What you capture when you integrate.
A position in the router
Every meaningful routing improvement your data produces mints tokens to you. Hold them, or redeem for USDC anytime.
Lower inference cost
A smarter router picks the right model for the task. You spend less on inference before counting any rewards.
Configurable value share
Decide at integration whether to keep token flow, pass it through to your users, or split. Make Hokusai your monetization layer or your user-acquisition feature.
Compounding ownership
Your early contributions keep paying as more harnesses route through the same model. You're not selling data; you're buying into an asset.
Inside a routing decision
Incoming task
Refactor the existing auth middleware so API keys can be restricted to specific scopes and routes. Requirements: - Keep the current middleware entrypoint stable for integrators. - Enforce scope checks before request handlers run. - Preserve existing admin flows while tightening least-privilege defaults. - Add tests covering missing scope, partial scope, and valid scope paths. - Document any new assumptions in code comments near the policy boundary.
Task packet
Refactor auth middleware to support scoped API keys
Available model families
Choice Layer
Historical outcomes with similar task packets are scored against the current budget and evaluation envelope, then narrowed to the route most likely to hold up end to end.
Output
Planner
claude-opus-4-7
Strong at shaping the migration plan and isolating policy boundaries.
Available for fallback
Coder
gpt-5.4
Best fit for the implementation pass and test repair inside budget.
Available for fallback
Reviewer
claude-sonnet-4-6
Good balance for regression review and policy edge-case coverage.
Available for fallback
Evaluation
How the run scored against the eval rubric.
Total cost
$18.42
Planner score
9.2 / 10
Coder score
8.7 / 10
Reviewer score
9.5 / 10
Wall clock
6m 52s
Coding / Multi-Model Routing
Coding Task Router
The first decision layer built on the Hokusai protocol, turning real coding tasks into a shared router that learns, compounds, and pays contributors back.
+3.2 pts
Cost-adjusted task success (illustrative)
12,400
Tasks routed, last 7d (illustrative)
27
Contributors (illustrative)
184,000
Tokens minted to date (illustrative)
Integration
Drop-in middleware.
Route tasks through Hokusai, execute them in your harness, then report the result back so the shared router can keep improving.
import { route } from '@hokusai/router'
const { model, reasoning } = await route({
task: userTask,
context: harnessContext,
})
const result = await models[model].run(userTask)
await route.reportOutcome(result) // mints tokens proportional to performance liftWhere does your routing data go today?
| Lab-owned auto-routing | Hokusai | |
|---|---|---|
| Who captures the optimization signal | The lab | You and the contributors |
| Who keeps the inference cost savings | The lab keeps margin | You |
| What you build over time | Nothing transferable | A token position in the router |
| Portability across harnesses | Locked in | Take your position with you |
| Auditability | Opaque | On-chain attribution |
What else the factory makes
The router is the first decision layer built on Hokusai. The same protocol applies to every model where shared, incentive-aligned learning beats siloed effort.
Tool-selection router
Same factory, same upside structure: which tool/MCP to call, learned from outcomes.
Learn moreMemory retrieval optimizer
A learned policy for what to surface from a user's context, contributors share in the lift.
Learn moreCode-review critic router
Which reviewer model catches which class of bug, improved by outcome data from real PRs.
Learn morePrompt-policy optimizer
A shared model for prompt strategies, owned by the engineers who improve it.
Learn more