is the heaviest thing the layout does; traversal queries are the only part that spends database time per request. They tier separately.

KEEL

Sessions per day and turns per session bound spend directly, because a session’s length is yours to decide. The deliverables are the value cut.

Custom Skills

Requesting a Skill nobody measured a need for. Its own entitlement, because it is an open-ended prompt you write — unlike regenerating, which re-runs a role the analysis already justified.

• The whole deterministic pipeline — structure, technology, architecture, dependencies, findings, constraints and the change set.
• The Skill security audit, on every analysis, because no model is involved in it.
• Regenerating a generated Skill, because it re-runs a role the analysis already justified.
• Any model your own connected key can reach. Your credential is not ours to ration.

If you have not run anything yet, start here. The current numbers for your account, and your standing against each limit, are on your plan page rather than in this chapter — a number published in documentation is a number that goes stale silently.