Paper Compute Enterprise / AI platform in your cloud
The right model.
For every task.
Paper Compute Enterprise is a platform for AI-native teams, running in your own cloud. Understand how your team builds from recorded coding sessions. Use that evidence to improve model choices, set decision policies, and add compliance controls where needed.
01 / Model routing
Know what’s worth routing.
Your coding workflows may use your most capable model for work a more efficient one could handle. Routable spend estimates the potential cost reduction in that work.
Start with historical context in tapes: the prompts, tool calls, model choices, and costs behind your team’s coding sessions. Evaluate alternatives on that work before setting a decision policy.
The diagram shows how session evidence informs model choice. Your team evaluates the opportunity, approves a decision policy, and reviews new sessions to see whether it holds up.
How routable spend is estimated
The current tapes test prototype identifies overhead turns from live capture and short, read-only calls on frontier models. It prices token types separately and counts a candidate only when it still saves money after the first switch to the target model’s cache.
This is a screening estimate, not proof of equivalent quality or realized savings. Switching back to the frontier model is not yet charged, so the estimate can run high. Recurring task clusters proven on more efficient models are not yet computed.
Recorded coding sessions show how your team works. Evaluate model alternatives on that history, then approve a decision policy.
02 / Your enterprise platform
Your cloud.
One control plane.
Deploy the enterprise platform in your own cloud. Give teams a shared control plane for model access, decision policies, and session history, with clear ownership of each environment.
Connect hosted or private models and keep session records inside your perimeter. Start with routing, then add compliance controls or shared memory as your teams need them. The example below shows a private model running in your VPC.
Apply decision policies informed by historical evidence. In this private-model setup, requests and session records stay inside your VPC.
Illustrative private-model architecture. Deployment boundaries and policy delivery are agreed during enterprise scoping.
Know why a request went there.
Connect each model choice to the decision policy and the historical evidence behind it. Review new sessions to see whether the policy still delivers on cost and quality.
Keep what your teams learn.
Build on open-source tapes for the underlying record. Use historical context to propose decision policies. Validate them, roll them out, and refine them as new evidence arrives.
Add to the platform
Optional / Compliance controls
A policy check before the provider.
For workflows that handle sensitive data, add checks before a request reaches a model. Block restricted requests or flag them for review.
Keep the rule, verdict, and relevant session context as evidence for your reviewers. Scope enforcement, access, and retention with your compliance team.
Scope your controlsOptional / Shared memory
Share what one team learns.
When one team finds a better way to work, other teams usually start over. Add a memory service that distills session history into shared knowledge your agents can recall.
Memory is layered by organization, team, user, and agent, and each layer is visible only to the people and agents it belongs to. It runs on Postgres in your VPC, next to your session records, and improves as agents use it by resolving conflicting knowledge instead of just adding more.
Scope shared memoryDeployment options, enforcement coverage, and support are scoped with your team.
Discuss your decision policy03 / From opportunity to savings
10%less model spend?
Use routable spend to size the opportunity. Then test whether changing the routing can reduce total model spend by 10% while maintaining quality.
Pilot target · not a measured resultExplore routable spend- 01
Measure routable spend
Use recorded sessions to estimate the savings opportunity. Account for each token type and the cost of changing model caches.
- 02
Prove the model fit
Test candidates on real tasks. Compare completion, corrections, repeated requests, and latency before changing the route.
- 03
Verify realized savings
Roll out the decision policy gradually. Compare cost per completed task, including routing overhead, cache changes, and retries, while checking quality.
Enterprise AI enablement
Start with savings.
Add the controls you need.
Start with your team’s coding sessions. We’ll scope the platform in your cloud and evaluate where different model choices could reduce spend while preserving quality.
Talk to Paper Compute