No classifier API
The router reads the current prompt in memory and makes a deterministic local decision.
Native Codex routing
Route the work. Reserve the depth.
TokenX classifies each root task locally, selects a native Codex profile, and assigns a bounded sibling-agent budget. It spends deeper reasoning only when the work calls for it.
X
0-4 siblings
Default profiles
One policy, three depths.
Every route is configurable. TokenX fails visibly when a requested model is unavailable instead of silently swapping profiles.
Economy
Luna
Low
Bounded routine work
Standard
Terra
Medium
Normal engineering work
Deep
Sol
Xhigh
Complex or high-risk work
Operating principles
Bounded by design.
TokenX is a routing layer, not a hosted model service. Users remain responsible for their Codex plan, usage, and review of hook behavior.
No prompt archive
Routing state contains metadata and opaque identifiers, not raw prompts or model responses.
Controlled parallelism
Independent tasks can use one to four sibling agents within a per-turn global ceiling.